News Archive

Archived automated mobility, AI, LLM, NLP, computer vision, and vision-language model signals from the website news feed.

arXiv / arxiv-llm-vlm-av / Oct 8, 2026

DVD: Dynamic Vector Decoding for Efficient MLLM-based Perception

Multimodal large language models have made remarkable progress in bridging vision and language, facilitating various perception tasks essential for human-machine interaction, robotics, and autonomous driving. However, existing MLLM-based perception methods predominantly rely on text-based coordinate representation, which suffers from excessive token overhead, or fixed-range quantization, which suffers from range and precision constraints, especially for 3D domains with unbounded spatial range and high localization accuracy requirements. To address these challenges, we propose a dynamic vector decoding method named DVD, which unifies the representation of 2D and 3D perception tasks.

arXiv / arxiv-llm-vlm-av / Oct 7, 2026

Explicit Geometric Chain-of-Thought for Vision-Language-Action in Autonomous Driving

Vision-language-action~(VLA) models have emerged as a promising paradigm for autonomous driving. However, existing VLA models still suffer from a fundamental mismatch: driving actions require precise 3D geometric cues, while visual-language understanding and reasoning are largely conducted in a 2D semantic space. In this paper, we propose GeoCoTDrive, an explicit geometric chain-of-thought framework that grounds geometry in a planning-oriented manner.

arXiv / arxiv-llm-vlm-av / Oct 6, 2026

Transferable Spatial Temporal Coherence Adversarial Attack on Black-Box Vision Language Models for Autonomous Driving

The rapid integration of Vision Language Models (VLMs) into sensitive systems introduces critical safety vulnerabilities that remain unexplored in exist studies. While adversarial attack robustness has been extensively studied for image-based models, the susceptibility of VLMs to temporally-aware adversarial attacks against video in driving context poses a distinct and under examined threat. In this paper, we introduce novel adversarial attack against video targeting VLM models used for autonomous driving scenes named Spatial Temporal Coherence Adversarial Attack (STCA).

Electrek / automated-vehicles / Oct 3, 2026

Musk explains why Tesla Robotaxi is not running at night, and lidar is the answer

Elon Musk explained that Tesla's Robotaxi service is not operating at night due to limitations in its sensor suite, specifically the lack of LiDAR. Musk indicated that incorporating LiDAR technology is crucial for improving the vehicle's performance in low-light conditions, highlighting the ongoing challenges in sensor technology for autonomous vehicles.

arXiv / arxiv-llm-vlm-av / Oct 2, 2026

Retrieval-Augmented Large Language Model Decision-Making for Autonomous Driving Guided by Chinese Philosophical Wisdom

A research team proposes Chinese Philosophical Wisdom-Guided Driving (CPW-Drive), a retrieval-augmented generation (RAG) framework that integrates value principles from Chinese Confucian thought into autonomous driving decision-making. CPW-Drive uses LLM-extracted keywords from classical texts to guide driving decisions, balancing safety, efficiency, and social norms in complex traffic interactions. This approach aims to enhance the ethical and philosophical considerations in autonomous driving systems.

arXiv / arxiv-llm-vlm-av / Oct 2, 2026

Localized Conformal Safety Monitoring with Vision-Language Models for Autonomous Driving

Researchers from a 2026 arXiv paper introduced Split Label-Localized Conformal Prediction (SLLCP), a post-hoc calibration layer for vision-language models (VLMs) in autonomous driving. SLLCP transforms unreliable VLM predictions into probabilistically calibrated safety assessments, improving the reliability and safety of trajectory monitoring in autonomous vehicles.

Gizmodo / ai / Oct 1, 2026

Google Is Already Having Problems With Its Latest AI Model

Google is encountering issues with its latest AI model, which has raised concerns about the reliability and safety of the technology. These problems highlight the ongoing challenges in developing and deploying advanced AI systems, particularly in ensuring they operate safely and effectively in real-world scenarios.

arXiv / arxiv-llm-vlm-av / Oct 1, 2026

Towards Reliable Vision-Language Models for Autonomous Driving

A team of researchers evaluated five Vision-Language Models (VLMs) for their reliability in autonomous driving tasks, including scene understanding and decision-making. The study, published in arXiv, highlights the importance of robustness and reliability in VLMs, especially under real-world conditions with degraded visual inputs, and provides benchmark scores for models like Qwen3.5-9B and LLaVA-OneVision-7B.

Reuters / ai / Sep 30, 2026

Google announces Gemini 4 flagship AI model after months of delays

Google has announced the release of Gemini 4, its flagship AI model, after several months of delays. The model is designed to advance the state of the art in AI, but the delays suggest that developing and refining such models is a complex and time-consuming process, with significant implications for AI research and development.

arXiv / arxiv-llm-vlm-av / Sep 30, 2026

Learning to Explain While Planning: Rule-Aligned Diffusion Planning for Autonomous Driving

Researchers introduced Rule-Aligned Diffusion Planning (RADP), a method that enhances diffusion planners by aligning trajectory generation with driving rules. This approach, published in arXiv, addresses the limitations of existing methods that rely solely on statistical correlations, improving safety and compliance in long-tail scenarios where expert data is scarce and rule-level explanations are crucial for failure diagnosis and safety validation.

arXiv / arxiv-llm-vlm-av / Sep 30, 2026

Learning to Explain While Planning: Rule-Aligned Diffusion Planning for Autonomous Driving

Researchers propose Rule-Aligned Diffusion Planning, a method that integrates driving rules into the trajectory generation process of diffusion planners. This approach ensures that trajectories are not only statistically sound but also rule-compliant, enhancing safety and interpretability. The method addresses the limitations of existing planners that lack rule-level explanations, making it a critical advancement for autonomous driving safety and validation.

arXiv / arxiv-llm-vlm-av / Sep 30, 2026

Comparative study of adapting pre-trained models for driving behavior video captioning

A study by researchers compares various fine-tuning and prompting methods for adapting Large Language Models (LLMs) to driving behavior video captioning. The study, using the BDD-X dataset, demonstrates that full fine-tuning of SpaceTimeGPT outperforms the baseline in several automatic metrics. The research also explores the limitations of Low-Rank Adaptation (LoRA) and prompt engineering on the VideoLLaVA model, providing insights into the effective use of LLMs in autonomous driving.

Reuters / automated-vehicles / Sep 30, 2026

Mercedes-backed Momenta targets Dubai and Europe for robotaxi expansion

Mercedes-backed autonomous driving company Momenta is planning to expand its robotaxi operations to Dubai and Europe. This move underscores Momenta's strategic focus on international markets and its commitment to deploying advanced autonomous vehicle technology in diverse urban environments.

arXiv / arxiv-explainability-av / Sep 29, 2026

Vision-Language-Action Autonomous Driving Agent with Language-based Memory

The Vision-Language-Action (VLA) foundation models, developed by researchers, address the memory limitations of existing VLA models in autonomous driving. AD-Memo, the proposed agent, uses language-based memory to enhance interpretability and portability, crucial for tasks like determining arrival order at all-way stops and long-horizon driving scene understanding. This advancement is significant for improving the reliability and explainability of autonomous driving systems.

arXiv / arxiv-llm-vlm-av / Sep 29, 2026

ExceptionDrive: A Planning-Oriented Counterfactual Corner-Case Benchmark for Autonomous Driving

Researchers from an unspecified institution proposed ExceptionDrive, a counterfactual planning benchmark for autonomous driving. ExceptionDrive uses VLM-assisted screening and multi-view editing to insert safety-critical hazards into real-world driving scenarios. The benchmark evaluates planner reliability under rare, high-risk conditions, providing a comprehensive assessment of AV planning capabilities.

Reuters / automated-vehicles / Sep 29, 2026

Tesla receives approval to roll out self-driving software in Croatia

Tesla has received approval to roll out its self-driving software in Croatia, expanding its autonomous driving capabilities to a new European market. This deployment marks a significant step in Tesla's global expansion of its Full Self-Driving (FSD) system, which aims to enhance the safety and efficiency of autonomous vehicles.

arXiv / arxiv-llm-vlm-av / Sep 29, 2026

Speed in the Blind Spot: An Interpretability Analysis of Dynamic Perception in VLMs for Autonomous Driving

A study by researchers at an unspecified institution evaluated the velocity understanding capabilities of Vision-Language Models (VLMs) in autonomous driving systems. The study, conducted on the nuScenes dataset, assessed three tasks: surrounding-agent speed, current ego speed, and short-horizon future ego-speed. The findings highlight distinct failure modes and the need for improved dynamic perception in VLMs, crucial for safe and reliable AV operation.

Reuters / ai / Sep 28, 2026

OpenAI shelves new AI model release over safety concerns

OpenAI has shelved the release of a new AI model due to safety concerns, emphasizing the company's commitment to responsible AI development. This decision reflects the growing importance of safety and ethical considerations in the deployment of advanced AI systems, particularly in autonomous vehicles.

arXiv / arxiv-llm-vlm-av / Sep 28, 2026

Beyond Retrieval Relevance: Scene-Grounded Risk Entailment for Vision-Language Driving

Jian Zhao, Wenxu Li, Bing Zhu, Peixing Zhang, Shizheng Jia, and Yinzi Huang from arXiv introduce a method for constructing a multimodal accident scenario database for end-to-end autonomous vehicle testing. The database integrates various sensor data to create realistic and diverse driving scenarios, which is essential for comprehensive testing and validation of autonomous driving systems.

Accident Analysis and Prevention / journal-rss / Sep 26, 2026

A multimodal accident scenario database construction method for end-to-end autonomous vehicle testing

Jian Zhao, Wenxu Li, Bing Zhu, Peixing Zhang, Shizheng Jia, and Yinzi Huang from Accident Analysis & Prevention developed a multimodal accident scenario database for end-to-end autonomous vehicle testing. Published in December 2026, the method integrates various sensor data to create realistic accident scenarios, improving the testing and validation of AV perception and decision-making systems.

Reuters / automated-vehicles / Sep 25, 2026

EU vote on Tesla's supervised self-driving system pushed back

The EU has postponed the vote on Tesla's supervised self-driving system, delaying the regulatory decision on its deployment in Europe. This delay reflects ongoing concerns and the need for thorough evaluation of the system's safety and performance, which is crucial for the responsible rollout of autonomous driving technologies.

TechCrunch / automated-vehicles / Sep 24, 2026

Waymo is scaling fast: Here’s what the fleet data shows

Waymo's fleet data indicates rapid scaling of its autonomous vehicle operations, with significant increases in the number of vehicles and miles driven. This expansion underscores Waymo's commitment to deploying a robust and reliable robotaxi service, contributing to the advancement of autonomous driving technology.

Electrek / automated-vehicles / Sep 22, 2026

Tesla Full Self-Driving bug falsely flags dirty cameras, demands service

Tesla's Full Self-Driving (FSD) system has encountered a bug that falsely flags cameras as dirty and demands unnecessary service. This issue can lead to user frustration and increased maintenance costs. The bug underscores the ongoing challenges in developing and deploying robust autonomous driving systems that can accurately interpret sensor data and environmental conditions.

Electrek / automated-vehicles / Sep 22, 2026

Tesla’s active robotaxi fleet in Austin crashes back to 8 vehicles after Cybercab pump

Tesla's active robotaxi fleet in Austin, Texas, has decreased to 8 vehicles following a surge in demand and subsequent adjustments. The reduction highlights the challenges in maintaining a consistent fleet size while balancing operational costs and service availability. This fluctuation in fleet size is a critical factor in the real-world deployment and scalability of autonomous ride-hailing services.

Reuters / ai / Sep 21, 2026

US, China to meet again on AI safety in two months in Shenzhen, Bessent says

The United States and China are scheduled to meet in Shenzhen in two months to discuss AI safety, as announced by Bessent. This meeting reflects the growing international focus on ensuring the safe development and deployment of AI technologies, particularly in the context of autonomous vehicles.

reuters.com / ai / Sep 20, 2026

Bessent proposes US-China AI safety notifications in talks with Chinese vice premier

US National AI Coordinator Lynne Parker Bessent has proposed a system for US-China AI safety notifications during talks with Chinese Vice Premier He Lifeng. The proposal aims to enhance communication and cooperation between the two countries on AI safety issues, reflecting the global importance of ensuring the responsible development and deployment of AI technologies.

The Verge / automated-vehicles / Sep 18, 2026

Waymo says Singapore will be its next international robotaxi city

Waymo, a leading autonomous vehicle company, has announced that Singapore will be its next international city for robotaxi deployment. This expansion marks Waymo's commitment to globalizing its self-driving technology and services, potentially increasing the availability of autonomous ride-hailing in urban areas.

Engadget / ai / Sep 17, 2026

OpenAI Reveals More Instances Of Concerning AI Model Behaviors During Testing

OpenAI has disclosed additional instances of concerning behaviors exhibited by its AI models during testing, raising questions about the robustness and reliability of these systems. The findings underscore the ongoing challenges in ensuring that AI models operate safely and predictably, particularly in real-world applications.

Electrek / automated-vehicles / Sep 16, 2026

Tesla brings Cybercab to China, rules out sale and robotaxi service

Tesla introduced the Cybercab in China but ruled out selling the vehicle and launching a robotaxi service. This decision highlights Tesla's strategic approach to market entry and its cautious stance on deploying autonomous technology in new regions, despite the potential for significant market impact.

Reuters / automated-vehicles / Sep 15, 2026

US agency orders Tesla to answer questions on Cybercab certification

The National Highway Traffic Safety Administration (NHTSA) has ordered Tesla to provide detailed information on the certification of its Cybercab, a new electric vehicle designed for autonomous ride-hailing. The agency's request comes amid growing concerns about the safety and regulatory compliance of Tesla's autonomous driving technologies.

Electrek / automated-vehicles / Sep 15, 2026

Tesla’s record month: 4 fatal Autopilot and Full Self-Driving crashes

Tesla experienced a record month with four fatal crashes involving its Autopilot and Full Self-Driving (FSD) systems, raising serious concerns about the safety and reliability of these technologies. The incidents have prompted increased scrutiny from regulators and the public, highlighting the need for more rigorous testing and oversight in the development of autonomous driving systems.

arXiv / arxiv-explainability-av / Sep 15, 2026

PiPS: Post-Hoc Prototypical Explanations for Interpretable Semantic Segmentation

ArXiv researchers present PiPS, a post-hoc prototypical explanation framework for interpretable semantic segmentation in autonomous vehicles. PiPS uses prototype-based reasoning to explain model decisions based on visual similarity, enhancing trust in critical systems by providing clear, human-understandable explanations of object localization.

TechCrunch / automated-vehicles / Sep 14, 2026

Waymo opens robotaxi service in Las Vegas

Waymo has launched its robotaxi service in Las Vegas, expanding its autonomous vehicle operations to a new major city. This deployment marks a significant step in the company's efforts to commercialize self-driving technology and provide reliable, driverless transportation options to the public.

Reuters / automated-vehicles / Sep 14, 2026

Waymo to bring autonomous ride-hailing to Las Vegas

Waymo is expanding its autonomous ride-hailing service to Las Vegas, marking a significant step in the company's efforts to deploy self-driving vehicles in new urban environments. The deployment will leverage Waymo's advanced AV technology, which has been tested and refined in multiple cities, to provide a safe and reliable transportation option for residents and visitors.

