A Continual Offline Reinforcement Learning Benchmark for Navigation Tasks
This work addresses the problem of catastrophic forgetting and adaptation in continual reinforcement learning for researchers and practitioners in domains like robotics and gaming, though it is incremental as it builds on existing advances to fill a gap in the literature.
The authors tackled the need for a standardized evaluation framework in continual reinforcement learning by introducing a benchmark suite of video-game navigation tasks, which includes datasets, evaluation protocols, and metrics to assess algorithms, including state-of-the-art baselines.
Autonomous agents operating in domains such as robotics or video game simulations must adapt to changing tasks without forgetting about the previous ones. This process called Continual Reinforcement Learning poses non-trivial difficulties, from preventing catastrophic forgetting to ensuring the scalability of the approaches considered. Building on recent advances, we introduce a benchmark providing a suite of video-game navigation scenarios, thus filling a gap in the literature and capturing key challenges : catastrophic forgetting, task adaptation, and memory efficiency. We define a set of various tasks and datasets, evaluation protocols, and metrics to assess the performance of algorithms, including state-of-the-art baselines. Our benchmark is designed not only to foster reproducible research and to accelerate progress in continual reinforcement learning for gaming, but also to provide a reproducible framework for production pipelines -- helping practitioners to identify and to apply effective approaches.