Get Started
Home
Topics
Search
Library
Research questionHow can robotic manipulation models recognize and recover from failures during task execution?Robots can complete benchmark tasks successfully yet fail when grasps slip, collisions occur, or objects move unexpectedly. Reliable manipulation therefore requires recognizing these disruptions and continuing the task from a changed state.
AI
Evaluation & Benchmarks
Multimodal Models
Robotics
Latest papersRecent research connected to this question, newest first.LIBERO-RECOVER: Beyond Task Success Towards Failure Recovery in Robotic Manipulation ModelsThe source presents LIBERO-Recover, a benchmark for vision-language-action and world-action models in robotic manipulation. It uses more than 1,000 real execution-failure scenarios spanning action retry, action adaptation, object-state recovery, and environmental recovery, and examines spatial, object-structure, interaction, and topological reasoning. The evidence is benchmark-based and built on LIBERO executions.research paper · Sep 4, 2026
Related questions
How can robots safely learn dynamic manipulation skills online despite sim-to-real mismatch?How can robots robustly execute changing tasks that coordinate humans, cranes, and manipulators?How can vision-language-action policies act reliably when irrelevant sensors are corrupted or only one informative sensor remains?How can robot manipulation policies be evaluated for execution quality without costly, unstable repeated hardware trials?