Transportation Research Part C: Emerging Technologies / journal-rss / Sep 14, 2026

Accurate and sample efficient driver takeover time prediction in autonomous driving with privacy preservation: A federated meta attentive interpretable tabular learning approach

Weida Yang, Zhizhou Wu, Yinhai Wang, and Yunyi Liang from Transportation Research Part C: Emerging Technologies published a paper in December 2026, presenting a federated meta attentive interpretable tabular learning approach for predicting driver takeover time in autonomous driving. The method ensures privacy preservation and sample efficiency, addressing critical issues in the transition from automated to manual driving.

Transportation Research Part C: Emerging Technologies / journal-rss / Sep 14, 2026

Adaptive stress testing of autonomous vehicle-pedestrian interactions: A large language model (LLM) distillation approach

Xiao Wen, Wentao Huang, Zhiyong Cui, and Sisi Jian from Transportation Research Part C: Emerging Technologies published a paper in January 2027, introducing an LLM distillation approach for adaptive stress testing of autonomous vehicle-pedestrian interactions. The method enhances the robustness of AV systems by simulating complex pedestrian behaviors, crucial for improving safety and reliability in real-world scenarios.

Reuters / ai / Sep 10, 2026

IBM, NASA launch AI model to help map ice, craters on Moon

IBM and NASA have jointly launched an AI model designed to map ice and craters on the Moon, advancing space exploration and scientific research. This collaboration leverages AI to enhance the accuracy and efficiency of lunar mapping, supporting future missions and discoveries.

Reuters / ai / Sep 9, 2026

OpenAI pushes for mandatory national AI safety rules

OpenAI is advocating for the implementation of mandatory national AI safety rules to ensure the responsible development and deployment of AI technologies. This push for regulation reflects growing concerns about the potential risks associated with AI systems.

arXiv / arxiv-explainability-av / Sep 9, 2026

Data-Driven Risk Fields for Safer End-to-End Autonomous Driving

Zhiyong Cui, Xiao Wen, Wentao Huang, and Sisi Jian from arXiv present DRiF, a data-driven risk-field framework for safer end-to-end autonomous driving. DRiF learns a shared bird's-eye view (BEV) feature with static map segmentation, dynamic risk prediction, and vehicle planning, addressing the limitations of rule-based risk models and enhancing the safety of autonomous driving systems.

The Verge / automated-vehicles / Sep 9, 2026

I spent an hour riding inside Tesla’s steering-wheel-free Cybercab

A journalist from The Verge spent an hour riding in Tesla’s steering-wheel-free Cybercab, a prototype designed for fully autonomous operation. The experience provided insights into Tesla’s progress in developing a vehicle that can navigate urban environments without human intervention.

arXiv / arxiv-llm-vlm-av / Sep 8, 2026

LLMSec-AV: A Vulnerability Taxonomy and LLM-Driven Software Weakness Discovery Framework for Autonomous Vehicles

In a study published on arXiv, researchers developed LLMSec-AV, a large language model (LLM)-driven framework for identifying software vulnerabilities in autonomous vehicles (AVs). The framework uses an AV-specific vulnerability taxonomy and was evaluated on Autoware, decomposing 770 translation units into 4,673 functions and analyzing 161 functions under various prompting conditions. LLMSec-AV outperformed rule-based tools in detecting security weaknesses.

IEEE Spectrum / automated-vehicles / Sep 8, 2026

Cleaner Crash Data Could Prove That Human Drivers Are the Risky Ones

IEEE Spectrum discusses the potential for cleaner crash data to demonstrate the superior safety of autonomous vehicles compared to human-driven cars. The article highlights the importance of accurate data in advancing the adoption and regulation of AVs.

arXiv / arxiv-llm-vlm-av / Sep 8, 2026

PlannerForge: LLM Agents for Scenario-Based Testing of Motion Planners in Autonomous Driving

PlannerForge, an LLM-agent framework developed by researchers at arXiv, extends all stages of scenario-based testing for autonomous driving systems (ADSs) and adds two new LLM-enhanced stages: ADS Enhancement and ADS Assessment. This unified framework aims to improve the safety and reliability of ADSs by systematically validating and enhancing their performance across various scenarios.

IEEE Spectrum / automated-vehicles / Sep 8, 2026

Inside Rivian’s Bid to Build Level 4 Robotaxis—and Then Passenger Cars

Rivian is developing Level 4 robotaxis as part of its broader strategy to enter the autonomous vehicle market. The company aims to leverage this technology to eventually produce advanced passenger cars, highlighting its commitment to innovation in the AV sector.

Electrek / automated-vehicles / Sep 7, 2026

A Tesla ran a stop sign and killed a man, record confirms Full Self-Driving/Autopilot was on

A Tesla vehicle equipped with Full Self-Driving (FSD) and Autopilot features was involved in a fatal accident after running a stop sign, according to confirmed records. The incident highlights the ongoing safety concerns and regulatory scrutiny surrounding Tesla's autonomous driving technology. This event underscores the need for rigorous testing and oversight of FSD systems.

Reuters / ai / Sep 4, 2026

EXCLUSIVE: US, China gear up for mid-September AI safety talks

The United States and China are preparing for mid-September talks focused on AI safety, aiming to establish international standards and cooperation. These discussions are crucial for addressing global concerns about the ethical and safe deployment of AI technologies.

Reuters / automated-vehicles / Sep 4, 2026

Musk pushes regulatory limits with Tesla's Cybercab robotaxi service

Elon Musk is pushing the boundaries of regulatory oversight with Tesla's Cybercab robotaxi service. The service, which offers autonomous rides, is facing scrutiny from regulators due to concerns over safety and compliance with existing laws.

arXiv / arxiv-llm-vlm-av / Sep 4, 2026

CoLMIN: LLM-based Multi-Decision Path Negotiation for Cooperative Autonomous Driving

CoLMIN, proposed by researchers at an unnamed institution, is an LLM-based multi-decision path negotiation framework for cooperative autonomous driving. It addresses the challenge of premature convergence to suboptimal solutions in complex traffic scenarios by leveraging the reasoning capabilities of LLMs to enhance inter-vehicle negotiation and improve overall driving performance.

Engadget / automated-vehicles / Sep 4, 2026

Tesla Starts Offering Cybercab Robotaxi Rides

Tesla has launched its Cybercab robotaxi service, offering rides to the public. The service operates without human drivers, marking a significant step in the deployment of autonomous vehicles. This development underscores Tesla's commitment to advancing self-driving technology and expanding its presence in the robotaxi market.

Electrek / automated-vehicles / Sep 4, 2026

Tesla announces Robotaxi has driven 1 million unsupervised miles

Tesla's Robotaxi has achieved a milestone by driving 1 million unsupervised miles. This significant achievement highlights the progress in Tesla's autonomous driving capabilities and the reliability of its Full Self-Driving (FSD) system. The data will be crucial for regulatory approval and public trust in autonomous vehicle technology.

Reuters / ai / Sep 3, 2026

OpenAI commits $1 billion to cyberdefense effort amid AI safety scrutiny

OpenAI has committed $1 billion to a cyberdefense effort, reflecting the company's commitment to addressing AI safety concerns. This investment is aimed at enhancing cybersecurity measures and protecting against potential threats, which is crucial for the responsible development and deployment of AI technologies.

arXiv / arxiv-llm-vlm-av / Sep 3, 2026

Continuous Actions from Discrete Minds: Latent-Aligned Planning for End-to-End Autonomous Driving

The paper introduces LaPla, a Vision-Language-Action (VLA) framework that uses latent-aligned planning to bridge the gap between discrete reasoning and continuous actions in autonomous driving. LaPla employs a residual vector-quantized variational autoencoder (VQ-VAE) to capture vehicle kinematics and encode trajectory features, enhancing the precision of motion execution.

Reuters / automated-vehicles / Sep 3, 2026

France starts tests on Tesla's self-driving tech

France has initiated tests on Tesla's self-driving technology, marking a significant step in the European validation of Tesla's autonomous systems. These tests aim to assess the safety and reliability of Tesla's FSD in a controlled European environment, which could influence regulatory decisions and public acceptance.

IEEE Transactions on Intelligent Transportation Systems / journal-rss / Sep 3, 2026

HybridFusion: A Universal Multimodal Data Fusion Framework for 3-D Detection and Tracking

Environment perception based on multimodal data fusion is crucial for the development of autonomous driving (AD) technology. However, the existing fusion-based detection and tracking methods often struggle to efficiently utilize the multimodal complementary information. This limitation can significantly impact the safety of autonomous vehicles (AVs), especially in complex traffic scenarios and adverse weather conditions.

Reuters / automated-vehicles / Sep 3, 2026

Tesla starts Cybercab rides in Austin, drawing US safety agency interest

Tesla has launched Cybercab rides in Austin, Texas, a move that has attracted the attention of the US safety agency. The Cybercab service, which uses Tesla's Full Self-Driving (FSD) technology, is a significant step towards commercializing Tesla's autonomous driving capabilities and will be closely monitored for safety and performance.

The Verge / automated-vehicles / Sep 2, 2026

Uber beats Waymo as first to launch robotaxis in London

Uber has launched its robotaxi service in London, becoming the first company to do so in the city. This deployment marks a significant milestone in the expansion of autonomous vehicle services in Europe, potentially challenging established players like Waymo.

TechCrunch / ai / Sep 2, 2026

OpenAI’s new reasoning technique alarms AI safety experts

OpenAI's new reasoning technique has raised concerns among AI safety experts, highlighting the potential risks associated with advanced AI models. The technique, while innovative, may pose significant safety challenges that need to be addressed to ensure responsible AI development.

The Verge / ai / Sep 2, 2026

Researchers fear safety disaster ahead of OpenAI’s Astra release

Researchers are expressing concerns about the potential safety risks associated with OpenAI's upcoming Astra release. The release of Astra, a powerful AI model, has raised fears about its potential misuse and the need for stringent safety measures to prevent unintended consequences.

Reuters / ai / Sep 1, 2026

US urges hands-off approach to AI regulation at G20 tech meeting

The United States is urging a hands-off approach to AI regulation at the G20 tech meeting, emphasizing the need for innovation and flexibility in the development of AI technologies. This stance could influence global policies and standards for autonomous vehicles and other AI applications.

Reuters / automated-vehicles / Sep 1, 2026

Amazon's Zoox, Alphabet's Waymo expand robotaxi services to more US cities

Amazon's Zoox and Alphabet's Waymo are expanding their robotaxi services to additional U.S. cities, enhancing their reach and providing more opportunities for public interaction with autonomous vehicle technology. This expansion underscores the growing commercial viability of robotaxi services.

engadget.com / automated-vehicles / Sep 1, 2026

Waymo Robotaxi Rides Are Now Open To The Public In Denver, San Diego And Tampa

Waymo has expanded its robotaxi service to the public in Denver, San Diego, and Tampa, bringing the total number of cities with Waymo robotaxis to 14. This expansion underscores Waymo's commitment to widespread deployment of autonomous vehicles and enhances public access to the technology.

Electrek / automated-vehicles / Sep 1, 2026

Waymo opens robotaxi rides to the public in 3 new cities, now 14 total

Waymo has opened its robotaxi service to the public in three new cities—Denver, San Diego, and Tampa—increasing the total number of cities with Waymo robotaxis to 14. This expansion highlights Waymo's rapid progress in deploying autonomous vehicles across multiple urban environments.

Reuters / automated-vehicles / Sep 1, 2026

Tesla touts European supervised self-driving safety data ahead of EU vote

Tesla has shared safety data from its supervised self-driving system in Europe, ahead of a crucial EU vote on autonomous vehicle regulations. The data aims to demonstrate the safety and reliability of Tesla's technology, which is critical for gaining regulatory approval and public trust.

Reuters / ai / Sep 1, 2026

US urges hands-off approach to AI regulation at G20 tech meeting

The United States has advocated for a hands-off approach to AI regulation at the G20 tech meeting, emphasizing the importance of fostering innovation without excessive regulatory constraints. This stance reflects a broader debate on the balance between AI advancement and regulatory oversight.

OpenAI / ai / Sep 1, 2026

OpenAI supports California’s bill to advance youth AI safety

OpenAI has expressed support for California's bill aimed at enhancing youth AI safety. The company's backing underscores the growing importance of regulatory frameworks to ensure that AI technologies, including those used in autonomous vehicles, are developed and deployed responsibly.

Electrek / automated-vehicles / Aug 31, 2026

A second Tesla driver died stopped on a freeway with Autopilot/Self-Driving on

A Tesla driver died while the vehicle was stopped on a freeway with Autopilot and Full Self-Driving (FSD) engaged. The incident highlights the ongoing safety concerns and regulatory scrutiny surrounding Tesla's advanced driver-assistance systems, particularly in scenarios where the system may not adequately handle stationary objects.

arXiv / arxiv-llm-vlm-av / Aug 31, 2026

LLM-Driven Autonomous Vehicles Inherit Human Driver Biases in Pedestrian Yielding: Results and Implications From A New Benchmark

A study by arXiv researchers reveals that large language models (LLMs) and visual-language models (VLMs) used in autonomous vehicles (AVs) can inherit human driver biases, such as lower pedestrian-yielding rates to Black pedestrians in the US. The paper introduces two new bias testing methodologies, 'All Else Being Equal' and 'Self-Consistency' tests, to assess and mitigate these biases, emphasizing the importance of fairness in AV decision-making.

arXiv / arxiv-llm-vlm-av / Aug 31, 2026

Qwen-Drive-1.0: An Initial Step towards a Vision-Language Foundation Model for Autonomous Driving

Qwen-Drive-1.0, a vision-language foundation model for autonomous driving, is introduced by arXiv researchers. This model integrates 3D perception, visual question answering, and motion planning within a unified framework. It includes a bird's-eye-view (BEV) perception head for 3D object detection and a Planning Expert for generating future ego trajectories, trained using a combination of driving supervision and general-purpose vision-language data.

arXiv / arxiv-explainability-av / Aug 31, 2026

Rethinking Language's Role in Efficient VLA for Autonomous Vehicles: Toward Smarter, Trustworthy Driving

Vision-Language-Action (VLA) models are reshaping autonomous driving by unifying perception, reasoning, and control through language, enabling semantic grounding and interpretable decisions. However, the computational cost of language models at inference time is a significant challenge. This work introduces the Language Residue taxonomy to categorize methods by their inference-time use of language, ranging from train-time-only supervision to full per-frame generation. The taxonomy aims to optimize the balance between computational efficiency and interpretability in VLA models for autonomous vehicles.

The Verge / automated-vehicles / Aug 27, 2026

In a swipe at Tesla, Waymo says ‘cameras… aren’t enough’

Waymo CEO John Krafcik stated that cameras alone are not enough for safe autonomous driving, directly challenging Tesla's camera-centric approach. Waymo's stance underscores the importance of multi-sensor fusion, including LiDAR and radar, in building reliable and safe autonomous vehicles.

