Recover, Discover, Plan: Learning Skills and Concepts from Robot Failures

2026-06-16

AI Summary

Current robotic systems typically recover passively from failures and struggle to extract abstract knowledge from experience to proactively avoid future errors. This work proposes ReSYNC, a recovery-driven, two-stage learning framework that jointly optimizes recovery skills and relational state abstractions, enabling the continuous discovery and refinement of relational predicates from failure–recovery experiences for the first time. By integrating reinforcement learning with symbolic reasoning, ReSYNC employs an incremental, alternating strategy for skill and concept acquisition, supporting abstract planning beyond grasping tasks and facilitating sim-to-real transfer. Experimental results demonstrate that ReSYNC achieves over 50% higher success rates than strong baselines across four simulated environments and successfully generalizes to previously unseen task scenarios in the real world.

Abstract

Intelligent robots should not only recover from failures, but also acquire the abstract knowledge needed to avoid them in the future. While reinforcement learning (RL) can learn reactive recovery behaviors, training a separate policy for every distinct failure mode is highly inefficient. We introduce Recovery-Driven Synthesis of Relational Concepts (ReSYNC), the first approach that progressively discovers and refines state abstractions (relational predicates) from failure-recovery experience to support abstract planning. Unlike purely reactive methods, ReSYNC jointly learns skills and concepts through an incremental dual-learning process. In the skill-learning phase, the robot uses RL to learn to recover from failures seen in training tasks. In the concept-learning phase, the robot discovers new relational predicates and refines its abstract planning model to explain and generalize the learned recovery behaviors. This interaction enables ReSYNC to convert local recoveries seen during training into global failure avoidance at test time. Across four simulated domains, we show that ReSYNC's ability to continually expand and refine its abstraction library allows it to solve long-horizon, previously unseen problems, outperforming strong baselines by over 50%. Additionally, we demonstrate sim-to-real transfer of ReSYNC, where it performs real-world non-prehensile manipulation skills and generalizes to unseen scenarios through abstract planning. Overall, ReSYNC represents a significant step toward robots that autonomously acquire abstractions for scalable, failure-aware planning in the physical world.

Problem

• robot failures
• state abstractions
• relational predicates
• abstract planning
• failure recovery

Innovation

• Recovery-Driven Learning
• Relational Concept Discovery
• Abstract Planning
• Dual-Learning Framework
• Sim-to-Real Transfer