Progressive Reinforcement Learning with Distillation for Multi-Skilled Motion Control

TL;DR

This paper introduces PLAID, a progressive reinforcement learning approach that combines multiple skills into a single policy using distillation, input augmentation, and transfer learning, demonstrated on simulated bipedal locomotion tasks.

Contribution

The paper extends policy distillation to continuous actions and proposes a novel framework for incremental skill integration in reinforcement learning.

Findings

PLAID effectively merges multiple skills into a single policy.

The method outperforms three baseline approaches in simulated locomotion tasks.

Input injection and transfer learning enhance skill acquisition efficiency.

Abstract

Deep reinforcement learning has demonstrated increasing capabilities for continuous control problems, including agents that can move with skill and agility through their environment. An open problem in this setting is that of developing good strategies for integrating or merging policies for multiple skills, where each individual skill is a specialist in a specific skill and its associated state distribution. We extend policy distillation methods to the continuous action setting and leverage this technique to combine expert policies, as evaluated in the domain of simulated bipedal locomotion across different classes of terrain. We also introduce an input injection method for augmenting an existing policy network to exploit new input features. Lastly, our method uses transfer learning to assist in the efficient acquisition of new skills. The combination of these methods allows a policy to be incrementally augmented with new skills. We compare our progressive learning and integration via distillation (PLAID) method against three alternative baselines.