The Verge / automated-vehicles / Aug 24, 2026

Robotaxis are real now — so is the pushback

The deployment of robotaxis by companies like Waymo, Tesla, and Uber is facing increasing public scrutiny and pushback. Concerns over safety, job displacement, and regulatory oversight are driving a growing debate about the role of autonomous vehicles in urban transportation.

Engadget / ai / Aug 22, 2026

OpenAI Calls For California To Strengthen Its AI Safety Laws

OpenAI has called for California to strengthen its AI safety laws, emphasizing the need for more stringent regulations to address growing concerns about AI safety. This advocacy reflects the increasing importance of regulatory frameworks in ensuring the responsible development and deployment of AI technologies.

TechCrunch / ai / Aug 22, 2026

OpenAI says California should strengthen its AI safety bill

OpenAI has urged California to strengthen its AI safety bill, advocating for stricter regulations to mitigate potential risks associated with AI. The organization's call for action underscores the need for comprehensive legal frameworks to ensure the safe and ethical development of AI, including applications in autonomous vehicles.

TechCrunch / automated-vehicles / Aug 21, 2026

Waymo hands over documents in NHTSA’s child collision probe

Waymo has submitted documents to the National Highway Traffic Safety Administration (NHTSA) as part of an ongoing investigation into a child collision incident. The documents are expected to provide insights into Waymo's safety protocols and the circumstances surrounding the incident.

arXiv / arxiv-llm-vlm-av / Aug 21, 2026

Roadside-Cooperative Autonomous Driving: From Data Platform to Vision-Language End-to-End Reasoning

Vehicle-to-Everything (V2X) cooperation enables beyond-line-of-sight perception, mitigating occlusions in single-vehicle sensing. However, existing V2X benchmarks provide limited support for closed-loop evaluation and language-grounded supervision, hindering the development of vision-language models (VLMs) for end-to-end cooperative driving. To address these limitations, we introduce V2XBench, a simulation platform featuring synchronized ego--roadside sensing and closed-loop evaluation, together with Chat-V2XBench, a progressively structured VQA dataset for cooperative reasoning.

InsideEVs / automated-vehicles / Aug 21, 2026

Nevada Opens The Door To Thousands Of Paid Robotaxis, And Tesla Has The Edge

Nevada has approved Uber, Tesla, and Waymo to launch paid robotaxi services, potentially allowing thousands of autonomous vehicles to operate in the state. This decision marks a significant step towards the commercialization of autonomous driving technology, providing a real-world testbed for these companies.

arXiv / arxiv-explainability-av / Aug 21, 2026

A Collaborative Multi-Modality Interaction for VLA-based End-to-End Autonomous Driving

Vision-Language-Action (VLA) models have emerged as a powerful paradigm for end-to-end autonomous driving by jointly integrating perception, reasoning, and decision making within a unified multimodal framework. However, most existing VLA models formulate end-to-end autonomous driving as a visual question answering task, leading to unreliable and less interpretable decision reasoning. In addition, they fail to establish effective multi-modal interaction across heterogeneous sensors, thereby limiting robust scene perception and reliable driving reasoning in long-tail driving scenarios.

Engadget / automated-vehicles / Aug 21, 2026

Nevada Allows Uber, Tesla And Waymo To Start Paid Robotaxi Service

Nevada has granted permission to Uber, Tesla, and Waymo to start paid robotaxi services. This regulatory approval marks a significant milestone in the commercial deployment of autonomous vehicles, allowing these companies to offer rides to the public for a fee.

TechCrunch / automated-vehicles / Aug 21, 2026

Tesla, Uber, and Waymo all get the OK to operate thousands of robotaxis in Nevada

Tesla, Uber, and Waymo have received approval to operate thousands of robotaxis in Nevada, marking a significant step towards the commercialization of autonomous driving technology. This decision will allow these companies to test and refine their autonomous systems in a real-world setting.

Engadget / automated-vehicles / Aug 20, 2026

Tesla's Robotaxi Fleet Might Finally Be Driving Around Austin Unsupervised

Tesla's robotaxi fleet is reportedly operating unsupervised in Austin, Texas. The vehicles, part of Tesla's Full Self-Driving (FSD) program, are being tested in real-world conditions, marking a significant step towards the company's goal of launching a fully autonomous ride-hailing service.

arXiv / arxiv-llm-vlm-av / Aug 20, 2026

Multi-Agent Orchestration with the Common-Sense Reasoning Capabilities of LLMs for Autonomous Driving

Autonomous vehicles require robust perception and decision-making capabilities to operate in diverse and unseen scenarios. While reinforcement learning and rule-based methods can provide effective control and safety mechanisms, their performance may degrade in situations requiring contextual reasoning. Large Language Models (LLMs) have demonstrated strong capabilities in understanding multimodal information and generating contextual reasoning, however, their use for direct vehicle control can introduce latency and hallucination risks.

Electrek / automated-vehicles / Aug 20, 2026

You can now ride a Chinese EV in the US as Waymo opens Zeekr Ojai to all

Waymo has opened its Zeekr Ojai electric vehicles to the public in the United States. These Chinese-made EVs, equipped with Waymo's autonomous driving technology, are now available for ride-hailing services, expanding the company's fleet and offering a new option for consumers.

The Verge / automated-vehicles / Aug 20, 2026

Tesla Robotaxis appear to go fully unsupervised in Austin ahead of Cybercab launch

Tesla's robotaxis in Austin, Texas, appear to be operating fully unsupervised ahead of the Cybercab launch. This development suggests that Tesla's Full Self-Driving (FSD) technology is progressing towards commercial readiness, with vehicles navigating urban environments without human intervention.

IEEE Open Journal of Intelligent Transportation Systems / journal-rss / Aug 20, 2026

What If I Cannot Always See: Misdetection-Aware MPC for Autonomous Vehicles

IEEE Open Journal of Intelligent Transportation Systems (2026) investigates misdetection-aware model predictive control (MPC) for autonomous vehicles. The paper identifies a critical detection-rate threshold below which collision risk increases sharply, particularly for conventional controllers. This research highlights the importance of addressing intermittent perception errors to ensure AV safety.

InsideEVs / automated-vehicles / Aug 20, 2026

Waymo’s New Robotaxi Minivans Are Now Taking Paying Passengers

Waymo's new robotaxi minivans have begun taking paying passengers. These vehicles, equipped with advanced autonomous driving technology, are now part of Waymo's commercial fleet, offering a new level of service to customers in select areas.

arXiv / arxiv-explainability-av / Aug 20, 2026

When Do LLM Agents Help? Deadline-Aware Mixed-Criticality Task Scheduling at the Autonomous-Vehicle Edge

Autonomous vehicles offload latency-sensitive perception tasks to nearby mobile edge computing (MEC) servers, where a missed safety-critical task is unsafe rather than merely degraded. Large language models (LLMs) are increasingly proposed as adaptive, explainable schedulers, yet evidence of when they help is scarce. We study deadline-aware, mixed-criticality scheduling on heterogeneous MEC servers, where time-critical (TC) tasks must be protected at a controlled cost to best-effort traffic, and ask whether a multi-agent LLM control layer improves on a strong heuristic.

TechCrunch / automated-vehicles / Aug 19, 2026

Waymo’s cheaper, next-gen robotaxi is now open to all riders in these three cities

Waymo has launched its next-generation robotaxi, which is now available to all riders in Phoenix, San Francisco, and Los Angeles. The new model is reportedly more affordable and features advanced autonomous driving technology, marking a significant step towards wider public adoption of robotaxis.

NVIDIA Developer / vlm / Aug 19, 2026

Building Federated Multimodal AI Workflows with NVIDIA FLARE | NVIDIA Technical Blog

NVIDIA has introduced a new framework, FLARE, for building federated multimodal AI workflows. This technology enables the training of AI models across multiple data sources while maintaining data privacy, which is crucial for advancing autonomous driving systems that require diverse and large datasets.

arXiv / arxiv-llm-vlm-av / Aug 18, 2026

GAPL: Grounded Action-effect Policy Learning for LLM-Based Trajectory Planning

Researchers from the University of Oxford proposed GAPL, a framework that integrates LLM-based effect estimation, simulation-based effect grounding, and policy optimization for trajectory planning in autonomous driving. GAPL addresses the limitations of LLMs in high-level reasoning and precise control, enhancing both interpretability and robustness in complex driving scenarios.

Electrek / automated-vehicles / Aug 18, 2026

Tesla ‘Robotaxi’ rams through bollards weeks after ‘impeccable’ claim

Tesla’s Robotaxi, which the company claimed to be 'impeccable,' was involved in an incident where it rammed through bollards. This event raises questions about the reliability and safety of Tesla's autonomous driving technology, particularly in real-world scenarios.

reuters.com / automated-vehicles / Aug 18, 2026

China's Pony.ai plans to deploy more than 4,000 robotaxis abroad

Pony.ai, a Chinese autonomous vehicle company, plans to deploy over 4,000 robotaxis in international markets. This ambitious expansion aims to bring advanced autonomous driving technology to a broader audience, potentially accelerating the global adoption of robotaxis.

arXiv / arxiv-llm-vlm-av / Aug 17, 2026

Structured Driving-State Narratives for Small Language Model-Based GNSS Spoofing Detection

A team from the University of Michigan developed a small language model (SLM)-based framework for detecting and classifying GNSS spoofing attacks in autonomous vehicles. The framework converts driving states from GNSS and other sensors into structured narratives for spoofing detection, achieving comparable performance to large language models (LLMs) on the same test set.

arXiv / arxiv-explainability-av / Aug 16, 2026

Identifying Confusion Trends in Concept-based XAI for Multi-Label Classification

Deep Neural Networks (DNNs) deployed in high-risk domains, such as healthcare and autonomous driving, must be not only accurate but also understandable to ensure user trust. In real-world computer vision tasks, these models often operate on complex images containing background noise and are heavily annotated. To make such models explainable, Concept-based Explainable AI (CXAI) methods need to be assessed for their applicability and problem-solving capacity.

Reuters / ai / Aug 14, 2026

EXCLUSIVE: Apple trains its own AI model for China market with Alibaba's support, sources say

NVIDIA has released DRIVE Sim 2.0, an updated version of its simulation software for autonomous vehicles, which includes improved simulation capabilities and support for more realistic scenarios. The software is intended to help developers test and validate their autonomous driving systems. This release is significant for the autonomous vehicle industry as it provides a new tool for developers to improve the performance of their systems. With DRIVE Sim 2.0, developers can simulate a wide range of driving scenarios and test their systems in a virtual environment.

arXiv / arxiv-llm-vlm-av / Aug 13, 2026

LLM-Assisted Dynamic Threat Analysis for Attacker-Reachable Software Weaknesses in Autonomous Vehicles

The arXiv paper proposes XCoT-VLA, a framework that replaces descriptive rationales with compact executable Chain-of-Thought tokens learned from automatically constructed Reason-Action supervision. XCoT-VLA connects scene understanding, semantic reasoning, and trajectory generation for autonomous driving, and the predicted XCoT sequence remains in context and conditions fixed trajectory queries through shared multimodal self-attention. This approach is crucial for real-time control in autonomous driving. The XCoT-VLA model is evaluated on logged trajectories, providing action evidence, while scene context supplies causal semantics.

IEEE Transactions on Intelligent Transportation Systems / journal-rss / Aug 13, 2026

Edge Case Detection in Automated Driving: Methods, Challenges, and Future Directions

IEEE Transactions on Intelligent Transportation Systems published a comprehensive survey by a research team on edge case detection in automated driving. The paper classifies and reviews various methodologies for detecting and assessing edge cases across different AV modules, including perception, prediction, planning, and control. The study highlights the importance of addressing rare and unexpected situations to enhance the reliability and safety of AVs in real-world conditions.

arXiv / arxiv-llm-vlm-av / Aug 12, 2026

Drive-to-Music: Context-Aware Generative Audio for In-Vehicle Experiences

The arXiv paper presents BrainWAM, a unified planner that leverages both semantic priors and predictive dynamics for autonomous driving. BrainWAM coordinates action-space semantic priors and predictive dynamics through a novel attention-allocation mechanism, inspired by neuroscience evidence on complex behavior. This approach addresses the attention-allocation mismatch in naive combinations of Vision-Language-Action models and World Action Models. The BrainWAM model is designed to provide a more comprehensive understanding of the driving environment, enhancing the decision-making capabilities of autonomous vehicles.

TechCrunch / ai / Aug 12, 2026

As AI safety concerns mount, three pioneers make the case for staying open

Three AI pioneers have made a case for maintaining open research and development practices despite growing concerns over AI safety. They argue that transparency and collaboration are essential for advancing AI technology responsibly and addressing potential risks. This perspective highlights the ongoing debate in the AI community about the balance between innovation and safety.

arXiv / arxiv-llm-vlm-av / Aug 11, 2026

XCoT-VLA: Executable Chain-of-Thought for Vision-Language-Action Driving

XCoT-VLA, proposed by arXiv researchers, is a novel executable Chain-of-Thought model for Vision-Language-Action driving, replacing descriptive rationales with compact executable tokens learned from automatically constructed Reason-Action supervision. This approach enables real-time control and improves the efficiency of vision-language models in autonomous driving. The model's performance is evaluated using logged trajectories and scene context, demonstrating its potential for robust autonomous driving. With XCoT-VLA, the predicted sequence remains in context and conditions fixed trajectory queries through shared multimodal self-attention.

Electrek / automated-vehicles / Aug 10, 2026

Tesla unveils first Cybercab with Starlink antenna integrated

Tesla unveiled its first Cybercab, featuring an integrated Starlink antenna for enhanced connectivity. This development underscores Tesla's commitment to integrating advanced communication technologies into its autonomous vehicle lineup.

Transportation Research Part F: Traffic Psychology and Behaviour / journal-rss / Aug 10, 2026

How do emotional cues influence drivers' pre-takeover readiness in automated driving? Evidence from olfaction-involved affective multimodal interaction

Yinjie Lv, Chunhui Jing, Guanyu Lei, and Teng Zhao from Transportation Research Part F: Traffic Psychology and Behaviour published a study on August 2026, investigating how emotional cues influence drivers' pre-takeover readiness in automated driving. The study explores the impact of olfaction-involved affective multimodal interaction on driver behavior. This research is crucial for understanding human-vehicle interaction and trust in autonomous vehicles. By examining the effects of emotional cues, the study provides insights into the complex relationships between human emotions, vehicle automation, and driver readiness.

arXiv / arxiv-llm-vlm-av / Aug 10, 2026

CRUISE: Vision-Language Model-Guided Uncertainty-Aware Cross-Modal Sensor Fusion for Robust Autonomous Driving

The arXiv paper proposes CRUISE, a novel uncertainty-aware cross-modal sensor fusion framework that integrates a vision-language model to address the challenge of robust cross-modal feature fusion in autonomous driving. CRUISE allows models to prioritize reliable signals and mitigate the issue of varying sensor reliability across diverse driving conditions. By leveraging uncertainty quantification, CRUISE enables effective fusion of features from multiple sensors, such as cameras, LiDAR, and radar, to improve the overall perception and decision-making capabilities of autonomous vehicles. This approach has significant implications for enhancing the safety and reliability of autonomous...

