# AI agents in autonomous vehicles

AI agents in autonomous vehicles are intelligent software systems that use sensor data and algorithms to interpret driving environments, make decisions, and control functions such as steering, braking, and obstacle detection.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## CP-Guard+: A New Paradigm for Malicious Agent Detection and Defense in Collaborative Perception

DevFeed: [CP-Guard+: A New Paradigm for Malicious Agent Detection and Defense in Collaborative Perception](<https://devfeed.tech/articles/cp-guard-a-new-paradigm-for-malicious-agent-detection-and-defense-in-collaborative-perception-144851.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2502.07807>)

Author: Senkang Hu, Yihang Tao, Zihan Fang, Guowen Xu, Hangcheng Cao, Yiqin Deng, Sam Kwong, Yuguang Fang

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [defense](<https://devfeed.tech/tags/defense.md>), [detection](<https://devfeed.tech/tags/detection.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [feature](<https://devfeed.tech/tags/feature.md>), [features](<https://devfeed.tech/tags/features.md>), [malicious](<https://devfeed.tech/tags/malicious.md>), [margin](<https://devfeed.tech/tags/margin.md>), [overhead](<https://devfeed.tech/tags/overhead.md>), [performance](<https://devfeed.tech/tags/performance.md>), [results](<https://devfeed.tech/tags/results.md>), [robust](<https://devfeed.tech/tags/robust.md>), [safe](<https://devfeed.tech/tags/safe.md>), [sensing](<https://devfeed.tech/tags/sensing.md>), [share](<https://devfeed.tech/tags/share.md>), [system](<https://devfeed.tech/tags/system.md>), [systems](<https://devfeed.tech/tags/systems.md>)

### AI overview

CP-Guard+ targets malicious agents in collaborative perception systems used by connected vehicles, where shared sensing can expose an ego vehicle to misleading information. Its detection paradigm identifies malicious agents from features without checking final perception results, which the authors say reduces computational overhead. The work also introduces CP-GuardBench for training and evaluating detection methods, and a defense method that uses Dual-Centered Contrastive Loss (DCCLoss) to increase the separation between benign and malicious feature representations. The authors report experiments on CP-GuardBench and V2X-Sim and say CP-Guard+ outperforms alternatives; the excerpt provides no quantitative results or further experimental details.

### Source excerpt

arXiv:2502.07807v2 Announce Type: replace-cross Abstract: Collaborative perception (CP) is a promising method for safe connected and autonomous driving, which enables multiple vehicles to share sensing information to enhance perception performance. However, compared with single-vehicle perception, the openness of a CP system makes it more vulnerable to malicious attacks that can inject malicious information to mislead the perception of an ego vehicle, resulting in severe risks for safe driving. To mitigate such vulnerability, we first propose a new paradigm for malicious agent detection that effectively identifies malicious agents at the feature level without requiring verification of final perception results, significantly reducing computational overhead. Building on this paradigm, we introduce CP-GuardBench, the first comprehensive dataset provided to train and evaluate various malicious agent detection methods for CP systems. Furthermore, we develop a robust defense method called CP-Guard+, which enhances the margin between the representations of benign and malicious features through a carefully designed Dual-Centered Contrastive Loss (DCCLoss). Finally, we conduct extensive experiments on both CP-GuardBench and V2X-Sim, and demonstrate the superiority of CP-Guard+.

## From Self-Play RL to Constrained Control: Distilling DiffMPC Students with DAgger

DevFeed: [From Self-Play RL to Constrained Control: Distilling DiffMPC Students with DAgger](<https://devfeed.tech/articles/from-self-play-rl-to-constrained-control-distilling-diffmpc-students-with-dagger-145022.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25197>)

Author: Luca M. Hartmann, Parinaz Naghizadeh

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Robotics](<https://devfeed.tech/topics/robotics.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [grpo](<https://devfeed.tech/topics/grpo.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [beats](<https://devfeed.tech/tags/beats.md>), [boundary](<https://devfeed.tech/tags/boundary.md>), [control](<https://devfeed.tech/tags/control.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [cs-sy](<https://devfeed.tech/tags/cs-sy.md>), [dagger](<https://devfeed.tech/tags/dagger.md>), [distilling](<https://devfeed.tech/tags/distilling.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [driving](<https://devfeed.tech/tags/driving.md>), [eess-sy](<https://devfeed.tech/tags/eess-sy.md>), [failures](<https://devfeed.tech/tags/failures.md>), [ppo](<https://devfeed.tech/tags/ppo.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [rl](<https://devfeed.tech/tags/rl.md>), [simulator](<https://devfeed.tech/tags/simulator.md>), [students](<https://devfeed.tech/tags/students.md>), [training](<https://devfeed.tech/tags/training.md>), [waymo](<https://devfeed.tech/tags/waymo.md>)

### AI overview

The approach adds explicit constraints to a trained driving policy by keeping a PPO teacher's backbone, replacing its action head with DiffMPC to enforce road-boundary half-planes and agent ellipses, then distilling the student with DAgger in the teacher's self-play simulator. Across nine seeds, the constrained student of a smaller teacher matched it on nuPlan and outperformed it on CARLA and out-of-distribution Waymo scenes. For a stronger teacher, the student had fewer failures on nuPlan and Waymo and similar results on CARLA. Distillation used no more than 30% of the teacher's training time.

### Source excerpt

arXiv:2611.25197v1 Announce Type: cross Abstract: Self-play reinforcement learning (RL) produces strong driving policies, but these policies are hard to verify or interpret. Differentiable model predictive control (DiffMPC) can enforce explicit constraints, but it is usually trained by imitating logged data or with costly RL. To add constraints to an already trained policy, we keep a PPO teacher's backbone and replace its action head with a DiffMPC that enforces road boundary half-planes and agent ellipses. We then distill the student with DAgger in the teacher's self-play simulator. Over 9 seeds, the constrained student of a smaller teacher matches its teacher on nuPlan and beats it on CARLA and on out-of-distribution Waymo scenes. For a stronger teacher, the student has fewer failures on nuPlan and Waymo, while staying level on CARLA. Its distillation takes no more than 30% of the teacher's training time.

## CaFe: Causally Faithful Conditional Traffic Scenario Generation

DevFeed: [CaFe: Causally Faithful Conditional Traffic Scenario Generation](<https://devfeed.tech/articles/cafe-causally-faithful-conditional-traffic-scenario-generation-141499.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22855>)

Author: Steven Yu Tung, Shubham Vaijanath Phoolari, Christoph Lauer, Abhinav Valada, Steven Peters

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [behaviour](<https://devfeed.tech/tags/behaviour.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cafe](<https://devfeed.tech/tags/cafe.md>), [conditional](<https://devfeed.tech/tags/conditional.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [events](<https://devfeed.tech/tags/events.md>), [example](<https://devfeed.tech/tags/example.md>), [existing](<https://devfeed.tech/tags/existing.md>), [failure](<https://devfeed.tech/tags/failure.md>), [generation](<https://devfeed.tech/tags/generation.md>), [ghost](<https://devfeed.tech/tags/ghost.md>), [grounded](<https://devfeed.tech/tags/grounded.md>), [traffic](<https://devfeed.tech/tags/traffic.md>)

### AI overview

CaFe targets intent leakage in language-conditioned traffic simulation: surrounding agents may react to a controlled vehicle's private instruction before its behavior is observable, changing the scenario. Its paired intervention test holds the controlled vehicle's executed trajectory fixed while changing its instruction to isolate these effects. The authors also present WOMD-PACT, a verifiable dataset of 531,099 WOMD scenes with timed behavior tags, reaction episodes identified by a counterfactual ghost test, and grounded private instructions. CaFe uses a dual-stream diffusion-forcing simulator, fine-tuned in closed loop, that limits inter-agent interactions to public beliefs. In the reported evaluation, existing conditioning designs displaced 23-83% of nearby uninstructed agents through leakage; CaFe eliminated measurable leakage while achieving a WOSAC realism score of 0.763, described as comparable.

