Debiased Model-based Representations for Sample-efficient Continuous Control

Researchers introduced DR.Q, a debiased model-based representation method for sample-efficient continuous control.

DR.Q maximizes mutual information between state-action and next-state representations, while using faded prioritized experience replay.

It is designed to reduce representation bias and overfitting, improving stability in off-policy Q-learning.

On continuous control benchmarks, DR.Q matched or exceeded strong baselines.