TechCrunch / ai / Aug 9, 2026

The AI safety test is becoming a safety risk

The AI safety test, designed to ensure the reliability of AI systems, is becoming a safety risk itself. Experts argue that the current testing methods are inadequate and may lead to false assurances, highlighting the need for more rigorous and comprehensive safety protocols in AI development.

Electrek / automated-vehicles / Aug 6, 2026

Tesla is racking up speeding tickets, and YOU are on the hook

Tesla vehicles equipped with Full Self-Driving (FSD) capabilities have been accumulating speeding tickets, raising concerns about the system's reliability and safety. The financial and legal implications of these incidents underscore the need for robust safety measures and regulatory oversight in autonomous vehicle technology.

arXiv / arxiv-llm-vlm-av / Aug 6, 2026

MEC-Patch: Visible-Infrared Cross-Modal Adversarial Attack Driven by Intrinsic Material Emissivity Laws

The arXiv paper 'MEC-Patch: Visible-Infrared Cross-Modal Adversarial Attack Driven by Intrinsic Material Emissivity Laws' proposes MEC-Patch, a cross-modal adversarial attack framework driven by intrinsic physical laws, specifically the Stefan-Boltzmann Law. MEC-Patch establishes a physics-grounded cross-spectral model to evaluate the cross-modal adversarial robustness of visible-infrared multimodal perception systems in autonomous driving. This framework addresses the limitations of existing approaches that simplify infrared modeling or neglect ambient temperature fluctuations. By leveraging physical laws, MEC-Patch provides a more realistic and effective method for assessing the...

reuters.com / automated-vehicles / Aug 5, 2026

Amazon's Zoox to start paid robotaxi rides in Vegas next week

Amazon's Zoox will initiate paid robotaxi services in Las Vegas starting next week. This deployment marks a significant milestone for Zoox, expanding its autonomous vehicle operations to a new city and offering real-world rides to the public.

TechCrunch / automated-vehicles / Aug 4, 2026

Waymo opens up robotaxi service in Dallas to everyone

NVIDIA has released a new AI model for autonomous vehicle perception, which is designed to improve the accuracy and efficiency of object detection and tracking. This model is part of NVIDIA's efforts to provide advanced AI models for the autonomous vehicle industry, and it is expected to accelerate the development of autonomous driving technology. The new AI model uses a combination of computer vision and machine learning algorithms to enable accurate and efficient object detection and tracking. NVIDIA's release of this model is significant for the industry as it provides a powerful tool for companies to improve their autonomous driving capabilities.

Electrek / automated-vehicles / Aug 4, 2026

Waymo CEO explains why Tesla’s camera-only self-driving falls short

Waymo is expanding its robotaxi service to Los Angeles, with plans to launch a fully autonomous ride-hailing service in the city. This expansion marks a significant milestone for Waymo, which has been testing its autonomous vehicles in various locations. The company's decision to launch its service in Los Angeles demonstrates its confidence in its technology and its ability to operate safely and efficiently in complex urban environments. Waymo's expansion is expected to increase the adoption of autonomous ride-hailing services and provide a new transportation option for residents and visitors in Los Angeles.

NVIDIA Blog / automated-vehicles / Aug 4, 2026

NVIDIA Alpamayo 2 Super, the Frontier Open Model for Robotaxis and Autonomous Vehicles, Now Available for Commercial Use

NVIDIA has made its Alpamayo 2 Super model available for commercial use, with Zoox engineers already working on integrating it into their robotaxi fleet. This model is designed for robotaxis and autonomous vehicles, and its availability is expected to accelerate the development of autonomous driving technology. The Alpamayo 2 Super model is part of NVIDIA's Frontier Open Model initiative, which aims to provide open and accessible AI models for the autonomous vehicle industry. This development is significant for the industry as it provides a powerful tool for companies to improve their autonomous driving capabilities.

Reuters / ai / Aug 3, 2026

Meta, Anthropic, Google, OpenAI to meet Trump officials about AI safety testing

Meta, Anthropic, Google, and OpenAI are scheduled to meet with Trump officials to discuss AI safety testing. The meeting comes in the wake of recent incidents, including an AI agent going rogue, and aims to establish voluntary safety standards. This collaboration between tech giants and government officials highlights the growing importance of AI regulation and safety.

arXiv / arxiv-llm-vlm-av / Aug 3, 2026

MoRAL: Sensor-Grounded BEV Reasoning for Compact VLMs toward Edge-Oriented Autonomous Driving

MoRAL, a two-stage fine-tuning pipeline, teaches Cosmos-Reason2-2B to read a physics-encoded Bird's Eye View (BEV) representation and reason over it for driving decisions, enabling compact model size and reliable metric grounding for autonomous driving. The BEV image encodes LiDAR metric distance, object class, and radar Doppler velocity, externalizing spatial perception into the input image. Stage 1 fine-tunes the vision encoder on 60,000 grounding records, and MoRAL achieves zero-shot baselines, making it suitable for resource-constrained autonomous driving platforms. This approach is crucial for safety-critical spatial reasoning in autonomous vehicles.

arXiv / arxiv-llm-vlm-av / Aug 3, 2026

TALSC: Timeliness-Aware Large-Small VLM Collaboration for Infrastructure-Assisted Autonomous Driving

TALSC, a Timeliness-Aware Large-Small VLM Collaboration framework, balances the accuracy gains of large VLMs with their latency-induced timeliness degradation, enabling collaboration between large VLMs at edge servers and small VLMs on-board vehicles. TALSC models the Age of Information (AoI) evolution for VLM inference and characterizes the coupling between AoI and VLM performance, making it essential for infrastructure-assisted autonomous driving. This framework is critical for dynamic vehicular environments where the utility of sensory data decays rapidly. By addressing the timeliness concern, TALSC improves the overall performance and safety of autonomous driving systems.

gizmodo.com / ai / Aug 2, 2026

OpenAI Smuggled the Announcement of Astra, Its Next AI Model, Into a Blog Post About Math

OpenAI has quietly announced the release of Astra, its next AI model, in a blog post about mathematical advancements. Astra is expected to build on the capabilities of previous models like GPT-4, potentially offering improved performance and efficiency. The discreet announcement highlights the company's strategy to balance innovation with public perception and regulatory concerns.

TechCrunch / automated-vehicles / Aug 1, 2026

Uber is building an autonomous vehicle empire, and here’s every company it’s using to do it

Tesla has released FSD Beta 11.4.1, which includes improved vision and enhanced safety features. This update is available to a limited number of users in the US and Canada, and it marks a significant improvement in Tesla's autonomous driving technology. The updated FSD Beta 11.4.1 includes new features and capabilities that are designed to improve the safety and efficiency of Tesla's autonomous vehicles. Tesla's release of FSD Beta 11.4.1 demonstrates the company's ongoing efforts to improve its autonomous driving technology and provide a safe and efficient transportation option for its customers.

Transportation Research Part C: Emerging Technologies / journal-rss / Jul 31, 2026

Semantics-guided hierarchical decision-making for autonomous driving via LLM-assisted reinforcement learning

The Transportation Research Part C: Emerging Technologies paper proposes a semantics-guided hierarchical decision-making framework for autonomous driving via LLM-assisted reinforcement learning, which enables the model to make decisions based on semantic information. The framework uses a large language model to guide the decision-making process and provide interpretable results. The authors demonstrate the effectiveness of the framework in improving the decision-making process in autonomous driving. The proposed framework can be used to improve the safety and efficiency of autonomous vehicles. The use of LLM-assisted reinforcement learning enables the model to learn from experience and...

arXiv / arxiv-explainability-av / Jul 31, 2026

Outcome-Guided Distillation: A Teacher-Student Framework to Advance VLM Reasoning in Autonomous Driving

The Outcome-Guided Distillation framework, proposed by researchers on arXiv, integrates structured reasoning and geometric precision through a teacher-student architecture for Vision-Language Models in autonomous driving. The teacher model generates logical explanations and reflectively refines the reasoning, enhancing interpretability and driving robustness. This framework addresses the limitations of pre-generated annotations and costly human labor, providing a new approach for advancing VLM reasoning in autonomous driving. The Outcome-Guided Distillation framework demonstrates its potential for improving the safety and trustworthiness of autonomous vehicles.

Electrek / automated-vehicles / Jul 29, 2026

Waymo adds Google’s Gemini AI assistant and new UI to Ojai robotaxi

Waymo has integrated Google’s Gemini AI assistant and a new user interface into its Ojai robotaxi service. The enhancements aim to improve passenger interaction and overall user experience. This deployment showcases Waymo's ongoing efforts to refine its robotaxi technology and enhance customer satisfaction.

The Verge / ai / Jul 29, 2026

We’re running out of reasons to ignore AI safety

A recent article argues that the urgency of addressing AI safety cannot be ignored. The piece highlights the increasing risks associated with AI, including potential failures in autonomous systems. This discussion is particularly relevant to the AV industry, where safety is a paramount concern and the consequences of AI failures can be severe.

Engadget / automated-vehicles / Jul 28, 2026

Fired Tesla Manager Says Full Self-Driving Cars Were 'Rolling Hazards'

A former Tesla manager, who was fired, claims that Tesla's Full Self-Driving (FSD) cars were 'rolling hazards' during testing. The manager's allegations highlight safety concerns and potential risks associated with the deployment of FSD technology.

Electrek / automated-vehicles / Jul 28, 2026

Tesla self-driving manager describes scary conditions in robotaxi testing

A Tesla self-driving manager described the conditions during robotaxi testing as 'scary,' raising concerns about the safety and reliability of the company's autonomous driving technology. The manager's account highlights the challenges and potential risks in the development and deployment of FSD systems.

Electrek / automated-vehicles / Jul 28, 2026

Baidu’s Apollo Go begins London robotaxi testing with Uber and Lyft

Baidu’s Apollo Go has begun robotaxi testing in London in collaboration with Uber and Lyft. This deployment marks a significant expansion of Baidu's autonomous driving technology into international markets, demonstrating the growing global reach of AV services.

ZDNET / ai / Jul 28, 2026

Microsoft's new AI model beats Mythos on security benchmark

Microsoft's new AI model has outperformed Mythos on a security benchmark, showcasing advancements in AI safety and security. This development is crucial for ensuring the reliability and trustworthiness of AI systems, particularly in critical applications like autonomous vehicles.

arXiv / arxiv-llm-vlm-av / Jul 27, 2026

Development of Vision-Language Model-based GNSS Spoofing Detection for Autonomous Vehicle Navigation

The arXiv paper proposes CARA, a concept-aware risk attention framework for interpretable collision anticipation in autonomous driving, which derives domain-grounded risk concepts from accident narratives and aligns them with video frames via vision-language similarity. CARA organizes these concepts into evolving concept trajectories, providing explicit risk evidence that can be used for collision anticipation. The framework is intrinsically interpretable and can be used for spatio-temporal reasoning in dynamic driving scenes. The authors demonstrate the effectiveness of CARA in anticipating collisions and providing interpretable reasoning.

arXiv / arxiv-llm-vlm-av / Jul 26, 2026

ObsDriveBench: Benchmarking Multimodal Understanding under Adverse Weather with Observability Awareness

The arXiv paper introduces ObsDriveBench, a real-world multi-modal benchmark for adverse-weather autonomous driving, which evaluates vision-language models under real-world adverse weather conditions with multi-modal inputs. ObsDriveBench is designed with three capability dimensions: observability, cross-modal consistency, and scene understanding. The benchmark is essential for evaluating the performance of autonomous driving systems in adverse weather conditions, which is a critical challenge in the development of autonomous vehicles. The ObsDriveBench dataset provides a comprehensive evaluation of vision-language models in adverse weather conditions, enabling the development of more...

Reuters / automated-vehicles / Jul 24, 2026

Waymo explores ending Uber partnership, Financial Times reports

Waymo is reportedly considering ending its partnership with Uber, a move that could signal a strategic shift in the company's approach to robotaxi services. The partnership, which began in 2021, has been a key part of Waymo's strategy to expand its autonomous vehicle operations.

arXiv / arxiv-explainability-av / Jul 24, 2026

Explainable Reinforcement Learning for assisting Air Traffic Controllers

The arXiv paper proposes CARA, an intrinsically interpretable spatio-temporal framework for collision anticipation in autonomous driving. CARA derives domain-grounded risk concepts from accident narratives and aligns them with video frames via vision-language similarity, providing explicit risk evidence. The framework organizes risk concepts into evolving concept trajectories, enabling the anticipation of potential collisions. CARA's approach addresses the need for interpretable reasoning in autonomous driving, making it a valuable tool for improving safety and trust in autonomous vehicles. The paper demonstrates the effectiveness of CARA in anticipating collisions and providing insights...

arXiv / arxiv-explainability-av / Jul 24, 2026

CARA: Concept-Aware Risk Attention for Interpretable Collision Anticipation

The researchers at arXiv proposed CARA, an intrinsically interpretable spatio-temporal framework for collision anticipation in autonomous driving, which derives domain-grounded risk concepts from accident narratives and aligns them with video frames via vision-language similarity. CARA organizes these concepts into evolving concept trajectories that provide explicit risk evidence, enabling interpretable reasoning about what risk factors are being tracked and how risk evolves over time. This framework addresses the limitations of existing methods, which are often opaque or lack fidelity. By providing a clear understanding of the risk factors, CARA can improve the safety and reliability of...

Reuters / ai / Jul 24, 2026

Anthropic rolls out Opus 5 AI model in efficiency upgrade

Anthropic has launched Claude Opus 5, a more efficient AI model designed for coding, agents, and enterprise workflows. The new model is cheaper and more powerful, making it a significant upgrade for developers and businesses. This advancement in AI efficiency could have implications for the development of autonomous driving systems, particularly in areas like perception and decision-making.

VentureBeat / ai / Jul 24, 2026

Anthropic launches Claude Opus 5, a cheaper AI model for coding, agents and enterprise workflows

Zoox has unveiled a new autonomous vehicle design with enhanced safety features, including a unique bi-directional design and a robust sensor suite. The new design is intended to provide a safe and comfortable ride for passengers, while also reducing the risk of accidents. This development is relevant for AV research as it demonstrates the importance of safety considerations in the design of autonomous vehicles. Zoox's new autonomous vehicle design is expected to be deployed in the company's ride-hailing service in the near future.

arXiv / arxiv-explainability-av / Jul 24, 2026

CommandLM: Data driven behavior level descriptor for ego vehicles

The arXiv paper introduces CommandLM, a multimodal large language model that generates human-readable behavior descriptions for ego vehicles from fused multi-sensor data, achieving CIDEr 0.67 and BERT-F1 0.88 on the CommandLM-nuScenes dataset. CommandLM processes temporally fused bird's-eye view representations from LiDAR and multi-camera inputs via a Q-Former adapter connected to a quantized, LoRA-fine-tuned large language model. This model is essential for safety, trust, and regulation in autonomous driving systems. The CommandLM-nuScenes dataset is used to train CommandLM, which produces intent-aware, interpretable captions suitable for planner supervision and safety auditing.

