Abstract:Training general agents to follow complex instructions (tasks) in intricate environments (levels) remains a core challenge in reinforcement learning. Random sampling of task-level pairs often produces unsolvable combinations, highlighting the need to co-design tasks and levels. While unsupervised environment design (UED) has proven effective at automatically designing level curricula, prior work has only considered a fixed task. We present ATLAS (Aligning Tasks and Levels for Autocurricula of Specifications), a novel method that generates joint autocurricula over tasks and levels. Our approach builds upon UED to automatically produce solvable yet challenging task-level pairs for policy training. To evaluate ATLAS and drive progress in the field, we introduce an evaluation suite that models tasks as reward machines in Minigrid levels. Experiments demonstrate that ATLAS vastly outperforms random sampling approaches, particularly when sampling solvable pairs is unlikely. We further show that mutations leveraging the structure of both tasks and levels accelerate convergence to performant policies.
Abstract:Ever since the concepts of dynamic programming were introduced, one of the most difficult challenges has been to adequately address high-dimensional control problems. With growing dimensionality, the utilisation of Deep Neural Networks promises to circumvent the issue of an otherwise exponentially increasing complexity. The paper specifically investigates the sampling issues the Deep Galerkin Method is subjected to. It proposes a drift relaxation-based sampling approach to alleviate the symptoms of high-variance policy approximations. This is validated on mean-field control problems; namely, the variations of the opinion dynamics presented by the Sznajd and the Hegselmann-Krause model. The resulting policies induce a significant cost reduction over manually optimised control functions and show improvements on the Linear-Quadratic Regulator problem over the Deep FBSDE approach.