### Source excerpt

arXiv:2611.22855v1 Announce Type: new Abstract: Language-conditioned traffic simulation enables scalable test scenarios steered by natural language. The goal is to control selected agents while surrounding traffic reacts to their observed behaviour. For example, a vehicle instructed to cut in late should cause the following vehicle to brake only when the manoeuvre becomes observable. Yet simulators can leak private instructions across agents, causing surrounding vehicles to react before the instructed behaviour becomes observable. Such premature reactions defuse desired events and alter the ensuing closed-loop scenario. Existing benchmarks measure what agents do, not what they could have known, leaving this failure unmeasured. We formalise it as intent leakage and address it through evaluation, data and modelling. Our paired intervention test holds an instructed agent's executed trajectory fixed while varying its private instruction, isolating leakage-induced changes in surrounding agents. We also construct WOMD-PACT, an automatically generated, verifiable dataset of 531,099 WOMD scenes with timed behaviour tags with reaction episodes identified through a counterfactual ghost test and grounded private instructions. Finally, we introduce CaFe, a dual-stream diffusion-forcing simulator fine-tuned in closed loop that restricts inter-agent interactions to public beliefs, preserving legitimate anticipation without leaking private intent. Our evaluation reveals that existing conditioning designs displace 23-83% of nearby uninstructed agents through private-intent leakage, even when the instructed agent's executed motion is held fixed. CaFe eliminates measurable leakage while maintaining comparable WOSAC realism (0.763), demonstrating that causal faithfulness can be enforced without sacrificing measured simulation realism.

## Test-Time Coverage: Test-Conditioned Data Curation for Deployment-Aware Learning

DevFeed: [Test-Time Coverage: Test-Conditioned Data Curation for Deployment-Aware Learning](<https://devfeed.tech/articles/test-time-coverage-test-conditioned-data-curation-for-deployment-aware-learning-139794.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.22697>)

Author: Nadine Chang, Maying Shen, Shizhe Diao, Jialiang Wang, Jingde Chen, Thomas Breuel, Pavlo Molchanov, Rafid Mahmood, Jose M. Alvarez

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Advanced AI workflows](<https://devfeed.tech/topics/advanced-ai-workflows.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [coverage](<https://devfeed.tech/tags/coverage.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [curation](<https://devfeed.tech/tags/curation.md>), [data](<https://devfeed.tech/tags/data.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [inference](<https://devfeed.tech/tags/inference.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [learning](<https://devfeed.tech/tags/learning.md>), [test](<https://devfeed.tech/tags/test.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

TTCov uses unlabeled deployment samples to guide training-data selection before training, without adapting model weights at inference. It builds a task Atlas of deployment-relevant atomic propositions, expands it with concepts found in deployment samples, and uses their observed frequencies to form a Knowledge Atlas target. It then selects a budgeted training set to approximate that target. In autonomous driving, the authors report greater deployment-relevant coverage, closer distribution matching, and stronger end-to-end driving performance than data-curation baselines, with city-to-city expansion for novel domains. The supplied abstract does not provide quantitative results or methodological details.

### Source excerpt

arXiv:2607.22697v2 Announce Type: replace Abstract: Deployed AI systems are often trained from broad candidate data pools, necessitating data curation towards the deployment test distribution. However, standard data curation methods score training-side criteria rather than directly optimizing deployment match. We introduce TTCov (Test-Time Coverage), a data-level test-conditioned curation method that uses test-side information before training instead of updating model weights at inference. TTCov decomposes deployment-conditioned curation into coverage and distribution. To represent coverage, it builds a task Atlas, a collection of LLM-based atomic propositions (APs) describing deployment-relevant concepts, seeded from open task knowledge and expanded with unmatched APs extracted from unlabeled deployment samples. To represent distribution, it instantiates the matched deployment APs with their frequencies, yielding a Knowledge Atlas (K-Atlas) that operationalizes the deployment distribution as a curation target. TTCov then selects a budgeted training set whose deployment APs distribution approximates this target. We apply TTCov towards autonomous driving (AD), keeping adaptation off the inference path while selecting data with greater deployment-relevant coverage, closer K-Atlas matching, and stronger downstream end-to-end driving performance than data-curation baselines, including seamless adaptability to novel domains via city-to-city expansion. Code available at https://github.com/NVlabs/TTCov.

## SAFER: Navigating the Exploration-Exploitation Dilemma in World-Model-Based Planning for Autonomous Driving

DevFeed: [SAFER: Navigating the Exploration-Exploitation Dilemma in World-Model-Based Planning for Autonomous Driving](<https://devfeed.tech/articles/safer-navigating-the-exploration-exploitation-dilemma-in-world-model-based-planning-for-autonomous-driving-137535.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20107>)

Author: Yansong Qu, Zilin Huang, Zihao Sheng, Sikai Chen, Samuel Labi

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [World models](<https://devfeed.tech/topics/world-models.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [autonomous](<https://devfeed.tech/tags/autonomous.md>), [based](<https://devfeed.tech/tags/based.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [dilemma](<https://devfeed.tech/tags/dilemma.md>), [driving](<https://devfeed.tech/tags/driving.md>), [exploitation](<https://devfeed.tech/tags/exploitation.md>), [exploration](<https://devfeed.tech/tags/exploration.md>), [model](<https://devfeed.tech/tags/model.md>), [planning](<https://devfeed.tech/tags/planning.md>), [world](<https://devfeed.tech/tags/world.md>), [world-model](<https://devfeed.tech/tags/world-model.md>)

### AI overview

SAFER combines offline learning from expert-judged failures in simulation with online model refinement as the vehicle encounters situations beyond the offline data. Partially trained agents and scripted recoverable perturbations stage failures; an expert labels visited states with actions and values. A chunked latent world model supports real-time planning, while a risk-aware sampling planner scores maneuvers by return minus a bounded collision penalty. In closed-loop comparisons, SAFER reports the highest route completion and reward per step on CARLA Leaderboard routes (0.95 and 0.77, using a privileged bird's-eye view) and highest episode reward in MetaDrive (53.4 in regular traffic; 49.0 in safety-critical traffic). The excerpt does not specify the compared methods or further evaluation limitations.

### Source excerpt

arXiv:2611.20107v1 Announce Type: cross Abstract: An autonomous vehicle must foresee the consequences of its actions, especially in the rare, time-critical situations where crashes happen. A world model offers this foresight, but only within the coverage of its training data, which poses an exploration-exploitation dilemma. Within this coverage a planner can hardly improve on the data, and beyond it the model's errors mislead it. Logs of competent driving rarely contain the failures that would widen the coverage, and agents that learn from interaction gain them only by failing themselves, without being shown what to do instead. We propose SAFER, a framework that learns offline from failures that others make, judged by an expert, and keeps learning online from its own driving. Offline, partially trained agents and scripted, recoverable perturbations Stage failures in simulation, and an expert Annotates every visited state with its action and value, so that much of what the planner explores is covered before it drives. In planning, a chunked latent world model Foresees the consequences fast enough for real-time use, and a risk-aware sampling-based planner Evaluates candidate maneuvers by their return minus a bounded collision penalty. Online, a loop continuously Refines the model wherever this driving leaves the coverage of the offline data. In closed-loop evaluation against the compared methods, SAFER reaches the highest route completion and reward per step on the CARLA Leaderboard routes, 0.95 and 0.77 from a privileged bird's-eye view, and the highest episode reward in MetaDrive, 53.4 in regular and 49.0 in safety-critical traffic. Project page: https://ys-qu.github.io/safer-website.

## Infrastructure-Based Multi-Agent Reinforcement Learning with Attention for Highway Ramp Merging

DevFeed: [Infrastructure-Based Multi-Agent Reinforcement Learning with Attention for Highway Ramp Merging](<https://devfeed.tech/articles/infrastructure-based-multi-agent-reinforcement-learning-with-attention-for-highway-ramp-merging-138988.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20078>)

Author: Chenhao Tong, Steffen M\\"uller

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [Transformer](<https://devfeed.tech/topics/transformer.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [autonomous-vehicles](<https://devfeed.tech/tags/autonomous-vehicles.md>), [communication](<https://devfeed.tech/tags/communication.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ma](<https://devfeed.tech/tags/cs-ma.md>), [cs-sy](<https://devfeed.tech/tags/cs-sy.md>), [eess-sy](<https://devfeed.tech/tags/eess-sy.md>), [flow](<https://devfeed.tech/tags/flow.md>), [highway](<https://devfeed.tech/tags/highway.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [multi-agent](<https://devfeed.tech/tags/multi-agent.md>), [ramp](<https://devfeed.tech/tags/ramp.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [safety](<https://devfeed.tech/tags/safety.md>), [sumo](<https://devfeed.tech/tags/sumo.md>), [traffic](<https://devfeed.tech/tags/traffic.md>), [transformer](<https://devfeed.tech/tags/transformer.md>)

### AI overview

The paper proposes a transformer-based multi-agent reinforcement learning controller for highway ramp merging. A roadside controller receives CAV states through V2I and sends commands; an attention-based encoder lets each vehicle select relevant information from global traffic. In SUMO comparisons with neighbor-only, global-information, and rule-based baselines, the authors report the highest traffic flow, safety, and smoothness. They also report avoiding jams during a four-hour dynamic-demand simulation, effectiveness with human-driven vehicles, and collision-free operation with maintained throughput under realistic V2I errors despite ideal-communication training. These are simulation results described in the supplied abstract.