Reuters / automated-vehicles / Jul 23, 2026

Tesla's once-bullish tone on robotaxis shifts

Tesla's tone on its robotaxi ambitions has shifted from bullish to more cautious. The company's recent statements and actions suggest a more realistic approach to the deployment of its autonomous vehicle fleet, reflecting the complexities and regulatory hurdles in the AV industry.

TechCrunch / automated-vehicles / Jul 23, 2026

Tesla’s robotaxis are moving in reverse

Tesla's robotaxi program is facing significant challenges, with the company's own data showing a lack of progress. Despite initial promises, the expansion of Tesla's autonomous ride-hailing service has not materialized, raising questions about the feasibility of its robotaxi vision.

arXiv / arxiv-explainability-av / Jul 23, 2026

HyWorldVLA: A Vision-Language-Action Model with Hybrid World Modeling for Autonomous Driving

The arXiv paper 'HyWorldVLA: A Vision-Language-Action Model with Hybrid World Modeling for Autonomous Driving' introduces a hybrid world-VLA framework that unifies pixel-level supervision and latent representation learning for end-to-end autonomous driving. The proposed HyWorldVLA model predicts video latents encoded by a pre-trained video VAE and reconstructs video frames to provide precise pixel-level grounding. This approach aims to reconcile the trade-off between robustness and interpretability in autonomous driving. By combining the strengths of pixel-level and latent-based world models, HyWorldVLA can improve the performance and safety of autonomous vehicles.

The Verge / automated-vehicles / Jul 23, 2026

Tesla’s robotaxi promises are clashing with reality

The head of the US AI safety agency has resigned, citing concerns over the regulatory framework's ability to keep pace with rapid advancements in AI technology. This departure underscores the growing tensions and challenges in ensuring the safe deployment of AI systems, including those in autonomous vehicles.

Electrek / automated-vehicles / Jul 23, 2026

Tesla still has no real plan for Hardware 3 owners promised FSD

Waymo is expanding its autonomous ride-hailing service to Los Angeles, marking the company's first major expansion outside of Phoenix, Arizona, with the service currently available in a limited area of Los Angeles. Waymo's autonomous vehicles have been tested extensively in Los Angeles and have demonstrated a high level of safety and reliability. This expansion is significant for AV research as it showcases the feasibility of autonomous ride-hailing services in new and complex urban environments. The service's availability in Los Angeles is a key step towards the widespread adoption of autonomous transportation.

Electrek / automated-vehicles / Jul 22, 2026

Tesla says ‘Robotaxi’ is expanding — its own chart shows it’s not

Tesla claims its Robotaxi service is expanding, but its own data shows limited growth. The company's chart indicates that the number of active Robotaxis has not increased significantly, raising questions about the progress of Tesla's autonomous ride-hailing ambitions. This discrepancy highlights the challenges in scaling autonomous vehicle fleets.

Reuters / automated-vehicles / Jul 22, 2026

Volkswagen's CARIZON and Horizon Robotics deepen self-driving partnership in China

Volkswagen's CARIZON and Horizon Robotics are expanding their collaboration on self-driving technology in China, with a focus on Level 4 autonomous vehicles. The partnership will enable the development and deployment of Level 4 autonomous vehicles in China, with the goal of commercializing the technology by 2028. This development is relevant for AV research as it demonstrates the growing importance of international collaborations in advancing autonomous driving technology. The partnership between Volkswagen's CARIZON and Horizon Robotics is a key step towards the widespread adoption of autonomous vehicles in China.

TechCrunch / automated-vehicles / Jul 21, 2026

Tesla spins up robotaxi pilots in Orlando and Tampa ahead of Q2 earnings

Tesla has launched robotaxi pilots in Orlando and Tampa, Florida, ahead of its Q2 earnings report. The company is expanding its autonomous vehicle service to new cities, aiming to demonstrate the readiness and reliability of its Full Self-Driving (FSD) technology. This deployment is significant for the advancement of autonomous ride-hailing services.

arXiv / arxiv-llm-vlm-av / Jul 21, 2026

Cognitive Dual-Process Planning for Autonomous Driving with Structured Scene Knowledge and Verifiable Reasoning-Action Consistency

The arXiv paper 'Cognitive Dual-Process Planning for Autonomous Driving with Structured Scene Knowledge and Verifiable Reasoning-Action Consistency' presents a cognitive dual-process planning framework that integrates perception foundation models, critical-path filtering, and an expert vision-language model to generate structured chain-of-thought supervision for high-level planning in autonomous driving. This approach aims to address the constraints of costly structured supervision and inconsistencies between generated rationales and driving actions. The framework represents planning-relevant scene knowledge in a machine-parsable structured chain-of-thought schema. The automated data...

arXiv / arxiv-llm-vlm-av / Jul 21, 2026

Cognitive Dual-Process Planning for Autonomous Driving with Structured Scene Knowledge and Verifiable Reasoning-Action Consistency

The arXiv paper 'Cognitive Dual-Process Planning for Autonomous Driving with Structured Scene Knowledge and Verifiable Reasoning-Action Consistency' presents a cognitive dual-process planning framework for autonomous driving, which represents planning-relevant scene knowledge in a machine-parsable structured chain-of-thought schema. This framework integrates perception foundation models, critical-path filtering, and an expert vision-language model to generate supervision without costly structured annotation. The approach enables more accurate and interpretable planning for autonomous vehicles. The framework's use of structured scene knowledge and verifiable reasoning-action consistency is...

Electrek / automated-vehicles / Jul 21, 2026

Tesla adds Robotaxi in Tampa and Orlando as Austin stalls at 17 cars

Tesla has added robotaxi services in Tampa and Orlando, Florida, while the Austin deployment remains limited to 17 cars. The expansion in Florida highlights Tesla's strategic push to scale its autonomous vehicle operations and gather more real-world data to improve its FSD technology. This move is crucial for the company's goal of achieving widespread commercial deployment.

Reuters / automated-vehicles / Jul 21, 2026

Tesla expands robotaxi service to Orlando, Tampa ahead of earnings

Tesla has expanded its robotaxi service to Orlando and Tampa, Florida, ahead of its Q2 earnings announcement. The company is rolling out its autonomous ride-hailing service to new cities to showcase the capabilities of its Full Self-Driving (FSD) technology. This expansion is a significant step towards achieving commercial viability in the robotaxi market.

Reuters / ai / Jul 20, 2026

Head of US AI safety agency resigns

The head of the US AI safety agency has resigned, raising concerns about the future of AI regulation and safety oversight. The resignation comes at a critical time as the US advances AI safety measures through state and federal actions.

Electrek / automated-vehicles / Jul 20, 2026

Tesla ‘Full Self-Driving’ got me a speeding ticket, and Tesla killed the fix

A Tesla Full Self-Driving (FSD) system reportedly led to a speeding ticket for the driver. Tesla subsequently disabled a feature that allowed the system to adjust speed based on traffic conditions, which was intended to prevent such issues. This incident highlights the ongoing challenges and regulatory scrutiny surrounding autonomous driving systems.

arXiv / arxiv-llm-vlm-av / Jul 20, 2026

A2RL V\textsubscript{max}: The A2RL autonomous racing dataset for long-range, high-speed perception and multi-vehicle interaction

The arXiv paper 'D3VL: Understanding Driving Scenes from 3D Time Series Data and Video with Language Models' proposes a novel MLLM framework that integrates 2D and 3D data for autonomous driving, using a pre-trained video VAE to encode video latents and a global semantic codebook to unify observations into a shared vocabulary. Published on July 21, 2026, this study demonstrates the effectiveness of MLLMs in using 3D sensors, particularly LiDAR and stereo cameras, for end-to-end autonomous driving. The D3VL framework improves performance in traditional 3D scene perception tasks, with a score of 96 for its relevance to AV research, highlighting the importance of multimodal perception in...

Transportation Research Part C: Emerging Technologies / journal-rss / Jul 19, 2026

Large language models for enhancing human-machine interaction in autonomous vehicles: A comprehensive survey

The journal paper Large language models for enhancing human-machine interaction in autonomous vehicles, published in Transportation Research Part C: Emerging Technologies on July 19, 2026, provides a comprehensive survey of the applications of large language models in autonomous vehicles. The authors, Zixuan Xu, Zihao Sheng, Sikai Chen, and Tiantian Chen, discuss the potential of large language models to improve human-machine interaction in autonomous vehicles. This survey is relevant for autonomous driving as it highlights the importance of human-machine interaction in ensuring safe and efficient transportation. The paper provides a thorough review of the current state of research in...

AP News / ai / Jul 17, 2026

Chinese AI model takes US tech industry by surprise with abilities rivaling Claude and ChatGPT

A new Chinese AI model has surprised the US tech industry with capabilities rivaling those of OpenAI's Claude and ChatGPT. The model, developed by a Chinese tech company, demonstrates advanced natural language processing and reasoning abilities, challenging the dominance of US-based AI models. This development underscores the growing competition in the global AI landscape.

arXiv / arxiv-llm-vlm-av / Jul 17, 2026

Vision-Language Assistant for Emotional Reactions to Risky Driving

The arXiv study introduces a vision-language pipeline, Keep Yelling Assistant (KYA), which detects risky driving behaviors and generates emotionally expressive responses to support driver awareness and comfort. KYA uses YOLOv8 variants to detect nearby vehicles and identify risky behaviors such as sudden cut-ins. The framework comprises two core modules, a vision module and a language module, to produce emotional responses tailored to driver preferences. This study matters for AV research as it highlights the importance of considering the emotional dimension and real-world user experience in autonomous driving systems. With KYA, drivers can receive real-time feedback on their driving...

Reuters / automated-vehicles / Jul 17, 2026

Zoox recalls self-driving cars because they may not detect smoke

Zoox has recalled its self-driving cars due to a potential issue where the vehicles may fail to detect smoke. The recall affects all Zoox vehicles currently in operation, highlighting a critical safety concern in the deployment of autonomous vehicles.

arXiv / arxiv-llm-vlm-av / Jul 17, 2026

Think at 5 Hz, Act at 20 Hz: Asynchronous Fast-Slow Vision-Language-Action Inference for Closed-Loop Driving

The arXiv study presents a fast-slow architecture for closed-loop driving, which removes the compromise of invoking the model on alternate simulation ticks and replaying the previous command in between. A frozen 7B vision-language backbone acts as the slow system, digesting navigation instructions and visual history at low frequency, while a lightweight action expert acts as the fast system, attending to the cache and the current camera frame at every simulation tick. This architecture is relevant for AV research as it enables more efficient and effective closed-loop driving, allowing for faster and more accurate decision-making. The study's findings can inform the development of more...

Engadget / automated-vehicles / Jul 16, 2026

NTSB Investigators Confirm Tesla Driver Overrode Full Self-Driving System In Fatal Crash

NTSB investigators have confirmed that a Tesla driver overrode the Full Self-Driving System in a fatal crash, highlighting the need for improved safety measures in autonomous vehicles. The incident occurred on a specific date and location, which has not been disclosed. Tesla's FSD system has been under scrutiny for its limitations and potential risks. This incident underscores the importance of addressing these concerns to ensure public safety.

arXiv / arxiv-llm-vlm-av / Jul 16, 2026

WorkDrive: Roadwork Chain of Causation for Autonomous Driving

The arXiv paper 'WorkDrive: Roadwork Chain of Causation for Autonomous Driving' proposes a framework that constructs perception-grounded causal reasoning for work zones and aligns it with trajectory prediction. WorkDrive redirects the annotator's attention to domain-specific work-zone observations, enabling the model to connect work-zone observations to correct planning decisions. This paper was published on July 16, 2026, and is relevant for AV research as it enhances the model's ability to handle roadwork zones. The proposed framework improves the model's performance in work zones by providing explicit guidance on work-zone observations.

arXiv / arxiv-llm-vlm-av / Jul 14, 2026

Adaptive Cross-Modal Fusion with Sparse Attention for Pedestrian Crossing Intention Prediction

The arXiv paper titled Adaptive Cross-Modal Fusion with Sparse Attention for Pedestrian Crossing Intention Prediction proposes a multimodal framework called ADAPT for predicting pedestrian crossing intention. ADAPT jointly models local and global visual context with temporal motion dynamics, processing four spatially aligned visual modalities and ego-vehicle speed, pedestrian bounding boxes, and skeleton pose information. This framework is relevant for AV research as it improves pedestrian crossing intention prediction, a safety-critical task for autonomous driving. The paper was published on July 14, 2026.

Electrek / automated-vehicles / Jul 13, 2026

Tesla says it’s building a wheelchair-accessible robotaxi in Texas

Tesla is developing a wheelchair-accessible robotaxi at its Giga Texas facility, emphasizing the company's commitment to expanding its autonomous vehicle offerings. The vehicle is currently in the testing phase, focusing on enhancing accessibility and safety for all users.

WIRED / automated-vehicles / Jul 13, 2026

Tesla Says It’s Building a Wheelchair-Accessible Robotaxi

Tesla is building a wheelchair-accessible robotaxi at its Giga Texas facility, aiming to enhance accessibility in autonomous transportation. The vehicle is currently in the testing phase, with a focus on safety and usability for all users.

IEEE Transactions on Intelligent Transportation Systems / journal-rss / Jul 13, 2026

Crosswalk Trajectory Prediction via an Intent-Trajectory Integrated Framework for Autonomous Driving

The increasing complexity of automated urban environments demands accurate modeling of pedestrian crossing behavior. Existing approaches often rely on human annotations or vehicle-centric views, limiting their ability to predict pedestrian intentions in dynamic scenarios. In this paper, we propose an Intent-Trajectory Integrated Pedestrian Crossing Prediction framework (ITI-PCP) that explicitly couples high-level semantic intention inference with multimodal trajectory generation.

TechCrunch / automated-vehicles / Jul 13, 2026

Uber’s robotaxi lobbying effort puts it on a collision course with Waymo

Uber's robotaxi lobbying efforts are creating tensions with Waymo, as both companies vie for regulatory approval and market dominance in autonomous vehicle services. The conflict highlights the competitive landscape and regulatory challenges in the AV industry.

Electrek / automated-vehicles / Jul 11, 2026

Tesla claims Cybercab driving employees at Giga Texas — in a parking lot

Tesla employees at Giga Texas are reportedly testing the Cybercab, a robotaxi version of the Cybertruck, in a parking lot. This deployment, while limited, marks a significant step towards the development and potential real-world application of Tesla's autonomous driving technology.

The Guardian / ai / Jul 10, 2026

Apple sues OpenAI, alleging artificial intelligence company stole trade secrets

Apple has filed a lawsuit against OpenAI, alleging that the artificial intelligence company stole trade secrets. The lawsuit claims that OpenAI misappropriated proprietary information and technology, which could have significant implications for the development and deployment of AI models in the industry.

InsideEVs / automated-vehicles / Jul 9, 2026

New Jersey Law Would Outlaw Tesla's Camera-Based Robotaxi Tech

New Jersey lawmakers have introduced a bill that would ban Tesla's camera-only robotaxi technology, citing safety concerns. The legislation specifically targets Tesla's approach, which relies solely on cameras, as opposed to multi-sensor systems used by other companies like Waymo.

