The authors trained a probabilistic, state-conditioned play policy through behavior cloning, using human play rather than demonstrations of specific tasks.
Minutes of supervised play bootstrap hours of autonomous robot practice
A cloned play policy collects contact-rich data and guides world-model planning for pushing and multifingered object reorientation.
Big Tech
Botao He · Yiming Dou · Yiannis Aloimonos · Jitendra Malik · Akash Sharma · Ruoshi Liu
Amazon Frontier AI & Robotics (FAR) · University of Maryland, College Park
Research Digest··2 min read
He and colleagues present DSP, a framework that learns safe, goal-agnostic play from minutes of human operation, then runs that policy autonomously to gather hours of robot experience.
Why this paper
From Amazon Frontier AI & Robotics (FAR) and University of Maryland, College Park · Part of Robot World Models, now 31 papers
In one line
A robot learns dexterous manipulation by mimicking minutes of human play, then autonomously practices to train a world model and guide control.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§