Robot Failure Recovery

Robot failure recovery aims to help robots continue a task or reach a safe state after actions go wrong, sensors fail, or the environment changes. Approaches include failure detection and diagnosis, learned corrective actions and retries, rollback to safe checkpoints, replanning with vision-language models, and fault-tolerant control for systems such as drones and humanoids. Work is tested in manipulation benchmarks and physical settings, with growing attention to long-horizon tasks, contact-rich interactions, and human assistance. A central challenge is that detecting an error does not guarantee that intervention will help: recovery must fit the specific state and task, remain safe, and generalize from simulation or training data to real robots.