Electrek / automated-vehicles / Jul 9, 2026

New bill would ban Tesla’s camera-only Robotaxi, not Waymo

A new bill in New Jersey aims to ban Tesla's camera-only robotaxis, while allowing other autonomous vehicle operators like Waymo. The legislation targets Tesla's reliance on cameras, which is seen as less safe compared to multi-sensor systems used by competitors.

WIRED / automated-vehicles / Jul 9, 2026

Self-Driving Cars Are Interfering With First Responders. Feds Aren’t Happy

The National Highway Traffic Safety Administration (NHTSA) has issued a warning to autonomous vehicle operators, including Waymo and Cruise, regarding their vehicles interfering with first responders. The federal agency is concerned about the safety implications and potential delays in emergency response times.

arXiv / arxiv-llm-vlm-av / Jul 9, 2026

Large Multimodal Model-Based Environment-Aware Mobility Management

The arXiv paper 'Large Multimodal Model-Based Environment-Aware Mobility Management' proposes an environment-aware mobility management scheme based on large multimodal models, which extend the capabilities of large language models to process multimodal sensing data. The proposed scheme extracts contextual information on the surrounding environment and predicts dynamic user trajectories, making real-time handover decisions across densely deployed small base stations. This approach is relevant for autonomous vehicles as it enables efficient mobility management, which is crucial for safe and efficient navigation. The proposed scheme has the potential to improve the performance of autonomous...

arXiv / arxiv-llm-vlm-av / Jul 9, 2026

Post-Training in End-to-End Autonomous Driving

Researchers on arXiv have proposed a post-training approach for end-to-end autonomous driving models, which aims to address the limitations of traditional machine learning applications in safety-critical and interaction-intensive environments, such as autonomous vehicles, by incorporating enhancements that enable the model to handle occlusions and recover from execution errors, and is relevant for AV research as it has the potential to improve the reliability and safety of autonomous vehicles. The approach is designed to handle the challenges of autonomous driving, including the accumulation of small execution errors and the scarcity of recovery behaviors in training data. With its focus...

The Verge / automated-vehicles / Jul 8, 2026

The robotaxi law that could ban Tesla

A proposed robotaxi law in California could potentially ban Tesla from operating its robotaxi service. The law aims to ensure safety and regulatory compliance for autonomous vehicles. This legislation highlights the ongoing regulatory challenges faced by companies like Tesla as they seek to deploy autonomous vehicles at scale.

arXiv / arxiv-explainability-av / Jul 7, 2026

Driving the Wrong Way: Leveraging Interpretability in End2End Autonomous Driving Models

The arXiv paper 'Driving the Wrong Way: Leveraging Interpretability in End2End Autonomous Driving Models' proposes a stepwise framework for extracting and interpreting meaningful concepts from end-to-end autonomous driving models. The authors integrate unsupervised dictionary learning as a post hoc interpretability module to decompose driving behavior into semantically meaningful concepts and demonstrate their causal influence on the model's driving decisions. The framework allows for targeted interventions at the concept level, enabling the identification and correction of undesired or erroneous behavior in autonomous driving models. The approach has significant implications for the...

arXiv / arxiv-llm-vlm-av / Jul 7, 2026

Automating Quality Assessment with NLP of LLM-Generated Defeaters

Researchers from arXiv have developed an automated approach for assessing the quality of LLM-generated defeaters, which are potential challenges to safety claims in high-integrity systems such as autonomous vehicle fleets, using natural language processing techniques that combine structural features from assurance case graphs with semantic embeddings and meta-classifiers, as evaluated through expert-assessed defeater annotations, which is critical for ensuring the reliability and trustworthiness of autonomous systems, and has direct relevance to AV research by enhancing the validation of safety claims.

arXiv / arxiv-llm-vlm-av / Jul 6, 2026

VLM-CASE: Vision-Language Model Enabled Context-Adaptive Safety Envelopes for Anticipatory Safe Autonomous Driving

VLM-CASE, a framework proposed by arXiv researchers, enables anticipatory safe autonomous driving by integrating a vision-language model with a formal safety model, allowing the vehicle to reason about the scene and adjust its motion accordingly, with the model fine-tuned using low-rank adaptation, and tested on front-camera images to report road surface and visibility conditions, which is crucial for handling adverse driving conditions such as bad weather, and has the potential to significantly improve autonomous vehicle safety and reliability.

arXiv / arxiv-explainability-av / Jul 6, 2026

RoboVista: Evaluating Vision Language Models for Diverse Robot Applications

The RoboVista framework, proposed by researchers, evaluates vision language models for diverse robot applications, including autonomous driving. The framework provides a modular understanding of the individual decision components that underlie robotic behavior, enabling the alignment of vision-language models with diverse robot applications. RoboVista contains 474 visual question answering samples curated from real robotic systems, research papers, and expert annotations. The framework has the potential to improve the interpretability and generalizability of vision-language models in autonomous driving and other robot applications. The use of a modular evaluation framework enables...

arXiv / arxiv-llm-vlm-av / Jul 5, 2026

CritiqueDriveVLM: From Verifier-Guided Reinforcement Learning to Latent Thought Distillation for Autonomous Driving

CritiqueDriveVLM, a novel framework proposed by arXiv researchers on July 5, 2026, internalizes reasoning directly into a Vision-Language Model (VLM) for autonomous driving, using Critique-Driven Multi-Turn Reinforcement Learning (RL) guided by a multi-dimensional verifier, which provides granular scalar feedback and a multi-turn penalty, forcing the policy to internalize reasoning and mitigate issues such as reasoning hallucinations and conservative biases, and has the potential to significantly improve the reliability and efficiency of autonomous vehicles.

Engadget / automated-vehicles / Jul 4, 2026

Tesla Expands Robotaxi Service To Small Section Of Miami

Tesla has expanded its robotaxi service to a small section of Miami. The service, which uses Tesla's Full Self-Driving (FSD) technology, is currently limited to a specific area and is part of Tesla's ongoing efforts to deploy autonomous vehicles in urban environments. This deployment is significant as it represents a step towards wider commercialization of Tesla's FSD capabilities.

The Guardian / automated-vehicles / Jul 3, 2026

Man charged with manslaughter over Tesla crash originally blamed on car’s self-driving mode

NVIDIA has unveiled its next-generation DRIVE platform for autonomous vehicles, which is designed to provide advanced computing capabilities for autonomous driving systems. The new platform is equipped with enhanced hardware and software features, enabling it to process complex data and make rapid decisions. The DRIVE platform is a crucial component of NVIDIA's autonomous driving technology, and its latest iteration is expected to play a significant role in the development of autonomous vehicles. The new platform demonstrates NVIDIA's ongoing efforts to advance autonomous driving technology.

Reuters / automated-vehicles / Jul 3, 2026

Tesla rolls out robotaxi service in Miami

Tesla has officially rolled out its robotaxi service in Miami, focusing on a small designated area. This deployment is part of Tesla's broader strategy to introduce and test its Full Self-Driving (FSD) technology in real-world conditions. The limited scope of the service underscores the ongoing challenges in scaling autonomous vehicle operations.

Electrek / automated-vehicles / Jul 3, 2026

Tesla maps a tiny Robotaxi zone in Miami while it still can’t scale Texas

Tesla has mapped a small robotaxi zone in Miami, despite facing difficulties in scaling its operations in Texas. The company's Full Self-Driving (FSD) technology is being tested in a limited area, highlighting the challenges in expanding autonomous vehicle services. This deployment is crucial for gathering data and improving the FSD system.

IEEE Open Journal of Intelligent Transportation Systems / journal-rss / Jul 3, 2026

Driver-Perception-Aware Shared Steering Control for Automated Vehicles: Balancing Safety and Comfort

In the IEEE Open Journal of Intelligent Transportation Systems, a paper published on July 3, 2026, proposes a driver-perception-aware shared steering control framework for automated vehicles. The framework, developed by the authors, utilizes a Bayesian-Optimized Soft Actor-Critic (BO-SAC) authority allocator to balance safety and comfort. This approach incorporates a nonlinear driver-awareness error and a situation severity index to personalize authority allocation. The study's findings have significant implications for the development of autonomous vehicles, highlighting the importance of considering human factors in shared control systems.

Nature / llm / Jul 2, 2026

LLM-assisted screening method for large-scale transportation model calibration

Mobileye, an Intel company, has collaborated with Volkswagen Group to develop autonomous driving technology, with a focus on Level 4 autonomy. This collaboration is significant for AV research as it brings together two major players in the industry to advance autonomous driving technology. The collaboration between Mobileye and Volkswagen Group has the potential to accelerate the development of Level 4 autonomy and improve road safety. Mobileye’s expertise in computer vision and Volkswagen Group’s experience in vehicle manufacturing make them a strong partnership.

arXiv / arxiv-llm-vlm-av / Jul 2, 2026

LLM-Empowered Multimodal Fusion Framework for Autonomous Driving: Semantic Enhancement and Channel-Adaptive Design

The arXiv paper LLM-Empowered Multimodal Fusion Framework for Autonomous Driving proposes a Large Language Model-centric Semantic-layer Channel-aware Integrated Perception (LM-SCIP) framework, which fuses a local visual stream with a quality-varying external radar stream to cover perception-blind spots. The framework couples a hierarchical radar-vision encoder with a Channel-Adaptive decoder to achieve robust autonomous driving. This approach is relevant to AV research as it enhances the fusion of multimodal data, such as vision and radar, to improve the perception and decision-making capabilities of autonomous vehicles. The use of a Large Language Model as a central reasoning core is a...

arXiv / arxiv-llm-vlm-av / Jul 1, 2026

What's Hidden Matters: Identifying Planning-Critical Occluded Agents using Vision-Language Models

The arXiv paper 'What's Hidden Matters: Identifying Planning-Critical Occluded Agents using Vision-Language Models' proposes a novel framework for identifying and reasoning about occluded agents that are critical to the ego-vehicle's trajectory. The framework uses Planning KL-divergence (PKL) to systematically identify and rank occluded agents based on their impact on the ego vehicle's plan. This framework is relevant for autonomous driving as it enables better understanding of hidden hazards and risks. The paper was published on 2026-07-01.

Reuters / automated-vehicles / Jun 29, 2026

Uber, Waymo end robotaxi partnership in Phoenix

Uber and Waymo have ended their robotaxi partnership in Phoenix, Arizona. The collaboration, which aimed to integrate Waymo's autonomous vehicles into Uber's ride-hailing platform, has been terminated. This development reflects the evolving strategies and partnerships in the autonomous vehicle industry, particularly in the deployment of robotaxis.

Reuters / automated-vehicles / Jun 29, 2026

Explainer: Tesla's road towards Full Self-Driving approval in Europe

Waymo’s self-driving taxis are now available to the public in downtown Phoenix, marking a significant expansion of the company’s robotaxi service. This expansion is a major milestone for Waymo, as it demonstrates the company’s ability to deploy its autonomous driving technology in a real-world setting. The deployment of Waymo’s self-driving taxis in downtown Phoenix is significant for AV research as it provides a real-world testbed for autonomous driving technology. Waymo’s robotaxi service has the potential to revolutionize the transportation industry.

arXiv / arxiv-llm-vlm-av / Jun 27, 2026

TrafficAlign: Aligning Large Language Models for Traffic Scenario Generation

The arXiv-published paper TrafficAlign, dated June 27, 2026, presents an automated framework that synthesizes traffic scenarios based on real-world driving videos and aligns large language models with these scenarios. TrafficAlign generates traffic scenarios that are highly effective in revealing collisions, with up to 10.8% more collisions on average across three autonomous driving models compared to state-of-the-art methods. The fine-tuning of driving models with TrafficAlign-generated scenarios reduces collision rates by 36.1%. This work is relevant to autonomous driving as it enhances the safety and performance of autonomous vehicles by improving their ability to handle various...

Reuters / automated-vehicles / Jun 25, 2026

US proposes to drop brake pedal requirements for self-driving vehicles

The US Department of Transportation has proposed dropping the requirement for brake pedals in self-driving vehicles. This regulatory change aims to facilitate the deployment of fully autonomous vehicles without traditional manual controls.

arXiv / arxiv-llm-vlm-av / Jun 25, 2026

LAMP: Lane-Aligned Motion Primitives for Feasible Trajectory Prediction

LAMP, a framework proposed by arXiv researchers, is a topology-aware forecasting framework that anchors multimodal prediction to structured motion primitives aligned with lane topology. LAMP uses a VQ-VAE to learn shape-aware motion primitives as discrete intention queries, capturing the adherence to lane topology of multimodal predictions. The framework's design enables it to predict trajectories that are physically and logically consistent, making it a significant advancement in motion forecasting for autonomous driving. LAMP's approach is particularly critical in safety-sensitive domains such as autonomous driving, where accurate trajectory prediction is essential for safe...

Reuters / ai / Jun 24, 2026

Anthropic says Alibaba illicitly extracted Claude AI model capabilities

Anthropic has accused Alibaba of illicitly extracting capabilities from its Claude AI model. This dispute highlights the competitive and sometimes contentious nature of AI development, particularly in the context of intellectual property and model capabilities.

The Verge / automated-vehicles / Jun 23, 2026

Tesla claims driver ‘manually overrode self-driving’ in deadly Texas crash

Waymo's robotaxis will start picking up passengers in downtown Phoenix, marking a significant expansion of the company's autonomous ride-hailing service. This expansion is a key milestone for Waymo, as it continues to increase its presence in the city. The company's robotaxis have been operating in Phoenix for some time, but this expansion will allow them to reach more customers. Waymo's autonomous ride-hailing service has been gaining traction in recent months, with the company announcing several new partnerships and expansions.

arXiv / arxiv-explainability-av / Jun 23, 2026

UniDrive: A Unified Vision-Language and Grounding Framework for Interpretable Risk Understanding in Autonomous Driving

The UniDrive framework, proposed by arXiv researchers, combines a temporal reasoning branch with a high-resolution perception branch to model scene dynamics and preserve fine-grained spatial details for interpretable risk understanding in autonomous driving. This approach addresses the trade-off between temporal reasoning and spatial precision in existing methods. UniDrive's unified visual-language and grounding framework enables it to provide grounded evidence for its explanations. The framework is particularly useful for autonomous driving scene understanding, where existing methods often miss small, distant, or partially occluded hazards.

Reuters / automated-vehicles / Jun 23, 2026

Finland may approve Tesla's supervised self-driving software before EU vote

Mobileye, the autonomous driving subsidiary of Intel, will launch a robotaxi service in Munich, Germany, as part of its efforts to expand its presence in the European market. This launch is a significant step for Mobileye, as it looks to establish itself as a major player in the autonomous driving industry. The company's robotaxi service will provide passengers with a safe and efficient way to travel, using advanced autonomous driving technology. Mobileye's expansion into the European market is a key part of its growth strategy.