### Source excerpt

arXiv:2611.20078v1 Announce Type: new Abstract: On-ramp merging is a typical safety and throughput problem on highways. Without proper coordination, human drivers and self-interested autonomous vehicles may make risky manoeuvres and cause accidents or congestion. With recent advances in connected autonomous vehicles (CAVs) and multi-agent reinforcement learning (MARL), vehicles can be trained to coordinate for safe and efficient merging. However, in dense traffic, forming an effective observation of the traffic state is challenging. For example, observing only the neighbouring vehicles may lead to locally optimal decisions, while observing the entire traffic can introduce much irrelevant information. In this paper, we propose a transformer-based MARL vehicle control model to address these problems. A roadside infrastructure controller receives the states of all CAVs via vehicle-to-infrastructure (V2I) communication and sends control commands back to them. An attention-based observation encoder is designed where each vehicle can filter the relevant vehicle information from the global traffic state. The proposed model is evaluated in SUMO against three baselines: a model that observes only the neighbouring vehicles, a model that encodes the global traffic information, and the rule-based SUMO model. The results show that our model achieves the highest traffic flow, safety and smoothness. Furthermore, in a four-hour simulation with dynamic traffic demand, our model successfully avoids traffic jams, while other baseline models fail at high traffic flow. We also show that our model remains effective when human-driven vehicles are present. We further show that our model, trained in ideal communication conditions, can still remain collision-free and maintain throughput under realistic V2I communication errors. Finally, we visualise the relevance weights that an ego vehicle assigns to other vehicles in multiple traffic situations.

## D'BEAST: Logged Bad Experiences as Explanation for Autonomous Driving Agents

DevFeed: [D'BEAST: Logged Bad Experiences as Explanation for Autonomous Driving Agents](<https://devfeed.tech/articles/d-beast-logged-bad-experiences-as-explanation-for-autonomous-driving-agents-137525.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20019>)

Author: Ahmad Alfan Alfian Irfan, Nur Ahmad Khatim, Mansur Arief

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [driving](<https://devfeed.tech/tags/driving.md>), [mistakes](<https://devfeed.tech/tags/mistakes.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [safety](<https://devfeed.tech/tags/safety.md>)

### AI overview

D'BEAST tests whether a tabular reinforcement-learning agent's driving decisions can be traced to logged training transitions linked to safety or task events ("bad experiences"). The authors replayed a fixed log from the same initial Q-table under three conditions: FULL keeps all updates, SKIP omits bad-experience updates, and NEUTRALIZE keeps them but removes their event penalties. In CARLA, FULL reached the goal from 33 of 40 test starts; SKIP and NEUTRALIZE reached it from none. In 46 of 47 decisions examined shortly before failures by the latter conditions, removing or neutralizing a small group of bad experiences (median one) made the trained agent choose the failing policy's action. The results support an explanation in terms of bad experiences for this agent, with evidence reported for this tabular setup and driving task.

### Source excerpt

arXiv:2611.20019v1 Announce Type: cross Abstract: Reinforcement learning agents learn from mistakes, yet their decisions can be difficult to trace back to those experiences. We investigate whether a tabular agent's decisions can be explained by its bad experiences, defined as the training transitions associated with safety or task events. A fixed training log is replayed from the same initial Q-table under three conditions. FULL retains all updates, SKIP omits bad-experience updates, and NEUTRALIZE retains those updates without their event penalties. We illustrate the method on six classic Markov decision processes and evaluate it in closed-loop driving in CARLA. On the full driving task, FULL reached the goal from 33 of 40 test starts, whereas SKIP and NEUTRALIZE reached it from none, indicating that the bad experiences carry decision-relevant information. At the local level, for 47 decisions shortly before a failure of SKIP or NEUTRALIZE, removing or neutralizing a small group of bad experiences, with a median size of one, caused the trained agent to select the failing policy's action in 46 cases. For this agent, the results indicate that its decisions can be explained by its bad experiences.

## Epigraph Form Reinforcement Learning-based Safe Trajectory Tracking Control for Connected and Autonomous Vehicles

DevFeed: [Epigraph Form Reinforcement Learning-based Safe Trajectory Tracking Control for Connected and Autonomous Vehicles](<https://devfeed.tech/articles/epigraph-form-reinforcement-learning-based-safe-trajectory-tracking-control-for-connected-and-autonomous-vehicles-132932.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.14903>)

Author: Xiaoke Yang, Jin Chen, Yafei Wang, Qi Zhang, Ming Li, Haiwen Wu, Chengliang Yin, Yong Zhang

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [autonomous-vehicles](<https://devfeed.tech/tags/autonomous-vehicles.md>), [control](<https://devfeed.tech/tags/control.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [ray-casting](<https://devfeed.tech/tags/ray-casting.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [safety](<https://devfeed.tech/tags/safety.md>), [tracking](<https://devfeed.tech/tags/tracking.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

STTC-EFRL targets safe trajectory tracking for connected and autonomous vehicles, aiming to improve tracking accuracy without sacrificing safety or training stability. It formulates tracking as an epigraph-form min-max problem and uses a decreasing epigraph-related scalar to adjust the policy-gradient direction without increasing its magnitude. A computationally efficient ray-casting method computes a polyhedron scaling factor for safety quantification. The reported experiments, including Vehicle-in-the-Loop tests, find higher tracking accuracy than the considered non-learning baseline and more stable training than widely used Lagrangian DRL methods, while meeting safety requirements in connected environments.

### Source excerpt

arXiv:2611.14903v1 Announce Type: new Abstract: Safe trajectory tracking (STT) aims to enable autonomous vehicles to follow a reference trajectory accurately while avoiding collisions. This problem is particularly challenging for Connected and Autonomous Vehicles (CAVs) in highly dynamic environments. Existing non-learning-based STT methods often use simplified vehicle and interaction models to meet real-time requirements, which may result in suboptimal tracking accuracy. Although Deep Reinforcement Learning (DRL) can help address this limitation, widely used Lagrangian DRL methods often suffer from unstable training because increasing Lagrange multipliers amplify policy-gradient magnitudes. To address these issues, we propose a novel reinforcement learning framework for STT, called STT Control Epigraph Form RL (STTC-EFRL). In this framework, the STT problem is formulated as an epigraph-form min-max optimization problem. Based on this formulation, we develop a learning mechanism that uses a decreasing scalar related to the epigraph to adjust the policy gradient descent direction without increasing the gradient magnitude, thereby improving training stability. We also develop a computationally efficient ray-casting method to compute a polyhedron scaling factor for safety quantification. Experimental results, including Vehicle-in-the-Loop tests, show that STTC-EFRL achieves higher tracking accuracy than the considered non-learning-based baseline and more stable training than widely used Lagrangian DRL methods, while satisfying safety requirements in connected environments.

## RiskFlow: Fast and Faithful Safety-Critical Traffic Scenario Generation

DevFeed: [RiskFlow: Fast and Faithful Safety-Critical Traffic Scenario Generation](<https://devfeed.tech/articles/riskflow-fast-and-faithful-safety-critical-traffic-scenario-generation-124437.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2606.06423>)

Author: Jie Li, Qi Lan, Yining Tang, Yu Shen, Yi Zhou, Yuhao Wei, Guofa Li, Hao Sun

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [ai safety](<https://devfeed.tech/topics/ai-safety.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [acceleration](<https://devfeed.tech/tags/acceleration.md>), [actions](<https://devfeed.tech/tags/actions.md>), [adversarial](<https://devfeed.tech/tags/adversarial.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [artifacts](<https://devfeed.tech/tags/artifacts.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [based](<https://devfeed.tech/tags/based.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [but](<https://devfeed.tech/tags/but.md>), [capability](<https://devfeed.tech/tags/capability.md>), [critical](<https://devfeed.tech/tags/critical.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [fast](<https://devfeed.tech/tags/fast.md>), [generation](<https://devfeed.tech/tags/generation.md>), [safety](<https://devfeed.tech/tags/safety.md>), [traffic](<https://devfeed.tech/tags/traffic.md>)

### AI overview

RiskFlow generates safety-critical traffic scenarios through one-step flow transport in action space, avoiding the repeated denoising corrections used by diffusion-based approaches. At test time, it ranks agents by time to collision, refines selected high-risk agents' actions with an adversarial interaction objective, then uses vehicle-dynamics rollout and map-aware regularization to produce feasible trajectories and discourage off-road behavior. In closed-loop nuScenes experiments, the authors report improved realism over representative baselines while retaining competitive safety-critical generation capability, with substantially lower inference time across multi-agent and long-horizon settings. The supplied abstract does not give quantitative results.

