Offline Learning

Training Safety-Critical Reinforcement Learning Agents Offline

Conservative Q-Learning achieves a 25% higher return mean than Behavior Cloning in safety-critical environments.