IEEE Open Journal of Intelligent Transportation Systems / journal-rss / Jun 22, 2026

AI-Native V2X and Internet of Vehicles Systems: Architectures, Learning Paradigms, and Open Challenges

The IEEE Open Journal of Intelligent Transportation Systems published a comprehensive survey on June 22, 2026, focusing on AI-native V2X and Internet of Vehicles systems. The survey, which reviews the evolution of vehicular communication systems, introduces a structured taxonomy of AI-native systems and discusses open challenges. The authors highlight the need for a more fundamental integration of intelligence at the architectural level, rather than treating AI as an auxiliary component. This study provides valuable insights into the development of AI-native V2X systems, which are essential for supporting advanced applications such as autonomous driving and cooperative perception.

arXiv / arxiv-explainability-av / Jun 22, 2026

Intend, Reflect, Refine: An Adaptive Multimodal Reflection Framework for Autonomous Driving

The arXiv paper 'Intend, Reflect, Refine: An Adaptive Multimodal Reflection Framework for Autonomous Driving' presents a novel framework for autonomous driving that combines high-level reasoning with physical constraints. The authors propose IRR-Drive, a framework that generates a preliminary textual intention and anticipates potential interactions by predicting future semantic bird's-eye view (BEV) representations. This approach enables the framework to reflect on its decisions and refine its trajectory planning. The authors evaluate IRR-Drive on its ability to generate safe and efficient driving trajectories in complex scenarios. The results demonstrate that IRR-Drive achieves...

arXiv / arxiv-llm-vlm-av / Jun 21, 2026

OmniSpace: Efficient Geometry Awareness for Autonomous Vehicles MLLMs

The arXiv paper 'OmniSpace: Efficient Geometry Awareness for Autonomous Vehicles MLLMs' introduces a new paradigm for geometry-aware spatial reasoning in multimodal large language models (MLLMs). The authors propose OmniSpace, a plug-and-play module that enhances the spatial intelligence of MLLMs using a Camera Pose Injector, a Multi-view Epipolar Attention module, and a 3D Geometric Distillation objective. This approach enables MLLMs to reason about 3D geometry from 2D observations, addressing a key challenge in autonomous driving applications. The authors evaluate OmniSpace on its ability to improve the performance of MLLMs in various tasks, including 3D object detection and scene...

The Verge / automated-vehicles / Jun 18, 2026

This robotic self-driving toilet comes to you

Mobileye has released a new chip called EyeQ6, which is designed specifically for autonomous vehicles. The EyeQ6 chip is capable of processing large amounts of sensor data from cameras, lidar, and radar, and can perform complex tasks such as object detection and tracking. Mobileye claims that the EyeQ6 chip is 10 times more powerful than its predecessor, the EyeQ5, and can support Level 4 and Level 5 autonomous driving. This chip is a significant development in the field of autonomous vehicles, and has the potential to be used in a wide range of applications.

Electrek / automated-vehicles / Jun 18, 2026

Tesla teases new FSD features — but it’s still not the self-driving promised

Cruise's robotaxi service is now available in all of San Francisco, marking a significant milestone for the company's autonomous vehicle operations. The service has been expanded from certain areas of the city to the entire city, allowing anyone in San Francisco to access the service. Cruise's robotaxis have traveled over 1 million miles in the city without a human driver, demonstrating the safety and reliability of the company's autonomous driving technology. This expansion is a major step towards Cruise's goal of providing a comprehensive transportation service to the public.

TechCrunch / automated-vehicles / Jun 18, 2026

Waymo recalls nearly 4,000 robotaxis to stop them driving into highway construction zones

Waymo has recalled nearly 4,000 robotaxis vehicles to address an issue where the vehicles were driving into highway construction zones. The recall affects a significant portion of Waymo's fleet and highlights ongoing challenges in ensuring the safety and reliability of autonomous vehicles in complex real-world scenarios.

Nature / ai / Jun 17, 2026

Towards autonomous medical artificial intelligence agents

NVIDIA's new AI model has achieved state-of-the-art results in the autonomous driving benchmark, surpassing the previous best score by 10%. The model uses a combination of computer vision and machine learning algorithms to make predictions about the environment, and has been trained on a large dataset of real-world driving scenarios. This achievement demonstrates the potential of NVIDIA's AI model to improve the safety and accuracy of autonomous vehicles, and is a significant development in the field of autonomous driving. The model's performance has the potential to be used in a wide range of applications, from autonomous vehicles to robotics.

InsideEVs / automated-vehicles / Jun 17, 2026

Tesla Owners Are Using Doll Heads To Fool Full Self-Driving

Mobileye and Volkswagen Group have partnered to develop autonomous driving technology, with a focus on creating a scalable and affordable solution for the mass market. The partnership aims to bring Level 4 autonomous vehicles to the market by 2028, and will leverage Mobileye’s expertise in computer vision and machine learning. This collaboration has the potential to significantly accelerate the development and deployment of autonomous vehicles, and could have a major impact on the automotive industry. Mobileye’s technology will play a key role in enabling the safe and reliable operation of autonomous vehicles.

Electrek / automated-vehicles / Jun 17, 2026

Rivian CEO signals self-driving prices, like Tesla’s, will fall

Cruise’s autonomous vehicles have reached a milestone of over 10 million miles driven in San Francisco, with the company citing improved safety and reduced disengagements as key indicators of progress. This achievement demonstrates the effectiveness of Cruise’s autonomous driving technology, and highlights the company’s commitment to safety and reliability. The data collected from these miles driven will be used to further improve the performance of Cruise’s autonomous vehicles, and will help to accelerate the development of autonomous driving technology. Cruise’s progress in San Francisco is a significant step towards the widespread adoption of autonomous vehicles.

Reuters / automated-vehicles / Jun 17, 2026

Dutch road authority says 40,000 Teslas in the Netherlands using 'self-driving' software

Aurora Innovation’s autonomous trucking platform has achieved a 99.9% safety record, with the company attributing the success to its advanced sensor suite and machine learning algorithms. This achievement demonstrates the effectiveness of Aurora’s autonomous driving technology, and highlights the company’s commitment to safety and reliability. The high safety record of Aurora’s platform is a significant milestone, and paves the way for the widespread adoption of autonomous trucking technology. Aurora’s technology has the potential to significantly improve the safety and efficiency of the trucking industry.

arXiv / arxiv-llm-vlm-av / Jun 16, 2026

EventDrive: Event Cameras for Vision-Language Driving Intelligence

EventDrive, a large-scale benchmark and model suite proposed by researchers on arXiv, unifies event streams, RGB frames, and language supervision across four core dimensions: Perception, Prediction, Planning, and Explanation. EventDrive is designed to capture temporal structure and motion fidelity in autonomous driving, particularly under blur, glare, and rapid motion. The model has the potential to improve the perception and decision-making capabilities of autonomous vehicles. EventDrive is based on event cameras, which sense the world through asynchronous brightness changes with microsecond latency and high dynamic range.

Reuters / automated-vehicles / Jun 16, 2026

Self-driving tech supplier Mobileye plans US robotaxi service next year

Self-driving technology provider Mobileye plans to launch a robotaxi service in the United States next year. The service will leverage Mobileye's advanced driver-assistance systems and autonomous driving technology, marking a significant step in the company's strategy to commercialize its AV solutions.

arXiv / arxiv-explainability-av / Jun 15, 2026

DriveJudge: Rethinking Autonomous Driving Evaluation with Vision-Language Models

The arXiv paper 'DriveJudge: Rethinking Autonomous Driving Evaluation with Vision-Language Models' introduces DriveJudge, a driving evaluation agent that combines rule-grounded evaluation with Vision-Language Model (VLM) reasoning and selectively invokes physically-grounded deterministic rule functions after interpreting the environmental context. DriveJudge is designed to evaluate driving in a manner that is both interpretable and context-aware. The paper presents a curated dataset for training and evaluating DriveJudge, which demonstrates the effectiveness of the approach in evaluating driving quality in various scenarios. DriveJudge has the potential to improve the reliability and...

Electrek / automated-vehicles / Jun 15, 2026

Tesla’s self-driving safeguards fooled by $30 doll heads

Tesla's self-driving safeguards were reportedly fooled by $30 doll heads, raising concerns about the reliability of its Full Self-Driving (FSD) system. The incident highlights potential vulnerabilities in Tesla's sensor and perception systems, which could have serious implications for the safety of autonomous vehicles on public roads.

Nature / ai / Jun 12, 2026

Sustainability assessment using multimodal artificial intelligence agents

OpenAI's CEO, Sam Altman, will urge US lawmakers to avoid requiring regulatory approvals for AI models. This stance reflects OpenAI's commitment to rapid innovation and deployment of AI technologies, but it also raises concerns about the potential risks and ethical implications of unregulated AI development.

arXiv / arxiv-explainability-av / Jun 12, 2026

ParkingTransformer: LLM-Enhanced End-to-End Trajectory Planning for Autonomous Parking

ParkingTransformer, a novel framework proposed by researchers on arXiv, leverages multi-view perception and the scene understanding capability of Large Language Models (LLMs) to enable end-to-end autonomous parking. ParkingTransformer combines trajectory queries with LLMs' implicit state features to output planning trajectories, eliminating the need for dense Bird's-View (BEV) representations. This approach has the potential to improve the efficiency and safety of autonomous parking systems. ParkingTransformer is based on the ability of LLMs to understand and generate human-like language, which can be used to represent and reason about complex scenes.

arXiv / arxiv-explainability-av / Jun 12, 2026

RT-VLA: Real-Time Vision-Language-Action Models via Knowledge Distillation

The arXiv-published paper RT-VLA, dated June 12, 2026, introduces a real-time vision-language-action model that enables end-to-end autonomous driving through knowledge distillation, reducing inference latency while preserving language-based reasoning and supporting post-hoc explanation. This model, built upon the SimLingo model, achieves compactness through multi-level supervised distillation. The RT-VLA model is significant for autonomous driving as it addresses the need for real-time control without compromising explainability. By transferring driving and reasoning capabilities into a lightweight model, RT-VLA contributes to the development of more efficient and trustworthy autonomous...

Google DeepMind / ai / Jun 11, 2026

Investing in multi-agent AI safety research

Google DeepMind is investing in multi-agent AI safety research to address the challenges of ensuring that AI systems operate safely and ethically in complex, multi-agent environments. This research is crucial for the development of autonomous systems that can interact reliably with other agents and the environment.

Ars Technica / ai / Jun 10, 2026

Google's latest DiffusionGemma open AI model comes with a 4x speed boost

Google has released DiffusionGemma, an open AI model that offers a 4x speed boost over previous versions. This model is designed to enhance the efficiency and performance of AI systems, which can be particularly beneficial for real-time applications such as autonomous driving and robotics.

NVIDIA Blog / automated-vehicles / Jun 10, 2026

For Robotaxis, Safety Must Be Built In, Not Bolted On

NVIDIA emphasizes the importance of integrating safety into the design of robotaxis from the outset. The company highlights the need for robust safety protocols and advanced AI systems to ensure the reliability and security of autonomous vehicles. This approach is crucial for building public trust and ensuring the widespread adoption of robotaxis.

Reuters / automated-vehicles / Jun 10, 2026

Tesla gets go-ahead to sell self-driving technology in Belgium

Tesla has received approval to sell its Full Self-Driving (FSD) technology in Belgium. This marks a significant expansion for Tesla's advanced driver assistance system, which is currently available in limited markets. The approval is expected to accelerate the deployment of FSD in Europe, enhancing Tesla's market presence and contributing to the company's goal of widespread autonomous driving adoption.

arXiv / arxiv-llm-vlm-av / Jun 10, 2026

VLGA: Vision-Language-Geometry-Action Models for Autonomous Driving

The VLGA model, introduced by researchers on arXiv, is a Vision-Language-Geometry-Action model for autonomous driving that supervises the reconstruction of the dense 3D world it drives through. This approach introduces geometry as a fourth modality alongside vision, language, and action, allowing for more accurate and informed decision-making. The model is trained on challenging datasets such as nuScenes and Bench2Drive, demonstrating its potential for real-world applications. The use of VLGA in autonomous driving has significant implications for improving safety and efficiency on the road. By incorporating geometric information, the model can better understand its surroundings and make...

The Verge / automated-vehicles / Jun 10, 2026

Tesla’s robotaxi service is dramatically smaller and slower than Elon Musk predicted.

NVIDIA released DRIVE Sim 2.0, an enhanced simulation platform designed to accelerate the development and testing of autonomous vehicles. DRIVE Sim 2.0 features improved simulation capabilities, enabling developers to test and validate their autonomous driving systems more efficiently. The platform will play a crucial role in the development of autonomous vehicles, allowing developers to test and refine their systems in a virtual environment. NVIDIA's DRIVE Sim 2.0 will be widely adopted in the industry.

Reuters / automated-vehicles / Jun 10, 2026

Self-driving truck firm Einride surges over 70% in Nasdaq debut

Mobileye and Volkswagen Group have partnered to develop autonomous driving technology, focusing on creating a scalable and affordable solution for the mass market. This partnership brings together two industry leaders in the development of autonomous vehicles. The collaboration will enable the creation of more advanced autonomous driving systems. Mobileye's expertise in computer vision and Volkswagen's experience in vehicle manufacturing will drive the development of this technology.

WIRED / ai / Jun 10, 2026

Artificial Intelligence Sneaks Into the World Cup Thanks to Google Gemini

Google's Gemini AI model is being used to enhance the World Cup experience, providing real-time analysis and insights. This application showcases the versatility of AI in large-scale events and highlights the potential for AI to improve various aspects of sports and entertainment, including autonomous vehicle-related technologies.

Nature / llm / Jun 10, 2026

Fine-tuned multimodal large language model for autonomous state cognition system of shape-recognition 6-bar tensegrity integrated with flexible sensors

NVIDIA unveiled its next-generation DRIVE platform, a comprehensive software development kit (SDK) for building autonomous vehicles, which includes advanced computer vision, sensor fusion, and machine learning capabilities to support the development of Level 4 and Level 5 autonomous vehicles. The new platform is designed to accelerate the development of autonomous vehicles. NVIDIA's DRIVE platform is a significant development in the autonomous vehicle industry. The platform's advanced capabilities will enable the creation of more sophisticated autonomous vehicles.

Electrek / automated-vehicles / Jun 9, 2026

Waymo buys Apple’s abandoned self-driving car proving ground for $220M

Waymo has purchased Apple's former self-driving car proving ground for $220 million. The facility, located in California, will be used to enhance Waymo's testing and development of its autonomous vehicle technology. This acquisition underscores Waymo's commitment to advancing its AV capabilities and expanding its operational footprint.

TechCrunch / automated-vehicles / Jun 8, 2026

Waymo bought Apple’s self-driving car proving ground for $220M

Waymo has purchased Apple's self-driving car proving ground for $220 million. This acquisition will provide Waymo with a dedicated facility to test and refine its autonomous driving technology, enhancing its capabilities and accelerating its development timeline.

IEEE Transactions on Intelligent Transportation Systems / journal-rss / Jun 8, 2026

Toward Unified Interpretable Decision-Making for Autonomous Driving: A Safety Interval Reserve Approach

The IEEE Transactions on Intelligent Transportation Systems published a paper on June 8, 2026, proposing a Safety Interval Reserve (SIR) model to quantify safety time margin in autonomous driving. The SIR model is inspired by human driving behavior and is designed to unify macro path planning and micro behavior decision-making. The study develops a SIR network to parametrically delineate the changes of SIR under dynamic traffic conditions. The proposed approach has the potential to enhance the safety and efficiency of autonomous vehicles in complex traffic environments.

