Structured Policy Representation: Imposing Stability in arbitrarily conditioned dynamic systems

TL;DR

This paper introduces a new class of neural network-based dynamic systems that are globally stable and can be conditioned on arbitrary context states, enabling structured and reliable robot policies.

Contribution

The paper proposes a novel family of deep neural network dynamics that are globally stable and conditionable, enhancing robot policy design.

Findings

Dynamics are globally stable and can be conditioned arbitrarily

Applicable as structured robot policies with stability guarantees

Facilitates reasonable behaviors outside demonstration regions

Abstract

We present a new family of deep neural network-based dynamic systems. The presented dynamics are globally stable and can be conditioned with an arbitrary context state. We show how these dynamics can be used as structured robot policies. Global stability is one of the most important and straightforward inductive biases as it allows us to impose reasonable behaviors outside the region of the demonstrations.