Dreamer-CDP: Improving Reconstruction-free World Models Via Continuous Deterministic Representation Prediction
This work addresses the problem of learning robust world models for MBRL agents in high-dimensional observation spaces, particularly for researchers exploring alternatives to reconstruction-based methods.
This paper introduces a JEPA-style predictor for continuous, deterministic representations to improve reconstruction-free world models in model-based reinforcement learning. The proposed method achieves performance comparable to Dreamer on the Crafter environment, demonstrating effective world model learning without relying on reconstruction objectives.
Model-based reinforcement learning (MBRL) agents operating in high-dimensional observation spaces, such as Dreamer, rely on learning abstract representations for effective planning and control. Existing approaches typically employ reconstruction-based objectives in the observation space, which can render representations sensitive to task-irrelevant details. Recent alternatives trade reconstruction for auxiliary action prediction heads or view augmentation strategies, but perform worse in the Crafter environment than reconstruction-based methods. We close this gap between Dreamer and reconstruction-free models by introducing a JEPA-style predictor defined on continuous, deterministic representations. Our method matches Dreamer's performance on Crafter, demonstrating effective world model learning on this benchmark without reconstruction objectives.