Multi-Agent Reinforcement Learning

We introduce Continuous Training with Discrete Execution (CTDE2), a paradigm for continuous multi-agent reinforcement learning that learns compact discrete action codebooks for efficient execution.