Tag
This work proposes using privileged information to actively sample rollouts in reinforcement learning, improving on typical blind sampling methods.