Reproduce “Same Label, Different Interface: Beyond Action-Space Labels in Robot Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Arshia Sangwan, Zhongyang Zhang
The authors test how a robot policy’s position scale affects simulated manipulation outcomes. They compare calibrated and joint-limit-based scales within a fixed position-feedback law, while also matching the action-driven raw-torque gain between position and velocity interfaces. On PegInsert, median success was 1.00 versus 0.00 across ten policies per condition; this simulation does not establish an optimal scale or real-robot performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.