Offline NEURO-LOOP
Brain-computer interfaces for human-in-the-loop RL
NEURO-LOOP is a reinforcement learning framework that integrates implicit neural feedback into robotic and artificial agent policy learning. It pulls from previous work by usinga person’s brain signals and an agent’s behavior without requiring explicit guidance. Offline NEURO-LOOP uses fNIRS data from a dataset instead of a real-time loop.
Overview
Traditional interactive RL often depends on overt feedback such as ratings, demonstrations, or button presses. Instead, NEURO-LOOP treats passive neural activity as a supervisory signal. This enables agents to adapt its behaviorto a user’s expectations as they naturally observe the agent’s behavior.
The offline NEURO-LOOP framework uses fNIRS data from a dataset, exploring how to interpret neural state and how to update the policy to align with human intent outside of a real-time loop.
Focus
This work focuses on how to reliably use fNIRS signals to update the policy of a robotic or artificial agent by using offline data from a dataset.