MoSA: Motion-constrained Stress Adaptation for Mitigating the Real-to-Sim Gap in Continuum Dynamics via Learning Residual Anisotropy

Researchers introduced MoSA, a physics-informed framework that learns residual stress operators on top of an isotropic base model.

The method uses motion-based supervision on deformation derivatives to capture mild anisotropy and heterogeneity in continuum dynamics.

MoSA outperforms prior approaches in prediction and generalization, narrowing the real-to-sim gap.

The approach also improves sim-to-real transfer in a robot manipulation task, making calibration more robust to material irregularities.