Quality-controllable robot policies use iterative hindsight relabeling
Original title: Learning Quality-Controllable Robot Policies with Iterative Hindsight Relabeling
Speed control raises conveyor throughput, and stability control improves success while carrying wine glasses.
Quality-Controllable Policy lets a robot policy set execution qualities such as speed and motion smoothness, alone or combined, at inference time. An iterative hindsight relabeling loop collects online rollouts under target quality labels, keeps successful trajectories, relabels them by measured quality and fine-tunes the policy. On Robomimic and RoboEval the authors report higher execution quality and success across iterations, including rollouts above the initial demonstrations. Real manipulation tests use speed control on conveyor work and transport-stability control when carrying wine glasses.
- QCP controls individual quality attributes and combinations at inference time.
- The loop extends the controllable quality range beyond the initial demonstrations.
- Successful rollouts exceed the quality of the starting demonstration data.
- Real tests cover conveyor throughput and carrying wine glasses.
- AI models
- Quality-Controllable Policy
From the original article
1 min read at huggingface.co
Robot policy evaluation often relies on a binary success signal for task completion. In real-world deployment, however, desired execution behaviors are also defined by quality attributes such as execution speed and motion smoothness, which need to be dynamically controlled according to task requirements.
By Sinjae Kang, Dong Kyu Shin, Jaehong Kim · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.