Adversarial Motion Priors Make Good Substitutes for Complex Reward Functions

TL;DR

This paper introduces adversarial motion priors as a way to replace complex reward functions in reinforcement learning, enabling naturalistic, transferable behaviors in simulated and real quadrupedal robots with minimal data.

Contribution

It demonstrates that style rewards learned from motion capture data can effectively substitute complex rewards, simplifying training and improving transferability to real robots.

Findings

Style rewards enable natural gait transitions.

Policies trained with style rewards transfer to real robots.

Few seconds of motion data suffice for effective style learning.

Abstract

Training a high-dimensional simulated agent with an under-specified reward function often leads the agent to learn physically infeasible strategies that are ineffective when deployed in the real world. To mitigate these unnatural behaviors, reinforcement learning practitioners often utilize complex reward functions that encourage physically plausible behaviors. However, a tedious labor-intensive tuning process is often required to create hand-designed rewards which might not easily generalize across platforms and tasks. We propose substituting complex reward functions with "style rewards" learned from a dataset of motion capture demonstrations. A learned style reward can be combined with an arbitrary task reward to train policies that perform tasks using naturalistic strategies. These natural strategies can also facilitate transfer to the real world. We build upon Adversarial Motion Priors -- an approach from the computer graphics domain that encodes a style reward from a dataset of reference motions -- to demonstrate that an adversarial approach to training policies can produce behaviors that transfer to a real quadrupedal robot without requiring complex reward functions. We also demonstrate that an effective style reward can be learned from a few seconds of motion capture data gathered from a German Shepherd and leads to energy-efficient locomotion strategies with natural gait transitions.