Discrete Policy: Learning Disentangled Action Space for Multi-Task Robotic Manipulation

TL;DR

This paper introduces Discrete Policy, a novel method using vector quantization to learn disentangled, discrete action representations for multi-task robotic manipulation, improving performance over existing approaches in simulation and real-world settings.

Contribution

The paper proposes Discrete Policy, a new approach that maps action sequences into a discrete latent space for multi-task robotic manipulation, enabling better generalization and task-specific action learning.

Findings

Outperforms Diffusion Policy and state-of-the-art methods in success rate.

Achieves 26% higher success in 5-task real-world setting.

Performance gap widens to 32.5% as tasks increase to 12.

Abstract

Learning visuomotor policy for multi-task robotic manipulation has been a long-standing challenge for the robotics community. The difficulty lies in the diversity of action space: typically, a goal can be accomplished in multiple ways, resulting in a multimodal action distribution for a single task. The complexity of action distribution escalates as the number of tasks increases. In this work, we propose \textbf{Discrete Policy}, a robot learning method for training universal agents capable of multi-task manipulation skills. Discrete Policy employs vector quantization to map action sequences into a discrete latent space, facilitating the learning of task-specific codes. These codes are then reconstructed into the action space conditioned on observations and language instruction. We evaluate our method on both simulation and multiple real-world embodiments, including both single-arm and bimanual robot settings. We demonstrate that our proposed Discrete Policy outperforms a well-established Diffusion Policy baseline and many state-of-the-art approaches, including ACT, Octo, and OpenVLA. For example, in a real-world multi-task training setting with five tasks, Discrete Policy achieves an average success rate that is 26\% higher than Diffusion Policy and 15\% higher than OpenVLA. As the number of tasks increases to 12, the performance gap between Discrete Policy and Diffusion Policy widens to 32.5\%, further showcasing the advantages of our approach. Our work empirically demonstrates that learning multi-task policies within the latent space is a vital step toward achieving general-purpose agents.