Group Distributionally Robust Reinforcement Learning with Hierarchical Latent Variables
This addresses task ambiguity in multi-task RL for applications like robotics and gaming, but it is incremental as it builds on existing robust RL methods with a novel hierarchical formulation.
The paper tackles the problem of multi-task reinforcement learning without task indicators by proposing Group Distributionally Robust Markov Decision Process (GDR-MDP), which balances worst-case robustness and average performance using a hierarchical latent variable model, and shows that their algorithms outperform classic robust training methods in experiments on control tasks and benchmarks.
One key challenge for multi-task Reinforcement learning (RL) in practice is the absence of task indicators. Robust RL has been applied to deal with task ambiguity, but may result in over-conservative policies. To balance the worst-case (robustness) and average performance, we propose Group Distributionally Robust Markov Decision Process (GDR-MDP), a flexible hierarchical MDP formulation that encodes task groups via a latent mixture model. GDR-MDP identifies the optimal policy that maximizes the expected return under the worst-possible qualified belief over task groups within an ambiguity set. We rigorously show that GDR-MDP's hierarchical structure improves distributional robustness by adding regularization to the worst possible outcomes. We then develop deep RL algorithms for GDR-MDP for both value-based and policy-based RL methods. Extensive experiments on Box2D control tasks, MuJoCo benchmarks, and Google football platforms show that our algorithms outperform classic robust training algorithms across diverse environments in terms of robustness under belief uncertainties. Demos are available on our project page (\url{https://sites.google.com/view/gdr-rl/home}).