Representation Convergence:
Mutual Distillation is Secretly a Form of Regularization

bigfish with MDPO (ours)

bigfish with PPO

coinrun with MDPO (ours)

coinrun with PPO

fruitbot with MDPO (ours)

fruitbot with PPO

Policy Robustness Testing
(brightness, contrast, saturation, and hue)

Policy Robustness Testing
(random CNNs)