On the applicability of brain reading for predictive human-machine interfaces in robotics

Area of Science:

• Neuroscience and Human-Machine Interaction
• Robotics and Artificial Intelligence

Background:

• Current robots have limited autonomous capabilities in supporting daily human activities.
• Predictive human-machine interfaces (HMIs) are crucial for enhancing human-robot interaction.
• Detecting human brain states via brain reading (BR) is key for inferring context-based behavior.

Purpose of the Study:

• To demonstrate that brain reading (BR) can detect concrete human states.
• To improve the robustness and applicability of BR for real-world scenarios.
• To showcase a dual BR application for simultaneous detection of cognitive processes.

Main Methods:

• Utilized supervised machine learning (ML) for single-trial electroencephalogram (EEG) analysis.
• Identified and combined relevant training data for robust single-trial classification.
• Applied classifier transfer techniques to adapt BR models with limited training data.
• Demonstrated a dual BR system for simultaneous detection of target recognition and movement preparation.

Main Results:

• BR successfully detected patterns in EEG, including event-related potentials like P300, indicative of specific brain processes.
• Improved BR robustness by optimizing training data selection and employing classifier transfer.
• Showcased successful classifier transfer even when training and testing classes lacked specific patterns.
• Validated a dual BR application detecting simultaneous cognitive processes during a robotic arm teleoperation task.

Conclusions:

• Brain reading (BR) is capable of detecting concrete human brain states relevant for predictive HMIs.
• Optimized BR methods enhance robustness and applicability in scenarios with limited training data.
• The developed dual BR system enables simultaneous monitoring of distinct cognitive processes, advancing predictive HMIs for complex tasks.