An offline approach to fNIRS-guided reinforcement learning for robot behavior

arXiv cs.AI · 6 hours ago

Researchers developed a method to train robots using reinforcement learning guided by brain signals from functional near-infrared spectroscopy, comparing passive observation and active demonstration approaches. The system improved learning by augmenting trajectory priorities and Q-values with neural signals, and achieved successful results working with offline data rather than requiring real-time brain-computer interface setups. This approach enables robot behavior training in scenarios where continuous brain signal monitoring is impractical.