### Source excerpt

arXiv:2606.06423v2 Announce Type: replace-cross Abstract: Safety-critical traffic scenario generation is essential for evaluating autonomous driving systems under rare but high-risk interactions. Existing diffusion-based methods offer strong controllability in closed-loop generation, but their iterative denoising process is computationally expensive and may accumulate sampling and guidance errors over long rollouts, causing unrealistic motion artifacts such as abnormal acceleration and off-road behavior. To address these issues, we propose RiskFlow, a closed-loop safety-critical traffic generation framework that reformulates future behavior generation as one-step flow transport in the action space during training. At test time, RiskFlow generates safety-critical interactions through a selective guidance procedure in which time-to-collision-based critical-agent ranking selects high-risk agents and an adversarial interaction objective directly refines their output actions. Vehicle-dynamics rollout and map-aware regularization then translate guided actions into feasible trajectories while discouraging off-road behavior. By intervening on selected agents in the output space, RiskFlow avoids repeated corrections of intermediate denoising states, thereby mitigating accumulated guidance errors and accelerating closed-loop scenario generation. Experiments on the nuScenes dataset under closed-loop evaluation show that RiskFlow achieves a strong adversariality-realism trade-off across multi-agent and long-horizon settings. Compared with representative baselines, RiskFlow consistently improves realism while maintaining competitive safety-critical generation capability, and substantially reduces inference time for evaluation.

## Jev with Foresight: Safe Language-Guided Autonomous Driving via a Language World Model

DevFeed: [Jev with Foresight: Safe Language-Guided Autonomous Driving via a Language World Model](<https://devfeed.tech/articles/jev-with-foresight-safe-language-guided-autonomous-driving-via-a-language-world-model-118327.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.05448>)

Author: Yansong Qu, Jiancong Chen, Samuel Labi

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [World models](<https://devfeed.tech/topics/world-models.md>)

Tags: [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [commands](<https://devfeed.tech/tags/commands.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [decision-model](<https://devfeed.tech/tags/decision-model.md>), [jev](<https://devfeed.tech/tags/jev.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [predictions](<https://devfeed.tech/tags/predictions.md>), [vehicles](<https://devfeed.tech/tags/vehicles.md>), [world-model](<https://devfeed.tech/tags/world-model.md>)

### AI overview

Jev-AD combines Jev, a typed decision model that answers predefined questions with probabilities, with a language world model that predicts how nearby vehicles will move over the next two seconds for each candidate action. Code converts those predictions into consequences for Jev to compare, rather than requiring it to infer outcomes from scene coordinates. In simulated highway merging, foresight reduced Jev's collision rate from 26.0% to 2.4% at 0.25 seconds per decision, and the system achieved passenger intent in 93% of free-form command trials. Across 500 replayed NGSIM episodes, foresight reduced collisions from 48.2% to 2.8%; the world model, learned from that traffic, foresaw 93% of dangerous actions. These are reported simulation and replay results.

### Source excerpt

arXiv:2611.05448v1 Announce Type: cross Abstract: Large language models understand passengers' driving commands, but as driving agents they reason in generated text for seconds and must infer each action's consequences implicitly from the scene. We present Jev-AD, which pairs Jev, a typed decision model that answers predefined questions with probabilities, not generated text, with a language world model that predicts for every candidate action how nearby vehicles move over the next two seconds. Code turns these predictions into foreseen consequences, which Jev compares instead of inferring them from coordinates. In simulated highway merging (500 car-episodes), this foresight lowers the same decision model's collision rate from 26.0% to 2.4% at 0.25 seconds per decision, and Jev-AD achieves the passenger's intent in 93% of free-form command trials. In 500 replayed real-world NGSIM episodes, foresight lowers the collision rate from 48.2% to 2.8%, and the world model, learned from this traffic, foresees 93% of dangerous actions. Project page: https://ys-qu.github.io/jev-ad-website.

## Beyond Infraction Scores: Failure-Boundary Profiles for Closed-Loop Evaluation of Driving and Robot Navigation Agents

DevFeed: [Beyond Infraction Scores: Failure-Boundary Profiles for Closed-Loop Evaluation of Driving and Robot Navigation Agents](<https://devfeed.tech/articles/beyond-infraction-scores-failure-boundary-profiles-for-closed-loop-evaluation-of-driving-and-robot-navigation-agents-118077.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.06453>)

Author: Milan Bulat, Gino Ardelean, Ellington Kirby

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [Robot Navigation](<https://devfeed.tech/topics/robot-navigation.md>), [Isaac Sim](<https://devfeed.tech/topics/isaac-sim.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [boundary](<https://devfeed.tech/tags/boundary.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [driving](<https://devfeed.tech/tags/driving.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [failure](<https://devfeed.tech/tags/failure.md>), [isaac-sim](<https://devfeed.tech/tags/isaac-sim.md>), [navigation](<https://devfeed.tech/tags/navigation.md>), [robot](<https://devfeed.tech/tags/robot.md>), [robot-navigation](<https://devfeed.tech/tags/robot-navigation.md>), [robotics](<https://devfeed.tech/tags/robotics.md>), [simulator](<https://devfeed.tech/tags/simulator.md>)

### AI overview

Failure-boundary profiles distinguish agents that stop at blocked routes from those that force passage, a difference the CARLA driving score can obscure. The method varies one measured factor within each scenario family, certifies when an agent should proceed or stop, and estimates commitment and contact-free competence curves, including uncertainty from simulator run-to-run variation. Its decision-accuracy metric rewards clean passage on passable variants and stopping on blocked ones. Across seven CARLA agents and seven scenario families, the rule-based expert scored 0.74 and TFv6, the highest-scoring learned agent, 0.60; similarly scored agents showed different failure patterns, and agents that stopped at blocked corridors rarely waited for oncoming-traffic gaps. Applied unchanged to a warehouse robot in Isaac Sim, the tested policy stopped at blocked gaps in only one of three families; two fine-tuned versions passed more gaps cleanly but stopped at none. These findings are limited to the evaluated agents and scenarios.

### Source excerpt

arXiv:2611.06453v1 Announce Type: cross Abstract: Closed-loop driving benchmarks in CARLA rank agents by a driving score that multiplies route completion by an infraction penalty. This score does not distinguish an agent that stops in front of an obstacle from one that pushes past it. In a corridor closed by two parked cars, for example, one agent that drives through with contact averages 52.2 on the blocked widths, while the agents that stop average between 37.8 and 40.6. We propose failure-boundary profiles, an evaluation in which each scenario family varies one measured factor, such as corridor width or the time gap in oncoming traffic, and each variant is certified from measurement as one where the agent should proceed or one where it should stop. Along this axis we estimate two curves per agent, commitment (the probability of attempting the passage) and competence (the probability of passing without contact), and locate where each crosses one half, with intervals that account for the simulator's run-to-run variance. The certified labels also give a summary metric, decision accuracy, which credits clean passes on passable variants and stops on blocked ones. We evaluate seven driving agents on seven CARLA families. The rule-based expert reaches a decision accuracy of 0.74 and TFv6, the highest-scoring learned agent, 0.60. Agents with similar driving scores fail in different ways, and agents that reliably stop at a blocked corridor still rarely wait for a gap in oncoming traffic. We also apply the method, unchanged, to a warehouse robot in Isaac Sim. The navigation policy we test stops at blocked gaps in only one of three families, and two fine-tuned versions of it pass more gaps cleanly but no longer stop at all.

## Large Multimodal Models for Embodied Intelligent Driving: Enabling the Next Generation of Autonomous Mobility

DevFeed: [Large Multimodal Models for Embodied Intelligent Driving: Enabling the Next Generation of Autonomous Mobility](<https://devfeed.tech/articles/large-multimodal-models-for-embodied-intelligent-driving-enabling-the-next-generation-of-autonomous-mobility-117196.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2601.08434>)

Author: Long Zhang, Yuchen Xia, Bingqing Wei, Zhen Liu, Shiwen Mao, Zhu Han, Mohsen Guizani

Published: 2026-11-04T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Machine Intelligence](<https://devfeed.tech/topics/machine-intelligence.md>), [Machine Perception](<https://devfeed.tech/topics/machine-perception.md>)

Tags: [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [decision-making](<https://devfeed.tech/tags/decision-making.md>), [deep-reinforcement-learning](<https://devfeed.tech/tags/deep-reinforcement-learning.md>), [framework](<https://devfeed.tech/tags/framework.md>), [multimodal-models](<https://devfeed.tech/tags/multimodal-models.md>), [planning](<https://devfeed.tech/tags/planning.md>), [simulations](<https://devfeed.tech/tags/simulations.md>)

### AI overview

The article proposes a hybrid decision framework for embodied intelligent driving that combines Large Multimodal Models (LMMs), which provide semantic understanding and cognitive representation, with deep reinforcement learning for real-time policy optimization. The authors argue that LMMs alone do not provide the joint decision-making needed for continuous learning through closed-loop interaction. In a simulated lane-change planning case study, the framework improves performance by at least 19% over the baselines. The abstract also identifies open challenges and future research directions, but does not specify them.

