World models adapt optimistically at test time to improve planning success
Adapt for Planning, Not just Prediction: Optimistic Test-Time World Model Adaptation
Machine LearningArtificial IntelligenceRobotics
Summary
Planning with AI models can get worse when the environment changes, because the models are trained beforehand and then frozen. The authors show that updating these models during use, guided not just by prediction accuracy but also by an optimistic goal-reaching bias, helps the AI plan better. Their method adjusts the model to be hopeful about reaching goals based on imagined scenarios, which improves the success rate of planning tasks like manipulation and navigation. This approach works across different models and planners, demonstrating that being optimistic helps fix problems caused by unexpected changes.
What this means in practice
- •For robotics engineers: Improve robot navigation and manipulation by adapting world models during deployment for better goal-directed planning under changing conditions.
- •For autonomous vehicle developers: Enhance route planning in autonomous vehicles by updating predictive models on the fly to cope with unexpected environment changes.
Authors
Akshay Mete, Dileep Kalathil, P. R. Kumar
Abstract
Latent world models are typically pre-trained offline and kept frozen at test time, so planning with them can degrade under distribution shift. Test-time adaptation (TTA) can mitigate this degradation by updating the model during deployment. Unlike pre-training on fixed data, TTA for planning is closed-loop: the planner uses model rollouts to select actions, and the resulting transitions drive the next update. This raises the question: which objective should guide TTA for planning? Under a prediction-only objective, the model is fit only to the outcomes of its own actions, so it can remain wrong about the actions that would reach the goal, and low prediction error need not translate into better planning. Inspired by reward-biased maximum likelihood from adaptive control, we introduce Optimistic Test-Time Adaptation (OpTTA). Its objective has two terms, the prediction loss on observed transitions and the planner's goal-reaching cost on its candidate rollouts, both minimized over the model parameters. The planning term biases adaptation toward models that are optimistic about reaching the goal, without any additional environment interaction. Across visual and environment shifts on manipulation, reaching and navigation tasks, OpTTA raises final planning success over the frozen model by 13.5 percentage points on average, compared with 6.9 for the prediction-only objective (AdaJEPA), when only the predictor is adapted. When the encoder is adapted as well, the gains are 13.5 and 7.5 points. These experiments span three JEPA world models with different encoders and training objectives and, for temporal straightening, both CEM and gradient-based planners. Our results suggest that test-time adaptation of world models for planning benefits from an optimistic, task-aware term in addition to the prediction loss.