dr-mas

Enable stable multi-agent reinforcement learning by normalizing advantages per-agent rather than globally, preventing gradient-norm inflation in heterogeneous multi-agent systems.

Independent GitHub snapshot · Automated classification · Original documentation governs