Hierarchical Imitation and Reinforcement Learning
This addresses the problem of inefficient policy learning in complex environments for AI and robotics researchers, offering an incremental improvement by combining existing methods hierarchically.
The paper tackles the challenge of learning sequential decision-making policies in sparse-reward, long-horizon environments by proposing a hierarchical guidance framework that integrates imitation and reinforcement learning. It demonstrates faster learning than hierarchical RL and greater label efficiency than standard IL, with significant reductions in expert effort and exploration costs on benchmarks like Montezuma's Revenge.
We study how to effectively leverage expert feedback to learn sequential decision-making policies. We focus on problems with sparse rewards and long time horizons, which typically pose significant challenges in reinforcement learning. We propose an algorithmic framework, called hierarchical guidance, that leverages the hierarchical structure of the underlying problem to integrate different modes of expert interaction. Our framework can incorporate different combinations of imitation learning (IL) and reinforcement learning (RL) at different levels, leading to dramatic reductions in both expert effort and cost of exploration. Using long-horizon benchmarks, including Montezuma's Revenge, we demonstrate that our approach can learn significantly faster than hierarchical RL, and be significantly more label-efficient than standard IL. We also theoretically analyze labeling cost for certain instantiations of our framework.