Reproduce “SOFTMAP: Sim2Real Soft Robot Forward Modeling via Topological Mesh Alignment and Physics Prior” 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

Ziyong Ma, Uksang Yoo, Jonathan Francis, Weiming Zhi, Jeffrey Ichnowski, Jean Oh

Researchers at Carnegie Mellon University developed SOFTMAP, a sim-to-real learning framework for real-time 3D forward modeling of tendon-actuated soft finger manipulators, combining physics-based simulation pre-training with a data-efficient residual correction network and topological mesh alignment. The framework achieved millimeter-level shape prediction accuracy and improved performance in trajectory tracking and teleoperation tasks on physical hardware by up to 89.5% IoU compared to baseline methods.

Key insights

  • ARAP-based topological alignment is crucial for establishing consistent vertex correspondences across simulated meshes and noisy real-world point clouds, enabling effective cross-domain supervision.
  • Leveraging a strong, physics-consistent prior learned from abundant simulation data significantly reduces the real-world data burden required for accurate soft robot forward modeling.
  • A lightweight residual correction mechanism, using LayerNorm and GELU activations, can effectively adapt a simulation-pretrained model to real-world complexities with minimal data, mitigating the 'reality gap'.

Summary by alphaXiv.

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