Reproduce “AIM: Adaptive Interaction Modeling Networks for Real-to-Sim Soft-Body Simulation” 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

Tiancheng Yang, Dingshuo Chen, Tianle Chen, Zhaocheng Liu, Qiang Liu

AIM predicts deformable-object motion by adapting particle-to-particle communication from motion history and geometry, while sharing object-wide context. It trains with multi-step autoregressive supervision, feeding predicted states back without intermediate ground-truth resets. On a 37-step interaction test, AIM’s step-30 displacement error is 24.8 mm versus PGND’s 32.1 mm; errors still accumulate, and results do not establish safe robotic deployment.

Summary by alphaXiv.

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