Reproduce “Visuo-tactile Diffusion Policy for Robotic Manipulation with Auxiliary Tactile Dynamics Prediction” 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
Kaihong Ouyang, Zihao Su, Chengxuan Zheng, Weiyi Lai, Dong Zhou
The authors train an assembly policy to use vision and rendered fingertip images, with an auxiliary task predicting how those images change under action. This prediction supervises a temporal representation for action generation. In simulation, the combined model achieves 59.0% success, versus 42.7% for vision-only across three seeds; results apply to rendered proxies and geometries seen in training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.