RAIL: A modular framework for Reinforcement-learning-based Adversarial Imitation Learning

TL;DR

This paper introduces RAIL, a modular framework for Adversarial Imitation Learning, and proposes two new algorithms, demonstrating improved performance on locomotion tasks through extensive evaluation.

Contribution

The paper presents RAIL, a flexible framework that unifies existing AIL approaches and introduces two novel algorithms, SAIfO and SILEM, for imitation learning from observation.

Findings

SAIfO outperforms existing RAIL algorithms on locomotion benchmarks.

The modular design of RAIL facilitates understanding and development of AIL algorithms.

SILEM addresses embodiment mismatch in imitation learning.

Abstract

While Adversarial Imitation Learning (AIL) algorithms have recently led to state-of-the-art results on various imitation learning benchmarks, it is unclear as to what impact various design decisions have on performance. To this end, we present here an organizing, modular framework called Reinforcement-learning-based Adversarial Imitation Learning (RAIL) that encompasses and generalizes a popular subclass of existing AIL approaches. Using the view espoused by RAIL, we create two new IfO (Imitation from Observation) algorithms, which we term SAIfO: SAC-based Adversarial Imitation from Observation and SILEM (Skeletal Feature Compensation for Imitation Learning with Embodiment Mismatch). We go into greater depth about SILEM in a separate technical report. In this paper, we focus on SAIfO, evaluating it on a suite of locomotion tasks from OpenAI Gym, and showing that it outperforms contemporaneous RAIL algorithms that perform IfO.