Reproduce “Objectives as Policies: Vision-Language Model Cost Programs for Contact-Rich Manipulation” 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

Zihao Lu, Trong Gia Hung Nguyen, Fuquan Wang, Yifan Zhu

OaP lets a vision-language model write staged objectives for a shared physics-based controller. The controller uses model-predictive control to plan contact-rich actions, while simulated rollouts let the model revise objectives before deployment. In four tabletop tasks, OaP succeeds in 36/40 simulation trials, compared with 31/40 for adapted CoRAL. On hardware, selected simulation-successful programs succeed in 32/40 perturbed executions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.