Latent Action World Models for Control with Unlabeled Trajectories

TL;DR

This paper introduces latent-action world models that effectively learn from both labeled and unlabeled data, improving control performance with fewer action-labeled samples by combining passive observations and active interactions.

Contribution

It proposes a novel latent-action representation that unifies action-conditioned and action-free data, enabling more efficient training of world models for control tasks.

Findings

Achieves strong performance on DeepMind Control Suite

Uses about ten times fewer action-labeled samples than baselines

Enables training on passive and interactive data simultaneously

Abstract

Inspired by how humans combine direct interaction with action-free experience (e.g., videos), we study world models that learn from heterogeneous data. Standard world models typically rely on action-conditioned trajectories, which limits effectiveness when action labels are scarce. We introduce a family of latent-action world models that jointly use action-conditioned and action-free data by learning a shared latent action representation. This latent space aligns observed control signals with actions inferred from passive observations, enabling a single dynamics model to train on large-scale unlabeled trajectories while requiring only a small set of action-labeled ones. We use the latent-action world model to learn a latent-action policy through offline reinforcement learning (RL), thereby bridging two traditionally separate domains: offline RL, which typically relies on action-conditioned data, and action-free training, which is rarely used with subsequent RL. On the DeepMind Control Suite, our approach achieves strong performance while using about an order of magnitude fewer action-labeled samples than purely action-conditioned baselines. These results show that latent actions enable training on both passive and interactive data, which makes world models learn more efficiently.