Forager: Lightweight RL Testbed for Continual Learning

Forager is a lightweight, partially-observable continual reinforcement learning (CRL) environment introduced in arXiv paper 2605.01131. It addresses the gap in CRL research where most experiments focus on loss of plasticity in fully observable MDPs with added non-stationarity, ignoring partial observability and memory-based agents. Forager maintains a constant memory footprint, making it accessible for repeated experiments. The paper provides sample tasks demonstrating that Forager challenges current CRL agents while enabling further research.