DIDI: Diffusion-Guided Diversity for Offline Behavioral Generation

TL;DR

DIDI introduces a diffusion-guided approach to offline behavioral generation, enabling the learning of diverse, discriminative, and reward-guided skills from offline data across multiple decision-making domains.

Contribution

The paper presents a novel diffusion-guided method for offline skill learning, incorporating diversity and regularization to produce a versatile and generalist skill space.

Findings

Effective in discovering diverse behaviors in four decision-making domains

Enables skill stitching and interpolation for flexible behavior generation

Incorporates reward guidance to optimize behaviors from sub-optimal data

Abstract

In this paper, we propose a novel approach called DIffusion-guided DIversity (DIDI) for offline behavioral generation. The goal of DIDI is to learn a diverse set of skills from a mixture of label-free offline data. We achieve this by leveraging diffusion probabilistic models as priors to guide the learning process and regularize the policy. By optimizing a joint objective that incorporates diversity and diffusion-guided regularization, we encourage the emergence of diverse behaviors while maintaining the similarity to the offline data. Experimental results in four decision-making domains (Push, Kitchen, Humanoid, and D4RL tasks) show that DIDI is effective in discovering diverse and discriminative skills. We also introduce skill stitching and skill interpolation, which highlight the generalist nature of the learned skill space. Further, by incorporating an extrinsic reward function, DIDI enables reward-guided behavior generation, facilitating the learning of diverse and optimal behaviors from sub-optimal data.