Safe Multi-Agent Reinforcement Learning on Pistonball 10 agents
4.919Constraint Violation RateScalable Primal-Dual Actor-Critic
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Scalable Primal-Dual Actor-CriticAlgorithm=Ours2023.05 | 4.919 | — | |
| MAPPO-LInformation Access=Global2023.05 | 6.884 | — | |
| Decentralized MAPPO-LInformation Access=Local neighborhood2023.05 | 9.303 | — | |
| Decentralized Aggregate MAPPO-LInformation Access=Local neighborhood, Reward structure=Sum of rewards in local neighborhood2023.05 | 21.79 | — |