HYDRA: Hybrid Robot Actions for Imitation Learning

TL;DR

HYDRA introduces a hybrid action space with dynamic switching and relabeling to improve imitation learning, significantly reducing distribution shift and enhancing performance in complex robot tasks.

Contribution

HYDRA's novel hybrid action space and dynamic switching mechanism address distribution shift, outperforming prior methods in diverse long-horizon robot tasks.

Findings

Outperforms prior IL methods by 30-40% in various environments

Enables both coarse and fine control through dynamic action abstraction switching

Effective in real-world tasks like making coffee and toasting bread

Abstract

Imitation Learning (IL) is a sample efficient paradigm for robot learning using expert demonstrations. However, policies learned through IL suffer from state distribution shift at test time, due to compounding errors in action prediction which lead to previously unseen states. Choosing an action representation for the policy that minimizes this distribution shift is critical in imitation learning. Prior work propose using temporal action abstractions to reduce compounding errors, but they often sacrifice policy dexterity or require domain-specific knowledge. To address these trade-offs, we introduce HYDRA, a method that leverages a hybrid action space with two levels of action abstractions: sparse high-level waypoints and dense low-level actions. HYDRA dynamically switches between action abstractions at test time to enable both coarse and fine-grained control of a robot. In addition, HYDRA employs action relabeling to increase the consistency of actions in the dataset, further reducing distribution shift. HYDRA outperforms prior imitation learning methods by 30-40% on seven challenging simulation and real world environments, involving long-horizon tasks in the real world like making coffee and toasting bread. Videos are found on our website: https://tinyurl.com/3mc6793z