### Source excerpt

arXiv:2601.08434v4 Announce Type: replace-cross Abstract: The advent of Large Multimodal Models (LMMs) offers a promising technology to tackle the limitations of modular design in autonomous driving, which often falters in open-world scenarios requiring sustained environmental understanding and logical reasoning. Besides, embodied artificial intelligence facilitates policy optimization through closed-loop interactions to achieve the continuous learning capability, thereby advancing autonomous driving toward embodied intelligent (EI) driving. However, such capability will be constrained by relying solely on LMMs to enhance EI driving without joint decision-making. This article proposes a novel semantics and policy dual-driven hybrid decision framework to tackle this challenge, ensuring continuous learning and joint decision. The framework merges LMMs for semantic understanding and cognitive representation, and deep reinforcement learning for real-time policy optimization. This article particularly details the foundational principles of EI driving and LMMs, and also introduces the emerging opportunities this framework enables, encompassing potential benefits and representative use cases. The proposed framework, evaluated through a lane-change planning case study, achieves at least a 19% performance improvement over the baselines, as verified by simulations. Finally, key open challenges and future research directions are identified to guide subsequent work.

## Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Driving

DevFeed: [Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Driving](<https://devfeed.tech/articles/distill-to-think-foresee-to-act-cognitive-physical-reinforcement-learning-for-autonomous-driving-116710.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.21139>)

Author: Yang Wu, Qiang Meng, Zhaojiang Liu, Youquan Liu, Jian Yang, Jin Xie

Published: 2026-11-04T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [World models](<https://devfeed.tech/topics/world-models.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>)

Tags: [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [safety](<https://devfeed.tech/tags/safety.md>), [world-model](<https://devfeed.tech/tags/world-model.md>)

### AI overview

CoPhy combines a distilled cognitive model with an action-conditioned physical world model to train autonomous driving policies using reinforcement learning. It distills vision-language model knowledge into a bird's-eye-view (BEV) encoder, then removes the VLM to retain the learned capability without inference cost; an optional language interface accepts human commands. An autoregressive BEV model predicts future semantic maps for candidate actions, providing rollouts from which safety metrics are derived. GRPO training uses a dual reward: physical rewards from BEV rollouts enforce safety constraints, while a language-aligned scorer rewards intent compliance. The abstract reports state-of-the-art results on NAVSIM v1 and v2, plus safer scene compliance and flexible control through user-defined language instructions; it gives no numerical results or implementation details.

### Source excerpt

arXiv:2605.21139v3 Announce Type: replace-cross Abstract: Current end-to-end autonomous driving models are fundamentally constrained by the behavioral cloning ceiling of imitation learning. While reinforcement learning offers a path to smarter autonomy, it demands two missing pieces of infrastructure: (1) a cognitive foundation that understands traffic semantics and driving intent, and (2) a foresighted physical environment that can anticipate the consequences of candidate actions. To this end, we propose CoPhy, a CognitivePhysical reinforcement learning framework for autonomous driving. To distill to think, we distill VLM knowledge into the BEV encoder and then discard the VLM entirely, retaining cognitive ability at zero inference cost while releasing the cognitive channel as a pluggable interface for optional human language commands. To foresee to act, we build an auto-regressive BEV world model that explicitly predicts future semantic maps conditioned on candidate actions, serving as an interpretable physical sandbox from which safety metrics are directly derived. Built upon this dual infrastructure, we optimize the driving policy via GRPO with a novel dual-reward mechanism: a physical reward derived from BEV rollouts enforces hard safety constraints, while a cognitive reward from a language-aligned scorer ensures intent compliance. Extensive experiments demonstrate that CoPhy not only achieves state-of-the-art results on NAVSIM v1 and v2 benchmarks, but also enables safer driving via cognitively informed scene compliance and flexible intent control through user-defined language instructions. Project page: https://wuyang98.github.io/CoPhy/.

## Scene-Adaptive Readout and Decoding in Diffusion Driving VLAs

DevFeed: [Scene-Adaptive Readout and Decoding in Diffusion Driving VLAs](<https://devfeed.tech/articles/scene-adaptive-readout-and-decoding-in-diffusion-driving-vlas-115265.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.02076>)

Author: Seungjun Yu, Jiho Choi, Inha Kang, Hojun Choi, Seungwook Kim, Hyunjung Shim

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [diffusion-transformers](<https://devfeed.tech/topics/diffusion-transformers.md>), [Machine Perception](<https://devfeed.tech/topics/machine-perception.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [adaptive](<https://devfeed.tech/tags/adaptive.md>), [after](<https://devfeed.tech/tags/after.md>), [all](<https://devfeed.tech/tags/all.md>), [also](<https://devfeed.tech/tags/also.md>), [analyze](<https://devfeed.tech/tags/analyze.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [based](<https://devfeed.tech/tags/based.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [benefit](<https://devfeed.tech/tags/benefit.md>), [code](<https://devfeed.tech/tags/code.md>), [combines](<https://devfeed.tech/tags/combines.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [driving](<https://devfeed.tech/tags/driving.md>), [e2e](<https://devfeed.tech/tags/e2e.md>), [error](<https://devfeed.tech/tags/error.md>), [first](<https://devfeed.tech/tags/first.md>)

### AI overview

Diffusion driving VLAs spend substantial inference effort rerunning the backbone at each denoising stage, but the useful readout layer and refinement depth vary by scene. The authors find that late-layer representations can yield useful trajectories after one backbone sweep, while early-layer representations benefit more from continued decoding. Their framework uses a scene-conditioned mixture of experts to combine late-layer predictions, then applies additional native refinement only when needed. Across two diffusion driving VLAs, Waymo E2E, and nuScenes, it reports validation RFS gains of up to 0.56, improved test performance, up to 6.0x end-to-end speedup on Waymo E2E, and consistently lower trajectory error on nuScenes.

### Source excerpt

arXiv:2611.02076v1 Announce Type: cross Abstract: Diffusion driving vision-language-action models repeatedly execute the backbone at every denoising stage, incurring substantial inference cost, while it remains unclear whether all scenes require the same decoding computation. To distinguish when a trajectory becomes readable from how much further refinement it requires, we analyze trajectory readout across network depth and decoding progress. We identify a depth-decoding dissociation in which late-layer representations already support useful trajectory readout after the first forward sweep, whereas early-layer representations benefit more from continued decoding. We further find that both the preferred readout layer and decoding depth vary across scenes. Based on these observations, we propose a scene-adaptive readout and decoding framework. A scene-conditioned readout MoE adaptively weights and combines trajectory predictions from layer-specific late-layer representations after a single backbone sweep, while an adaptive decoding policy applies additional native refinement only when required by the scene. We evaluate our framework on two diffusion driving VLAs across the Waymo E2E and nuScenes benchmarks. Our method improves validation RFS by up to 0.56 while also improving test performance, and achieves an end-to-end speedup of up to $6.0\times$ on Waymo E2E, while consistently reducing trajectory error on nuScenes. These results demonstrate that scene-level readout and decoding allocation improves planning quality while avoiding unnecessary decoding. Our code is available at https://github.com/sjyu001/SARD

## Mechanistic Interpretability for End-to-End Self-Driving

DevFeed: [Mechanistic Interpretability for End-to-End Self-Driving](<https://devfeed.tech/articles/mechanistic-interpretability-for-end-to-end-self-driving-111553.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.41185>)

Author: Manav Gagvani, Benjamin Namikas, Sivamurugan Velmurugan, Nysa Kumar, Shrey Sharma, Thomas Peterson, Pratyush Mathur, Hyunseo Chang, Rohan Paleja

Published: 2026-11-02T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [Tensorflow](<https://devfeed.tech/topics/tensorflow.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [adapter](<https://devfeed.tech/tags/adapter.md>), [application](<https://devfeed.tech/tags/application.md>), [apply](<https://devfeed.tech/tags/apply.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [autoencoders](<https://devfeed.tech/tags/autoencoders.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [build](<https://devfeed.tech/tags/build.md>), [but](<https://devfeed.tech/tags/but.md>), [camera](<https://devfeed.tech/tags/camera.md>), [changes](<https://devfeed.tech/tags/changes.md>), [clustering](<https://devfeed.tech/tags/clustering.md>), [code](<https://devfeed.tech/tags/code.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [driving](<https://devfeed.tech/tags/driving.md>), [e2e](<https://devfeed.tech/tags/e2e.md>), [interpretability](<https://devfeed.tech/tags/interpretability.md>), [mechanistic-interpretability](<https://devfeed.tech/tags/mechanistic-interpretability.md>), [self-driving](<https://devfeed.tech/tags/self-driving.md>)

### AI overview

The study applies Sparse Autoencoders (SAEs) to activations from a single-camera end-to-end path-planning model. Clustering SAE latents recovered route intent with 98.3% accuracy without supervision, and counterfactual image edits showed features responding to traffic light state and stop sign presence. Intervening on individual latents caused predictable, monotonic changes in lateral displacement, curvature, and braking. A lightweight adapter using SAE latents reduced trajectory selection error by 3.44% and outperformed an equivalent adapter without them. These results suggest the sparse representations expose behaviorally useful structure in this model; the supplied abstract does not establish performance beyond the studied setup.

