Learning human behaviors from motion capture by adversarial imitation

TL;DR

This paper introduces an adversarial imitation learning method that trains neural policies to generate humanlike movements from limited motion capture data, even across different physical parameters, enabling reusable sub-skill policies.

Contribution

It extends generative adversarial imitation learning to work with limited, partially observed data and unknown physical parameters, improving humanlike motion generation.

Findings

Successfully learned humanlike movement patterns from limited data

Reused sub-skill policies for task solving with hierarchical control

Produced more natural movements than traditional reinforcement learning methods

Abstract

Rapid progress in deep reinforcement learning has made it increasingly feasible to train controllers for high-dimensional humanoid bodies. However, methods that use pure reinforcement learning with simple reward functions tend to produce non-humanlike and overly stereotyped movement behaviors. In this work, we extend generative adversarial imitation learning to enable training of generic neural network policies to produce humanlike movement patterns from limited demonstrations consisting only of partially observed state features, without access to actions, even when the demonstrations come from a body with different and unknown physical parameters. We leverage this approach to build sub-skill policies from motion capture data and show that they can be reused to solve tasks when controlled by a higher level controller.