Fundamental Challenges in Deep Learning for Stiff Contact Dynamics

TL;DR

This paper investigates the difficulties deep learning models face in accurately capturing the non-smooth, discontinuous behavior of contact dynamics in robotics, highlighting the impact of mechanical stiffness on training and generalization.

Contribution

It provides empirical evidence that current deep learning approaches struggle with contact non-smoothness and shows how stiffness affects model training and transferability.

Findings

Stiffness significantly degrades training and generalization.

Deep learning models are not well-suited for contact non-smoothness.

Simulated environments may misrepresent real hardware stiffness.

Abstract

Frictional contact has been extensively studied as the core underlying behavior of legged locomotion and manipulation, and its nearly-discontinuous nature makes planning and control difficult even when an accurate model of the robot is available. Here, we present empirical evidence that learning an accurate model in the first place can be confounded by contact, as modern deep learning approaches are not designed to capture this non-smoothness. We isolate the effects of contact's non-smoothness by varying the mechanical stiffness of a compliant contact simulator. Even for a simple system, we find that stiffness alone dramatically degrades training processes, generalization, and data-efficiency. Our results raise serious questions about simulated testing environments which do not accurately reflect the stiffness of rigid robotic hardware. Significant additional investigation will be necessary to fully understand and mitigate these effects, and we suggest several avenues for future study.