### Source excerpt

arXiv:2610.41185v1 Announce Type: cross Abstract: End-to-end (E2E) autonomous driving models achieve strong planning performance but remain largely opaque, limiting observability and trust. We apply mechanistic interpretability to E2E driving by training Sparse Autoencoders (SAEs) on the internal activations of a simple, single-camera end-to-end path planning model. Unsupervised clustering of SAE latents reveals interpretable structure aligned with route intent, vehicle speed, and braking behavior, recovering route intent with 98.3% accuracy without supervision. Through paired counterfactual image edits, we further show that specific SAE features are visually grounded and systematically respond to scene changes such as traffic light state and stop sign presence. Steering individual SAE latents produces predictable, monotonic changes in planner outputs including lateral displacement, curvature, and braking. Finally, we demonstrate a practical application of these representations: we build a lightweight adapter trained on SAE latents that reduces trajectory selection error by 3.44%, outperforming an equivalent adapter that omits SAE latents, confirming that the sparse representation captures useful information for interpretability. Code and visualizations are available at https://mgagvani.github.io/mechanistic-interpretability-for-e2e-driving/

## Modern Cooperative Autonomous Driving Ecosystem

DevFeed: [Modern Cooperative Autonomous Driving Ecosystem](<https://devfeed.tech/articles/modern-cooperative-autonomous-driving-ecosystem-106127.md>)

Original publisher: [Read original article](<https://hackernoon.com/modern-cooperative-autonomous-driving-ecosystem?source=rss>)

Author: Alexander Borschel

Published: 2026-10-29T08:27:22Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [Data Space](<https://devfeed.tech/topics/data-space.md>), [trust](<https://devfeed.tech/topics/trust.md>), [responsible-ai](<https://devfeed.tech/topics/responsible-ai.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [automotive-communication](<https://devfeed.tech/tags/automotive-communication.md>), [automotive-tech](<https://devfeed.tech/tags/automotive-tech.md>), [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [autonomous-vehicles](<https://devfeed.tech/tags/autonomous-vehicles.md>), [autonomy](<https://devfeed.tech/tags/autonomy.md>), [awareness](<https://devfeed.tech/tags/awareness.md>), [car-tech](<https://devfeed.tech/tags/car-tech.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [collective](<https://devfeed.tech/tags/collective.md>), [communication](<https://devfeed.tech/tags/communication.md>), [communications](<https://devfeed.tech/tags/communications.md>), [compute](<https://devfeed.tech/tags/compute.md>), [ecosystem](<https://devfeed.tech/tags/ecosystem.md>), [framework](<https://devfeed.tech/tags/framework.md>), [local](<https://devfeed.tech/tags/local.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [self-driving-cars](<https://devfeed.tech/tags/self-driving-cars.md>), [sharing](<https://devfeed.tech/tags/sharing.md>), [smart-car](<https://devfeed.tech/tags/smart-car.md>), [the-future-of-smarter-care](<https://devfeed.tech/tags/the-future-of-smarter-care.md>), [trust](<https://devfeed.tech/tags/trust.md>), [vehicles](<https://devfeed.tech/tags/vehicles.md>)

### AI overview

Cooperative perception can extend an autonomous vehicle's awareness beyond what its own sensors can observe, while leaving driving decisions under local control. Vehicles and infrastructure could share timely, privacy-conscious reports of objects, hazards, signal states, and intent through a common external interface; manufacturers would remain free to use their own sensors, models, and confidence systems. The article distinguishes established V2V/V2X communication and standards work from making shared observations a usable, manufacturer-neutral evidence layer. It emphasizes that messages need provenance, timestamps, uncertainty, bounded lifetimes, plausibility checks, and revocable credentials. Stale, contradictory, unavailable, or invalid information should not prevent a vehicle from operating on local perception. Passive road users need not identify themselves, while optional active participation could add observations. Deployment still depends on incentives, shared requirements, governance, and clear accountability.

### Source excerpt

Why autonomous vehicles need cooperative perception: a cross-ecosystem framework for sharing observations while preserving local autonomy, trust and privacy.

## DriveMind: Mind-Evolving Belief Tracking for Closed-Loop Autonomous Driving

DevFeed: [DriveMind: Mind-Evolving Belief Tracking for Closed-Loop Autonomous Driving](<https://devfeed.tech/articles/drivemind-mind-evolving-belief-tracking-for-closed-loop-autonomous-driving-99939.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.32069>)

Author: Yan Shi, Kehua Chen, Jiarui Qi, Yihong Tang, Haichen Liao, Yinhai Wang

Published: 2026-10-26T04:00:00Z

Content type: news

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [Machine Perception](<https://devfeed.tech/topics/machine-perception.md>), [Physical AI](<https://devfeed.tech/topics/physical-ai.md>)

Tags: [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [interactive](<https://devfeed.tech/tags/interactive.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [tracking](<https://devfeed.tech/tags/tracking.md>)

### AI overview

DriveMind improves the same MLLM driving pipeline by maintaining multiple scene hypotheses and updating their weights as new frames arrive. Inspired by particle filtering, it propagates hypotheses through a VLM, scores them against each observation, and resamples to preserve diversity. On Bench2Drive, belief tracking raises Driving Score from 68.50 to 81.37 and Success Rate from 41.36% to 54.09%; the largest reported gains are at unsignalized intersections (+20.0 DS) and in oncoming-traffic interactions (+20.5 DS). The abstract also reports improvements in merging, emergency braking, and give-way.

### Source excerpt

arXiv:2610.32069v1 Announce Type: new Abstract: Recent multimodal large language models (MLLMs) can perceive a traffic scene, reason about it, and propose a driving action within a single model. However, existing MLLM-based drivers reason from scratch at each frame and commit to a single explanation, discarding earlier hypotheses about the scene. When the scene is ambiguous, the agent locks in on the wrong interpretation and acts on it before evidence accumulates. In this paper, we propose DriveMind, an MLLM-based driving framework that keeps several hypotheses about the scene alive at the same time and updates their weights as new frames arrive. Inspired by particle filtering, DriveMind propagates each hypothesis through a VLM, scores it against the current observation, and resamples to maintain diversity. On Bench2Drive, adding belief tracking to the same driving pipeline improves the Driving Score (DS) from 68.50 to 81.37 and the Success Rate (SR) from 41.36% to 54.09%. The largest gains occur on ambiguous scenarios such as unsignalized intersections (+20.0 DS) and oncoming-traffic interactions (+20.5 DS). Beyond the overall metrics, DriveMind also improves fine-grained interactive driving abilities such as merging, emergency braking, and give-way.

## Super Agents and Confounders: Influence of surrounding agents on vehicle trajectory prediction

DevFeed: [Super Agents and Confounders: Influence of surrounding agents on vehicle trajectory prediction](<https://devfeed.tech/articles/super-agents-and-confounders-influence-of-surrounding-agents-on-vehicle-trajectory-prediction-96828.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2604.03463>)

Author: Daniel Jost, Luca Paparusso, Martin Stoll, J\\"org Wagner, Raghu Rajan, Joschka B\\"odecker

Published: 2026-10-23T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [driving](<https://devfeed.tech/tags/driving.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [robustness](<https://devfeed.tech/tags/robustness.md>)

### AI overview

The study finds that adding surrounding traffic agents can reduce vehicle trajectory prediction accuracy: many predictors appear to rely on unstable, non-causal signals that vary across training runs. The authors use Shapley-based attribution to analyze agent influence and propose a Conditional Information Bottleneck (CIB) that compresses agent features and filters out features unhelpful to prediction, without requiring additional supervision. Across multiple datasets and model architectures, they report improved overall performance in many cases and greater robustness to perturbations. The abstract also says they provide interpretable metrics for identifying non-robust behavior; it does not specify the datasets, metrics, or detailed results.