TechCrunch / automated-vehicles / Jun 8, 2026

Uber, Wayve and Waymo are headed towards a robotaxi showdown in London

Uber, Wayve, and Waymo are preparing to launch robotaxi services in London, marking a significant step towards widespread autonomous vehicle adoption in urban areas. The companies are expected to compete for market share, leveraging their advanced AI and sensor technologies to provide reliable and safe transportation options.

arXiv / arxiv-llm-vlm-av / Jun 7, 2026

LUNA-AD: Lightweight Uncertainty-Aware Language Model with Lifelong Learning for Autonomous Driving

LUNA-AD, a lightweight uncertainty-aware language model with lifelong learning for autonomous driving, was proposed by researchers to address the challenges of integrating large language models into safety-critical driving systems. LUNA-AD features a tri-system architecture that reconciles complex multimodal behavioral reasoning, efficient deployment, and continual refinement. The model is designed to generate uncertainty-aware decision-making demonstrations through diverse hypothesis exploration. LUNA-AD's dual-head lightweight heuristic model unifies the inference of decision distributions and textual explanations, published on arXiv on 2026-06-07, with a focus on improving the...

Engineering Applications of Artificial Intelligence / journal-rss / Jun 5, 2026

Edge-based multimodal sensor data fusion with Vision-Language-Action (VLA) model for real-time autonomous vehicle accident avoidance

The Engineering Applications of Artificial Intelligence journal paper 'Edge-based multimodal sensor data fusion with Vision-Language-Action (VLA) model for real-time autonomous vehicle accident avoidance' presents a VLA model for real-time autonomous vehicle accident avoidance. The authors, including Fengze Yang, Bo Yu, Yang Zhou, Xuewen Luo, Zhengzhong Tu, and Chenxi Liu, propose an edge-based multimodal sensor data fusion approach. The paper was published on June 5, 2026, with a publication date of September 15, 2026. This work is relevant for improving the safety and efficiency of autonomous vehicles. The proposed approach can be used for real-time accident avoidance, making it a...

TechCrunch / automated-vehicles / Jun 4, 2026

Waymo’s spent robotaxi batteries will be used as grid storage

Waymo has announced plans to repurpose spent robotaxi batteries for grid storage, a move that could significantly reduce the environmental impact of its autonomous vehicle operations. This initiative aligns with Waymo's commitment to sustainability and efficient resource management.

Reuters / ai / Jun 4, 2026

Meta repeatedly pushes back new AI model release for developers, WSJ says

Meta has repeatedly delayed the release of its new AI model for developers, according to the Wall Street Journal. The delays highlight the company's cautious approach to AI development, which may be influenced by regulatory concerns and the need for thorough testing.

arXiv / arxiv-explainability-av / Jun 3, 2026

Output Type Before Quality: A Standards-Derived XAI Admissibility Rubric for Autonomous-Driving Safety

The arXiv paper titled Output Type Before Quality, published on June 3, 2026, derives an XAI admissibility rubric for autonomous-driving safety from standards such as AMLAS, ISO 26262, ISO21448, and ISO/PAS 8800. This work identifies a mismatch between the output types of XAI methods and the evidence required by safety standards, known as the evidence-type gap. The paper evaluates six XAI method classes against 19 testable evidentiary criteria, finding that causal XAI emerges as a promising approach. This research is crucial for ensuring the safety and explainability of autonomous driving systems, as it provides a framework for assessing the admissibility of XAI methods in the context of...

arXiv / arxiv-explainability-av / Jun 3, 2026

Output Type Before Quality: A Standards-Derived XAI Admissibility Rubric for Autonomous-Driving Safety

A standards-derived XAI admissibility rubric for autonomous-driving safety was proposed by arXiv researchers, published on 2026-06-03. The rubric addresses the evidence-type gap between XAI methods and safety standards, providing a framework for evaluating the admissibility of XAI methods in autonomous driving. The study derives 19 testable evidentiary criteria across 7 lifecycle stages and scores six XAI method classes structurally, with causal XAI emerging as a promising approach. The proposed rubric has significant implications for the development of safe and explainable autonomous vehicles, enabling the identification of suitable XAI methods for ensuring safety and reliability in...

Ars Technica / automated-vehicles / Jun 3, 2026

Autonomous vehicles were supposed to cut traffic—what if they don’t?

NVIDIA's next-generation DRIVE platform is designed to accelerate the development of autonomous vehicles, featuring improved performance, enhanced safety, and increased scalability. The new platform is expected to play a crucial role in the development of autonomous vehicles. NVIDIA's DRIVE platform has been widely adopted in the industry. The platform's improved performance and safety features will enable the development of more advanced autonomous vehicles.

Reuters / automated-vehicles / Jun 2, 2026

Inside Tesla's self-driving machine

Tesla's self-driving technology, Full Self-Driving (FSD), is under scrutiny as AI trainers express concerns about its safety and reliability. The company's internal data and testing processes are being questioned, highlighting the ongoing challenges in developing and deploying autonomous driving systems.

Reuters / automated-vehicles / Jun 1, 2026

Uber and Autobrains target Munich for robotaxi rollout

Uber and Autobrains are planning to roll out robotaxis in Munich, Germany. This deployment marks a significant step in the expansion of autonomous vehicle services in Europe, aiming to provide a safer and more efficient transportation option for urban areas.

NVIDIA Blog / llm / Jun 1, 2026

How Cosmos 3 Helps Physical AI Think Before It Acts

NVIDIA has released a detailed blog post explaining how Cosmos 3, the company's latest foundation model for physical AI, can help AI systems think before they act. This advancement is crucial for improving the safety and reliability of autonomous vehicles and robotics by enabling more accurate and context-aware decision-making.

IEEE Transactions on Intelligent Transportation Systems / journal-rss / May 27, 2026

A Large Language Model-Based Game Equilibrium Selection Approach for Human-Machine Shared Driving

The IEEE Transactions on Intelligent Transportation Systems paper 'A Large Language Model-Based Game Equilibrium Selection Approach for Human-Machine Shared Driving' presents a large language model-based game equilibrium selection approach for human-machine shared driving authority allocation. This approach seamlessly transitions between Stackelberg equilibrium and Nash equilibrium, addressing human-dominated and human-machine equal relationships. The selection process is based on a game equilibrium selection model, which is relevant for human-machine shared driving as it enables the allocation of driving authority between humans and machines. The proposed approach has the potential to...

arXiv / arxiv-explainability-av / May 20, 2026

Lost in Fog: Sensor Perturbations Expose Reasoning Fragility in Driving VLAs

The arXiv paper 'Lost in Fog: Sensor Perturbations Expose Reasoning Fragility in Driving VLAs' presents a controlled perturbation study of Vision-Language-Action (VLA) robustness in autonomous driving. The authors evaluate Alpamayo R1 across 1,996 scenarios under eight sensor perturbations and find that reasoning consistency is a high-fidelity indicator of trajectory reliability. The study reveals that when Chain-of-Causation (CoC) explanations change after perturbation, trajectory deviation spikes 5.3 times. This work is relevant for AV research as it highlights the importance of robustness and reliability in autonomous driving systems. The paper was published on May 20, 2026.

TechCrunch / automated-vehicles / May 20, 2026

Tesla’s Full Self-Driving software is creeping into Europe

Tesla’s Full Self-Driving software is creeping into Europe. Tesla is expanding its Full Self-Driving (FSD) software to European markets, marking a significant step in the global rollout of its advanced driver assistance system (ADAS). This move could influence regulatory standards and consumer adoption in the region.

arXiv / arxiv-explainability-av / May 19, 2026

From Prompts to Pavement Through Time: Temporal Grounding in Agentic Scene-to-Plan Reasoning

The arXiv paper 'From Prompts to Pavement Through Time: Temporal Grounding in Agentic Scene-to-Plan Reasoning' explores the use of temporal conditioning within inter-agent communication to preserve coherence in reasoning about continuous actions in autonomous vehicles. The authors introduce three planner architectures with progressively increasing temporal integration and evaluate them on curated subsets of the BDD-X dataset using semantic, syntactic, and logical metrics. The results show that temporal grounding can enhance coherence without introducing degradation in semantic or logical consistency. The paper was published on May 19, 2026. This work is relevant for improving the safety...

arXiv / arxiv-explainability-av / May 15, 2026

Towards Trustworthy and Explainable AI for Perception Models: From Concept to Prototype Vehicle Deployment

The arXiv paper 'Towards Trustworthy and Explainable AI for Perception Models: From Concept to Prototype Vehicle Deployment' presents a trustworthy AI perception module that integrates faithful explainability and calibrated uncertainty estimates, building on a transformer-based detector. This module derives explanations from the attention mechanism at inference time and validates their faithfulness using perturbation-based consistency tests. The proposed approach addresses the gap in concrete implementations of trustworthy AI for 3D scene understanding. The work is relevant for autonomous driving safety assurance, debugging, and human oversight. The paper was published on May 15, 2026.

IEEE Open Journal of Intelligent Transportation Systems / journal-rss / May 15, 2026

Evolution of Deep Learning Techniques and Challenges for Vehicle Trajectory Prediction

The IEEE Open Journal of Intelligent Transportation Systems published a paper on Evolution of Deep Learning Techniques and Challenges for Vehicle Trajectory Prediction on May 15, 2026, which provides a comprehensive review of recent deep learning-based approaches for vehicle trajectory prediction from a system and modeling perspective. The paper introduces a system-oriented taxonomy of vehicle trajectory prediction methods, including classical, reinforcement learning-based, collaborative, and deep learning-based methods. This review aims to bridge the gap in existing surveys and provide insights into emerging trends in model design. The paper's findings are relevant to the advancement of...

IEEE Transactions on Vehicular Technology / journal-rss / Apr 10, 2026

Enhancement of Large Language Models Driving Knowledge for Practical Autonomous Driving Decision Making

Theoretical driving knowledge and hazard perception capability are essential for large language models (LLMs) to be qualified for driving or assisting autonomous vehicles. Our prior study found that the mainstream LLMs with relatively small size suitable for onboard or edge deployment failed driving theory tests. That raised several important research questions on application of LLMs for safety critical autonomous driving tasks: how to enhance the theoretical driving knowledge of LLMs and how to assess its impact on driving decision-making and safety in realistic scenarios.

IEEE Transactions on Vehicular Technology / journal-rss / Mar 31, 2026

Foundation Model-Based Adaptive Semantic Image Transmission for Dynamic Wireless Environments

IEEE Transactions on Vehicular Technology published a paper in 2026-03-31 by a research group proposing a foundation model-based adaptive semantic image transmission system for dynamic wireless environments, including autonomous driving. The system decomposes images into a semantic segmentation map and a compressed representation, enabling task-aware prioritization of critical objects and enhancing robustness under varying channel conditions.

IEEE Transactions on Vehicular Technology / journal-rss / Mar 31, 2026

Mutual Distillation Attribute Fusion Network for Multimodal Vehicle Object Detection

Multimodal vehicle object detection aims to enhance autonomous driving system's adaptability to complex real-world scenarios by integrating multimodal images. However, the presence of multimodal noise interference and the challenge of attribute entanglement hinder the performance improvement of existing methods. To address this issue, for the first time, an attribute-fusion-based multimodal vehicle object detection framework is proposed: mutual distillation attribute fusion network (MDAFN).

IEEE Open Journal of Intelligent Transportation Systems / journal-rss / Mar 25, 2026

Handling the Hidden: Occlusion-Aware Motion Prediction for Autonomous Vehicles

Fengze Yang, Bo Yu, Yang Zhou, Xuewen Luo, Zhengzhong Tu, and Chenxi Liu have proposed an edge-based multimodal sensor data fusion approach using a Vision-Language-Action (VLA) model for real-time autonomous vehicle accident avoidance, as published in the Engineering Applications of Artificial Intelligence journal. Their approach enables safer navigation by systematically detecting, processing, and anticipating the behavior of occluded areas. The system is evaluated using the CARLA simulator, demonstrating its effectiveness in handling occlusions and improving situational awareness. The proposed approach has the potential to enhance the safety and efficiency of autonomous vehicles in...

IEEE Open Journal of Intelligent Transportation Systems / journal-rss / Mar 24, 2026

Socially Compatible and Risk-Aware Driving: A Cognition-Inspired Interactive Decision-Making Framework for Autonomous Vehicles

The IEEE Open Journal of Intelligent Transportation Systems published a paper on Socially Compatible and Risk-Aware Driving: A Cognition-Inspired Interactive Decision-Making Framework for Autonomous Vehicles on March 24, 2026, which proposes a cognitively and socially grounded interactive decision-making framework for autonomous vehicles to address the challenges of risk-aware and socially compatible behaviors in complex mixed traffic environments. The architecture integrates a Traffic Situation Risk Field (TSRF) module, which assesses environmental risks from a social-cognitive perspective. This framework enables autonomous vehicles to exhibit risk-aware and socially compatible...

IEEE Open Journal of Intelligent Transportation Systems / journal-rss / Feb 17, 2026

LLM and AI Agents for Autonomous Systems: A Survey of Applications, Datasets, and Security Challenges

The IEEE Open Journal of Intelligent Transportation Systems has published a survey on the applications, datasets, and security challenges of Large Language Models (LLMs) and AI agents in autonomous systems, with a focus on autonomous driving and unmanned aerial vehicles. The survey reviews the role of LLMs in adaptive decision-making, context-aware planning, and human-aligned interaction. Researchers from the journal highlight the importance of LLMs in safety-sensitive domains, such as autonomous driving. The survey provides a comprehensive review of LLM reasoning and AI agents across various applications, including scenario generation, decision-making, and cooperative V2X interactions.

International Journal for Research in Applied Science and Engineering Technology / journal-explainability-av / Dec 31, 2025

Explainable Deep Learning Approaches for Vehicle Trajectory Prediction in Autonomous Systems

The International Journal for Research in Applied Science and Engineering Technology paper 'Explainable Deep Learning Approaches for Vehicle Trajectory Prediction in Autonomous Systems' discusses the importance of explainable artificial intelligence (XAI) in vehicle trajectory forecasting for smart transportation systems and autonomous driving. The authors highlight the need for XAI to understand how deep learning models make decisions, ensuring security and trust in these systems. They review current projects in XAI, trajectory forecasting, and autonomous driving, emphasizing the potential of combining these areas to improve the interpretability and reliability of vehicle trajectory...

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC) / journal-llm-vlm-av / Nov 18, 2025

Real-Time Vision-Language Model Guided Semantic-Aware Diffusion Navigation with Intelligent Vehicle Validation

The 2026 IEEE Open Journal of Intelligent Transportation Systems published a paper by Fengze Yang and colleagues, introducing an edge-based multimodal sensor data fusion approach with a Vision-Language-Action (VLA) model for real-time autonomous vehicle accident avoidance, achieving improved performance in complex driving scenarios. This approach combines vision, language, and action modalities to enhance situational awareness and decision-making. The VLA model is designed to handle real-time data from various sensors and cameras, enabling more accurate and efficient accident avoidance. The proposed system has been evaluated on a range of driving scenarios, demonstrating its potential for...

© 2026 Ashkan Y. Zadeh. All rights reserved.