Batch Constrained Q-learning (BCQ)

Offline Reinforcement Learning

Learning paradigm where the agent learns exclusively from a fixed set of previously collected data, without interaction with the environment. This approach is essential when real-time exploration is costly or dangerous.

← Back