Combining Brain-Computer Interfaces with Deep Reinforcement Learning for Robot Training: A Feasibility Study in a Simulation Environment

The study tested whether a robot in a 3D simulation could learn from people’s brain responses to its mistakes. The researchers compared EEG recording setups and classifiers, then used detected error responses as feedback for a deep reinforcement learning agent. With a suitable dry EEG system, training from implicit feedback accelerated learning and produced performance comparable to training with explicit feedback.

Frontiers in Neuroergonomics, 4, 1274730