VLM-Safe-RL is proposed, a framework that integrates frozen CLIP signals into PPO-Lagrangian through reward shaping and an augmented multiplier update, and shows a conditional reduction in observed catastrophe rate without evidence of collision anticipation.
Abstract
Safe reinforcement learning seeks policies that maximise task performance while satisfying safety constraints. In driving benchmarks, however, collision costs typically appear only at the time of collision, providing no advance warning of an approaching hazard. Frozen vision--language models can provide dense semantic feedback, yet it remains unclear whether their scores anticipate collisions and which component drives an observed safety improvement. Episodic cost can also favour policies that make little task progress. To address these gaps, we propose VLM-Safe-RL, a framework that integrates frozen CLIP signals into PPO-Lagrangian through reward shaping and an augmented multiplier update. On MetaDrive Hard, which combines the densest traffic with the largest map, the catastrophe rate falls from 31.6\% to 19.4\%. FormulaOne-L2 analysis finds no evidence that the CLIP signals anticipate collisions and shows that the VLM term has a negligible effect on the Lagrange multiplier. These findings show a conditional reduction in observed catastrophe rate without evidence of collision anticipation.
Frozen vision-language models increasingly provide safety signals for reinforcement learning. Their use assumes that similarity to language describing danger indicates the hazard itself. Yet policy return and collision rate cannot reveal whether a score detects hazards or responds to correlated features of the scene. V...
ShieldVLA is proposed, a safety-aligned fine-tuning framework for VLA models based on Hamilton-Jacobi (HJ) reachability that reduces cumulative safety cost by 57% on average and improves task success rate by +0.13 over SafeVLA.
This work presents Vision-Language-Poisson-Safe Actions (VLPSA), a safety filtering framework that provides full-body safety for VLA policies in cluttered and dynamic environments without retraining, and deploys VLPSA on a Franka FR3 in cluttered scenes with dynamic obstacles and human interference.
Meg Wilkinson, E. Fourney, Joel W. Burdick et al.· 0 citations
FIND is introduced, an agentic real-world RL framework that closes the loop between scene understanding, weakness-aware practice, self-evaluation, and policy improvement in a persistent workspace and reframes autonomous practice as a scene-conditioned, performance-aware task-selection problem.
Vision-Language-Action (VLA) models for autonomous driving rely heavily on successful expert demonstrations, leaving model-specific failures underexploited. Learning from these failures is hindered by unreliable diagnoses, poorly matched correction targets, and coarse rewards. We propose RefineDrive, a failure-guided p...
Zhe Sun, Zi-Yin Luo, Ye-Hao Lu et al.· 0 citations
Vision-language-action (VLA) models with billions of parameters now dominate the LIBERO manipulation benchmark, but the model capacity actually required by the benchmark remains unclear. We introduce MINERVA (MINimal Efficient Robotic Vision-Action policy), a family of deliberately compact visuomotor policies designed...
Kohei Sendai, T. Matsushima, Yusuke Iwasawa· 3 citations· ⚡1
Related blog posts
Microsoft Research Blog· microsoft.comNov 16, 2026
AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduNov 13, 2026