The authors built OCBench in MJWarp, a GPU-accelerated implementation of MuJoCo based on NVIDIA Warp.
Human-like scripted demonstrations reproduce behavioral cloning’s puzzling failures
OCBench recreates counterintuitive robot-learning behaviors in controllable, GPU-accelerated simulations by modeling structural features of human demonstrations.
Top University
Seohong Park · Sergey Levine
University of California, Berkeley
Research Digest··2 min read
Park and Levine introduce OCBench, a simulated robotic manipulation benchmark designed to reproduce behavioral cloning phenomena that usually appear only with costly human-collected data.
Why this paper
From University of California, Berkeley
In one line
Behavioral cloning performance can improve with overfitting, and fully closed-loop policies often fail without action chunking.
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 (2 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§