HILONet: Hierarchical Imitation Learning from Non-Aligned Observations
This addresses a practical limitation in imitation learning for robotics or AI agents where aligned demonstrations are hard to obtain, though it appears incremental as it builds on hierarchical and observation-based methods.
The paper tackles the problem of imitation learning from non-time-aligned observation-only demonstrations, which are common in real-world scenarios, by proposing HILONet, a hierarchical method that dynamically selects sub-goals, resulting in improved performance and learning efficiency across various environments.
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.