This website complements the paper "When Is Diversity Rewarded in Cooperative Multi-Agent Learning?" by presenting multimedia results from the experiments analyzed in the work.
In Multi-Goal-Capture, we show that when the heterogeneity gain is positive, only heterogeneous agents can go to different goals.
Heterogeneous
Homogeneous
By increasing observability, parameter-shared agent policies can learn the optimal policy.
Parameter-sharing agents with partner sensing
In Tag, we show that our theoretical results apply to discrete and sparse episode rewards.
Heterogeneous
Homogeneous
In Football, we show that our theory extends and applies to more complex real-world tasks.
Heterogeneous
Homogeneous