Mo-Gymnasium
Emerging4papers using it
2024first seen
The 'Mo-Gymnasium' is a simulated control task used to evaluate the effectiveness of the MO-GRPO algorithm in addressing reward hacking in multi-objective settings.
Papers using Mo-Gymnasium (4)
- MO-GRPO: Mitigating Reward Hacking of Group Relative Policy Optimization on Multi-Objective ProblemsA Single Deep Preference-Conditioned Policy for Learning Pareto Coverage SetsA Reward-Free Viewpoint on Multi-Objective Reinforcement LearningMOMAland: A Set of Benchmarks for Multi-Objective Multi-Agent
Reinforcement Learning