When should we prefer Decision Transformers for Offline Reinforcement Learning?
This work addresses the problem of algorithm selection in offline RL for researchers and practitioners, providing incremental insights into when to prefer DT over other methods based on data and task conditions.
The paper empirically compares Decision Transformers (DT), Conservative Q-Learning (CQL), and Behavior Cloning (BC) for offline reinforcement learning across benchmarks like D4RL and Robomimic, finding that DT is more robust and better in sparse-reward and low-quality data settings, while scaling data for DT by 5x improves Atari scores by 2.5x on average.
Offline reinforcement learning (RL) allows agents to learn effective, return-maximizing policies from a static dataset. Three popular algorithms for offline RL are Conservative Q-Learning (CQL), Behavior Cloning (BC), and Decision Transformer (DT), from the class of Q-Learning, Imitation Learning, and Sequence Modeling respectively. A key open question is: which algorithm is preferred under what conditions? We study this question empirically by exploring the performance of these algorithms across the commonly used D4RL and Robomimic benchmarks. We design targeted experiments to understand their behavior concerning data suboptimality, task complexity, and stochasticity. Our key findings are: (1) DT requires more data than CQL to learn competitive policies but is more robust; (2) DT is a substantially better choice than both CQL and BC in sparse-reward and low-quality data settings; (3) DT and BC are preferable as task horizon increases, or when data is obtained from human demonstrators; and (4) CQL excels in situations characterized by the combination of high stochasticity and low data quality. We also investigate architectural choices and scaling trends for DT on Atari and D4RL and make design/scaling recommendations. We find that scaling the amount of data for DT by 5x gives a 2.5x average score improvement on Atari.