### Source excerpt

arXiv:2604.03463v2 Announce Type: replace Abstract: In highly interactive driving scenes, trajectory prediction is conditioned on information from surrounding traffic participants such as cars and pedestrians. Our main contribution is a comprehensive analysis of state-of-the-art trajectory predictors, which reveals a surprising and critical flaw: many surrounding agents degrade prediction accuracy rather than improve it. Using Shapley-based attribution, we rigorously demonstrate that models learn unstable and non-causal decision-making schemes that vary significantly across training runs. Building on these insights, we propose to integrate a Conditional Information Bottleneck (CIB), which does not require additional supervision and is trained to effectively compress agent features as well as ignore those that are not beneficial for the prediction task. Comprehensive experiments using multiple datasets and model architectures demonstrate that this simple yet effective approach not only improves overall trajectory prediction performance in many cases but also increases robustness to different perturbations. Our results highlight the importance of selectively integrating contextual information, which can often contain spurious or misleading signals, in trajectory prediction. Moreover, we provide interpretable metrics for identifying non-robust behavior and present a promising avenue towards a solution.

## Measuring the Reasoning-Action Coupling in Driving Vision-Language-Action Models through Counterfactuals

DevFeed: [Measuring the Reasoning-Action Coupling in Driving Vision-Language-Action Models through Counterfactuals](<https://devfeed.tech/articles/measuring-the-reasoning-action-coupling-in-driving-vision-language-action-models-through-counterfactuals-98142.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.30953>)

Author: Gabriel Manalu, Saad Wazir, Thao Do, Seongah Kim, Hyeontaek Hwang, Tolcha Yalew Kidane, Daeyoung Kim

Published: 2026-10-23T04:00:00Z

Content type: news

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [vlm](<https://devfeed.tech/topics/vlm.md>), [Chain-of-thought](<https://devfeed.tech/topics/chain-of-thought.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [camera](<https://devfeed.tech/tags/camera.md>), [chain-of-thought](<https://devfeed.tech/tags/chain-of-thought.md>), [coupling](<https://devfeed.tech/tags/coupling.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [domain](<https://devfeed.tech/tags/domain.md>), [driving](<https://devfeed.tech/tags/driving.md>), [explainability](<https://devfeed.tech/tags/explainability.md>), [language](<https://devfeed.tech/tags/language.md>), [measuring](<https://devfeed.tech/tags/measuring.md>), [models](<https://devfeed.tech/tags/models.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [safety](<https://devfeed.tech/tags/safety.md>), [small](<https://devfeed.tech/tags/small.md>), [vision](<https://devfeed.tech/tags/vision.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>), [vision-language-action-models](<https://devfeed.tech/tags/vision-language-action-models.md>)

### AI overview

The study finds that driving VLA models' stated reasoning is weakly coupled to their trajectories, limiting the explanations those systems provide. Across four models and different benchmarks, changing the meaning of a model's reasoning caused smaller trajectory changes than blanking its camera inputs. The researchers also used photorealistic scene counterfactuals that removed key objects named in the reasoning, then examined whether the reasoning and trajectories tracked those changes. The abstract does not report detailed results for that second intervention.

### Source excerpt

arXiv:2610.30953v1 Announce Type: new Abstract: Driving vision-language-action (VLA) models bring an appealing form of explainability to a safety-critical domain, since a system that can state the reasons for its maneuvers offers the transparency that deployment and public trust in autonomous driving demand. These models read the world into language before they act on it, describing the scene and reasoning over it in text that ranges from chain-of-thought to multi-stage question answering on the way to a trajectory. That framing quietly assumes two separable properties, namely that the trajectory follows the generated reasoning and that the generated reasoning tracks the causes that actually produce the trajectory. To thoroughly investigate these properties, we apply causal, counterfactual interventions to four driving VLAs across different benchmarks. First, we perturb the models' own reasoning and show that altering its meaning produces small trajectory changes relative to blanking the same models' camera inputs, a pattern that persists across benchmarks and model architectures. We then construct scene-level counterfactuals through photorealistic removal of key objects named in the models' reasoning and analyze whether the generated reasoning and trajectories track these changes. Overall, our findings indicate that the reasoning these models present to account for their resulting driving behavior is weakly coupled to that behavior itself, pointing to limits on the explainability these systems are intended to provide.

## PCLA-Bridge: A Framework for Cross-Domain Deployment of CARLA Agents

DevFeed: [PCLA-Bridge: A Framework for Cross-Domain Deployment of CARLA Agents](<https://devfeed.tech/articles/pcla-bridge-a-framework-for-cross-domain-deployment-of-carla-agents-94614.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.28696>)

Author: Masoud Jamshidiyan Tehrani, Jinhan Kim, Paolo Tonella

Published: 2026-10-22T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [Tool](<https://devfeed.tech/topics/tool.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [actions](<https://devfeed.tech/tags/actions.md>), [adas](<https://devfeed.tech/tags/adas.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [all](<https://devfeed.tech/tags/all.md>), [analyze](<https://devfeed.tech/tags/analyze.md>), [apis](<https://devfeed.tech/tags/apis.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [bridge](<https://devfeed.tech/tags/bridge.md>), [changes](<https://devfeed.tech/tags/changes.md>), [clock](<https://devfeed.tech/tags/clock.md>), [code](<https://devfeed.tech/tags/code.md>), [control](<https://devfeed.tech/tags/control.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [domain](<https://devfeed.tech/tags/domain.md>), [framework](<https://devfeed.tech/tags/framework.md>)

### AI overview

PCLA-Bridge runs unmodified CARLA Leaderboard agents on different sensor sources and control destinations without an active CARLA server. It handles dependency isolation, tensor marshalling, routing, and synthetic timing; integrators provide a minimal interface, including any needed output translation and coordinate-frame transformations. The framework was evaluated in offline replay on CARLA recordings and nuScenes logs, and in live closed-loop control with MetaDrive, AWSIM, and AlpaSim. All eight evaluated agents ran across all three simulators without modification. The supplied abstract reports minimal glue code, but does not provide detailed performance results or establish deployment on physical vehicles.

### Source excerpt

arXiv:2610.28696v1 Announce Type: new Abstract: Most high-performing end-to-end autonomous driving agents (ADAs) are developed for the CARLA Leaderboard, tightly coupling them to CARLA's APIs, sensor layouts, coordinate systems, and internal clock. Reusing these models on real datasets, alternative simulators, or physical platforms necessitates substantial manual re-engineering. We present PCLA-Bridge, a deployable tool that runs unmodified CARLA Leaderboard agents from any sensor source to any control destination without an active CARLA server. With PCLA-Bridge handling dependency isolation, tensor marshalling, routing, and synthetic timing, integrators only need to provide a minimal interface. We validate the tool across two operational paradigms: offline replay on in-domain CARLA recordings and real-world nuScenes logs, and live closed-loop control across three diverse simulation engines: MetaDrive, AWSIM, and AlpaSim. We analyze the lightweight integration requirements for each platform, highlighting output translation (such as converting agent actions into target trajectory formats) and coordinate-frame transformations, showing that integration requires minimal glue code rather than structural changes. All eight evaluated agents execute across all three simulators without modification, confirming the reliability of PCLA-Bridge and establishing a foundation for systematic sim-to-real, cross-sim, and security testing.

## When Does Experience Memory Help Closed-Loop LLM Driving Agents? A Controlled Evaluation with Resampled Warmup Pools, Token-Matched Fillers, and Audited Baselines

DevFeed: [When Does Experience Memory Help Closed-Loop LLM Driving Agents? A Controlled Evaluation with Resampled Warmup Pools, Token-Matched Fillers, and Audited Baselines](<https://devfeed.tech/articles/when-does-experience-memory-help-closed-loop-llm-driving-agents-a-controlled-evaluation-with-resampled-warmup-pools-token-matched-fillers-and-audited-baselines-94335.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.27714>)

Author: Yingxin Jia, Xuanyi Li

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [driving](<https://devfeed.tech/tags/driving.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [experience](<https://devfeed.tech/tags/experience.md>), [llm](<https://devfeed.tech/tags/llm.md>), [memory](<https://devfeed.tech/tags/memory.md>), [token](<https://devfeed.tech/tags/token.md>)

### AI overview

On the evaluated HighwayEnv testbed (1 Hz, with gpt-4o-mini for decisive contrasts), apparent gains from LLM driving-agent memory depended on the sampled warmup experiences and did not exceed token-matched irrelevant filler. Across ten disjoint pools, both retrieval memories beat No-Memory in only one pool; the reported pool was the best-performing draw for each. A 15-token caution instruction matched the collision-only merge ceiling, and pricing caution erased the rulebook's lead. The tested memory representations showed no robust advantage, while audited baseline-port quality could reverse conclusions from single-pool comparisons. These findings are specific to this testbed; the authors say they release its setup, ports, seeds, and caches.

