Quality-Controllable Robot Policies via Iterative Hindsight Relabeling
October 26, 2026
The Quality-Controllable Policy (QCP) framework allows for inference-time control over robot execution attributes like speed and smoothness. It utilizes an iterative hindsight relabeling loop to expand the controllable range of behaviors beyond the initial demonstration dataset.
HOW THIS AFFECTS YOU
●
builderYou can dynamically adjust robot movement characteristics during deployment.
●
researcherThis method addresses the limitation of binary success signals in policy training.