Meta reinforcement learning (meta-RL) methods such as RL have emerged as promising approaches for learning data-efficient RL algorithms tailored to a given task distribution. However, they show poor asymptotic performance and struggle with out-of-distribution tasks because they rely on sequence models, such as recurrent neural networks or transformers, to process experiences rather than summarize them using general-purpose RL components such as value functions. In contrast, traditional RL algorithms are data-inefficient as they do not use domain knowledge, but they do converge to an optimal policy in the limit. We propose RL, a principled hybrid approach that incorporates action-values, learned per task through traditional RL, in the inputs to meta-RL. We show that RL earns greater cumulative reward in the long term, compared to RL, while maintaining data-efficiency in the short term, and generalizes better to out-of-distribution tasks. Experiments are conducted on both custom and benchmark discrete domains from the meta-RL literature that exhibit a range of short-term, long-term, and complex dependencies.
View on arXiv