### Source excerpt

arXiv:2610.27714v1 Announce Type: cross Abstract: When does experience memory help a closed-loop LLM driving agent? Published gains cannot say, because the two factors we find to matter most, which warmup experiences the pool contains and what any injected tokens do to the prompt, are controlled in none of them. We answer with a controlled evaluation on one audited HighwayEnv testbed (1Hz, gpt-4o-mini for the decisive contrasts). We treat the warmup pool as a resampled random factor (ten disjoint pools) and add token-matched irrelevant-filler controls, oracle-certified diagnostics, caution-pricing metrics, and quantified audits of the baseline ports. The probe suite spans flat-text retrieval (a DiLu port), a single-frame grid port (RESPOND-inspired), our episodic temporal-graph memory, a retrieval-free consolidated rulebook, and a flat K-step observation-history baseline. Four findings follow, each a property of this testbed. (1) Headline memory gains belong to the warmup draw. The reported pool is the best of ten for both retrieval memories, which beat No-Memory in only one pool of ten (sign test p=0.011), and even the rulebook's apparent pool-immunity collapsed on five fresh draws. (2) No retrieval design tested exceeds token-matched filler. The collision-only merge ceiling is reached by a single 15-token caution instruction, and metrics that price caution erase the rulebook's lead. (3) Representational sophistication is nowhere robustly operative. (4) Baseline port quality silently inverts conclusions. On this testbed, a single-pool comparison without filler controls would have reported memory gains that these controls reverse. We release the testbed, audited ports, seeds, and caches.

## Dream Labs raises $90M to help build more intelligent robots with simple video feeds

DevFeed: [Dream Labs raises $90M to help build more intelligent robots with simple video feeds](<https://devfeed.tech/articles/dream-labs-raises-90m-to-help-build-more-intelligent-robots-with-simple-video-feeds-90482.md>)

Original publisher: [Read original article](<https://siliconangle.com/2026/10/19/dream-labs-raises-90m-to-help-build-more-intelligent-robots-with-simple-video-feeds/>)

Author: Mike Wheatley

Published: 2026-10-20T01:45:10Z

Content type: news

Language: en

Sources: [SiliconANGLE](<https://devfeed.tech/sources/siliconangle.md>)

Topics: [FIRST Robotics Competition](<https://devfeed.tech/topics/first-robotics-competition.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [Frontier AI](<https://devfeed.tech/topics/frontier-ai.md>), [lunar foundation model](<https://devfeed.tech/topics/lunar-foundation-model.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [application-programming-interface](<https://devfeed.tech/tags/application-programming-interface.md>), [b-capital](<https://devfeed.tech/tags/b-capital.md>), [dream-1](<https://devfeed.tech/tags/dream-1.md>), [dream-labs](<https://devfeed.tech/tags/dream-labs.md>), [emerging-tech](<https://devfeed.tech/tags/emerging-tech.md>), [foundation-model](<https://devfeed.tech/tags/foundation-model.md>), [funding](<https://devfeed.tech/tags/funding.md>), [intelligent-robots](<https://devfeed.tech/tags/intelligent-robots.md>), [joel-jang](<https://devfeed.tech/tags/joel-jang.md>), [news](<https://devfeed.tech/tags/news.md>), [robot-intelligence](<https://devfeed.tech/tags/robot-intelligence.md>), [robot-training](<https://devfeed.tech/tags/robot-training.md>), [robotic-ai](<https://devfeed.tech/tags/robotic-ai.md>), [robotic-training](<https://devfeed.tech/tags/robotic-training.md>), [robots](<https://devfeed.tech/tags/robots.md>), [seed-round](<https://devfeed.tech/tags/seed-round.md>), [the-latest](<https://devfeed.tech/tags/the-latest.md>)

### AI overview

Dream Labs raised $90 million in seed funding as it launched a public API for Dream-1, a foundation model intended to help robots learn tasks from video and natural language instructions. Developers can use the API to translate visual inputs and prompts into robotic actions, or submit task videos for fine-tuning. The company also launched a browser-based Robot Playground.

### Source excerpt

Dream Labs, a startup trying to make robots more intelligent and capable of performing many different tasks, has raised $90 million in seed funding to coincide with the launch of a public application programming interface that provides access to its universal robot foundation model. Today's round was jointly led by B Capital and Kindred Ventures [...] The post Dream Labs raises $90M to help build more intelligent robots with simple video feeds appeared first on SiliconANGLE.

## End-to-End AI vs. Traditional Self-Driving Systems

DevFeed: [End-to-End AI vs. Traditional Self-Driving Systems](<https://devfeed.tech/articles/end-to-end-ai-vs-traditional-self-driving-systems-90141.md>)

Original publisher: [Read original article](<https://hackernoon.com/end-to-end-ai-vs-traditional-self-driving-systems?source=rss>)

Author: vishwa vimukthi gammuduwaththage

Published: 2026-10-19T14:45:51Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [autonomous vehicles](<https://devfeed.tech/topics/autonomous-vehicles.md>), [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [World models](<https://devfeed.tech/topics/world-models.md>), [Physical AI](<https://devfeed.tech/topics/physical-ai.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [autonomous-driving](<https://devfeed.tech/tags/autonomous-driving.md>), [autonomous-vehicle-regulation](<https://devfeed.tech/tags/autonomous-vehicle-regulation.md>), [autonomous-vehicle-safety](<https://devfeed.tech/tags/autonomous-vehicle-safety.md>), [autonomy](<https://devfeed.tech/tags/autonomy.md>), [av](<https://devfeed.tech/tags/av.md>), [cars](<https://devfeed.tech/tags/cars.md>), [debugging](<https://devfeed.tech/tags/debugging.md>), [driving](<https://devfeed.tech/tags/driving.md>), [end-to-end-autonomous-driving](<https://devfeed.tech/tags/end-to-end-autonomous-driving.md>), [neural-network-explainability](<https://devfeed.tech/tags/neural-network-explainability.md>), [nvidia-alpamayo](<https://devfeed.tech/tags/nvidia-alpamayo.md>), [safety](<https://devfeed.tech/tags/safety.md>), [self-driving](<https://devfeed.tech/tags/self-driving.md>), [systems](<https://devfeed.tech/tags/systems.md>), [tesla](<https://devfeed.tech/tags/tesla.md>), [tesla-cybercab](<https://devfeed.tech/tags/tesla-cybercab.md>), [vision-language-action-models](<https://devfeed.tech/tags/vision-language-action-models.md>), [waymo](<https://devfeed.tech/tags/waymo.md>), [wayve-ai-driver](<https://devfeed.tech/tags/wayve-ai-driver.md>)

### AI overview

The article compares modular and end-to-end systems for autonomous driving, focusing on debugging, validation, safety oversight, regulation, and sensing and mapping choices. It argues that deployed systems combine learned drivers with retained structure, independent safety layers, and evidence to support safety claims.

### Source excerpt

Tesla, Wayve, and Waymo are advancing AI-driven cars. Explore the challenges of debugging neural drivers, verifying safety, and regulating autonomy.

## Grounding AI agents in biological reality: Building blocks for autonomous wet labs

DevFeed: [Grounding AI agents in biological reality: Building blocks for autonomous wet labs](<https://devfeed.tech/articles/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs-88899.md>)

Original publisher: [Read original article](<https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/>)

Published: 2026-10-16T18:11:34Z

Content type: article

Language: en

Sources: [The latest research from Google](<https://devfeed.tech/sources/the-latest-research-from-google.md>)

Topics: [AI agents in autonomous vehicles](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>), [gui-agents](<https://devfeed.tech/topics/gui-agents.md>), [Ollama](<https://devfeed.tech/topics/ollama.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [automation](<https://devfeed.tech/tags/automation.md>), [digital-twin](<https://devfeed.tech/tags/digital-twin.md>), [general-science](<https://devfeed.tech/tags/general-science.md>), [health-bioscience](<https://devfeed.tech/tags/health-bioscience.md>), [robotics](<https://devfeed.tech/tags/robotics.md>)

### AI overview

Google Research describes three projects toward autonomous wet labs: VERID, a stateful digital twin and automation framework for lab robotics; closed-loop AI-guided aptamer discovery; and agent-guided protein binder design. The work reports improvements in experimental hit rates and a faster protein design cycle, while emphasizing human oversight, reliable protocols, and monitoring agent execution.

### Source excerpt

General Science

[Next page](<https://devfeed.tech/topics/ai-agents-in-autonomous-vehicles.md?cursor=WyIyMDI2LTEwLTE2VDE4OjExOjM0KzAwOjAwIiwgImQ0ZTJlODVhLTJhNGYtNDFiNC1hYmQwLTU2ZmUzYWZjY2ZlYiJd>)