Random Network Distillation

Random Network Distillation (RND) is an intrinsic motivation method for reinforcement learning used to estimate state novelty. It measures how frequently an agent has visited a given state via prediction error and is commonly employed in exploration-driven reinforcement learning algorithms.