Topic: corrective learning

RedFlow Turns Robot Failures Into Action-Level Corrections

The paper proposes converting failed actions into localized corrections for flow-matching vision-language-action policies.