HILONet: Hierarchical Imitation Learning from Non-Aligned Observations

TL;DR

HILONet introduces a hierarchical imitation learning framework that dynamically selects sub-goals from demonstrations, enabling efficient learning in non-aligned observation scenarios across various tasks.

Contribution

It presents a novel hierarchical approach for imitation learning from non-time-aligned observations, improving flexibility and sample efficiency over existing methods.

Findings

Enhanced performance in diverse tasks

Improved learning efficiency

Effective handling of non-aligned demonstrations

Abstract

It is challenging learning from demonstrated observation-only trajectories in a non-time-aligned environment because most imitation learning methods aim to imitate experts by following the demonstration step-by-step. However, aligned demonstrations are seldom obtainable in real-world scenarios. In this work, we propose a new imitation learning approach called Hierarchical Imitation Learning from Observation(HILONet), which adopts a hierarchical structure to choose feasible sub-goals from demonstrated observations dynamically. Our method can solve all kinds of tasks by achieving these sub-goals, whether it has a single goal position or not. We also present three different ways to increase sample efficiency in the hierarchical structure. We conduct extensive experiments using several environments. The results show the improvement in both performance and learning efficiency.