Value Functions Factorization with Latent State Information Sharing in Decentralized Multi-Agent Policy Gradients

The game replay is for demonstration purposes for readers to understand the SMAC environment and policy strategies.

Algorithms trained agents control the units on the left side, built-in AI controls the units on the right side with the hardest difficulty (7-Insane).

All combat videos are for a win/lose visualization of the learned policy and do not reflect actual win rate/performance.

After a finished round the environments reset by killing all units left alive and respawn the units as required by the map. This is achieved by using in-game local-player cheats.

The win rate is calculated with 32 independent evaluation runs. For more details, please refer to our paper and SMAC environment.

View in fullscreen with high resolution for better visibility.