Reproduce “SleepWalking: Privileged Representation Shaping for End-to-End Blind Locomotion in Legged Robots” 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
Zheng Pan, Tenghui Wang, Peilin Li, Shiyu Zhou, Hao Sun, Yan Ma, Liang Yu, Liang He
SWAQ trains a robot to navigate uneven terrain by using simulator-only physical targets to supervise its recurrent history representation. The targets guide what the policy’s memory retains, but reconstructed quantities are not inputs to the deployed controller. In humanoid simulation, SWAQ reaches terrain level 5.5410, versus 4.8168 for DWAQ, with 44.4% fewer inference MACs per control step; a real-world stair trial illustrates contact-based adjustment, not general obstacle capability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.