Context-aware Learned Mesh-based Simulation via Trajectory-Level Meta-Learning

TL;DR

This paper introduces M3GN, a meta-learning approach for mesh-based simulation that leverages trajectory-level data and movement primitives to achieve rapid adaptation, higher accuracy, and faster computation compared to existing methods.

Contribution

It proposes a novel trajectory-level meta-learning framework using Conditional Neural Processes for mesh simulation, enabling quick adaptation and improved accuracy.

Findings

M3GN outperforms state-of-the-art GNSs in accuracy.

M3GN achieves faster simulation runtimes.

The method effectively adapts to new scenarios with limited data.

Abstract

Simulating object deformations is a critical challenge across many scientific domains, including robotics, manufacturing, and structural mechanics. Learned Graph Network Simulators (GNSs) offer a promising alternative to traditional mesh-based physics simulators. Their speed and inherent differentiability make them particularly well suited for applications that require fast and accurate simulations, such as robotic manipulation or manufacturing optimization. However, existing learned simulators typically rely on single-step observations, which limits their ability to exploit temporal context. Without this information, these models fail to infer, e.g., material properties. Further, they rely on auto-regressive rollouts, which quickly accumulate error for long trajectories. We instead frame mesh-based simulation as a trajectory-level meta-learning problem. Using Conditional Neural Processes, our method enables rapid adaptation to new simulation scenarios from limited initial data while capturing their latent simulation properties. We utilize movement primitives to directly predict fast, stable and accurate simulations from a single model call. The resulting approach, Movement-primitive Meta-MeshGraphNet (M3GN), provides higher simulation accuracy at a fraction of the runtime cost compared to state-of-the-art GNSs across several tasks.