WorldAgen improves robot actions by learning from new environments fast
WorldAgen: Unified State-Action Prediction with Test-Time World Model Training
Artificial Intelligence
Summary
Robots and AI systems often struggle when facing new settings where things behave differently than before. The researchers created WorldAgen, a system that combines understanding how the world works and deciding what actions to take, using a single smart model. It can learn and adjust to new environments while it's working, by trying out actions and quickly updating its knowledge. This makes it better at predicting what to do next, even in unfamiliar situations. Tests showed that WorldAgen can perform as well or better than other methods, especially after it adapts in real time.
vision-language-action modelsworld modelingaction predictiontest-time trainingTransformer neural networksstate transitionsexploratory actionsCALVIN benchmarkLIBERO benchmarkunseen environment adaptation
Authors
Chi Wan, Kangrui Wang, Yuan Si, Pingyue Zhang, Manling Li
Abstract
How can vision-language-action (VLA) models adapt to new environments where world dynamics shift? While recent research has combined world modeling and action prediction to improve VLA performance, existing methods largely rely on pretraining on static datasets, without mechanisms for active adaptation at deployment time. As a result, these models often fail to generalize when deployed in unseen scenarios with novel object configurations or dynamics. We present WorldAgen, a unified framework that jointly learns world modeling and action prediction while enabling Test-Time Training (TTT) to adapt to new environments. WorldAgen employs a shared Transformer backbone with two heads: (1) a world model head that predicts future states from past state-action trajectories, and (2) an agent model head that predicts actions conditioned on task instructions. We design a Mixed Unidirectional Attention Mask to separate these two models. During test time, WorldAgen samples exploratory actions, collects ground-truth state transitions, and performs lightweight TTT updates to refine its world model. This adaptation improves the model's understanding of the environment and leads to more accurate action predictions. Experiments on the CALVIN and LIBERO benchmarks demonstrate that our baseline model achieves comparable, and in some cases superior, performance to current state-of-the-art approaches. Moreover, with TTT on a small number of samples, our method surpasses existing state-of-the-art models, highlighting the effectiveness of adapting world models at inference time.