FAR: Failure-Aware Retry for Test-Time Recovery and Continual Policy Improvement

This paper proposes FAR, a framework that allows robots to learn from failures during deployment, adapt their behavior, and complete tasks autonomously. FAR combines failure-contrastive preference adaptation with lightweight action perturbations to encourage local exploration and integrates successful recovery trajectories into a training loop for continual policy improvement. Experiments show substantial gains in success rates and data efficiency in both simulation and real-world settings.