NovelGym
A Flexible Ecosystem for Simulating Gridworld Environments
Modular
Features easy task and novelty design for developing and evaluating open-world agents, in single- and multi-agent scenarios.
Seamless
Works with agents of different architectures, including symbolic planning agents, reinforcement learners, and hybrid neurosymbolic architectures.
Exemplative
Offers benchmarks of state-of-the-art learning and hybrid methods for novelty handling.
Evaluative
Employs agent-agnostic evaluation metrics for novelty adaptation.