Fully-Decentralized MADDPG with Networked Agents
This work addresses the computational efficiency challenge in multi-agent systems for researchers and practitioners, though it is incremental as it builds on existing MADDPG methods.
The paper tackled the problem of decentralized training in multi-agent reinforcement learning for continuous action spaces by adapting MADDPG with networked communication and surrogate policies, achieving comparable results to the original MADDPG while reducing computational cost, especially with larger numbers of agents.
In this paper, we devise three actor-critic algorithms with decentralized training for multi-agent reinforcement learning in cooperative, adversarial, and mixed settings with continuous action spaces. To this goal, we adapt the MADDPG algorithm by applying a networked communication approach between agents. We introduce surrogate policies in order to decentralize the training while allowing for local communication during training. The decentralized algorithms achieve comparable results to the original MADDPG in empirical tests, while reducing computational cost. This is more pronounced with larger numbers of agents.