Hybrid Brain-Computer-Interfacing for Human-Compliant Robots: Inferring Continuous Subjective Ratings With Deep Regression.

摘要: Appropriate robot behavior during human-robot interaction is a key part in the development of human-compliant assistive robotic systems. This study poses question how to continuously evaluate quality hybrid brain-computer interfacing (BCI) task, combining brain and non-brain signals, use collected information adapt robot's accordingly. To this aim, we developed rating system compatible with EEG recordings, requiring users execute only small movements their thumb on wireless controller rate continuous scale. The ratings were recorded together dry EEG, respiration, ECG, joint angles ROS. Pilot experiments conducted three that had different levels previous experience robots. results demonstrate feasibility obtain data give insight into subjective user perception direct interaction. suggests differences for no, moderate, or substantial experience. Furthermore, variety regression techniques, including deep CNNs, allowed us predict ratings. Performance was better when using position hand than respiration. A consistent advantage features expected be related motor bias could not found. Across-user predictions showed models most likely learned combination general individual across-users. transfer pre-trained regressor new especially accurate more For future research, studies participants will needed methodology its practice. Data code reproduce are available at https://github.com/TNTLFreiburg/NiceBot.