Multi-Agent Deep RL

COMA (Counterfactual Multi-Agent Policy Gradients)

Algorithm that uses counterfactual baselines to estimate how individual actions affect the global reward by changing one agent's policy while keeping others fixed.

← Kembali