Self-adaptive shared control with brain state evaluation network for human-wheelchair cooperation.

Area of Science:

• Neuroscience
• Robotics
• Machine Learning

Background:

• Shared control systems require balancing robot autonomy and human operator input, particularly in brain-computer interface (BCI) applications.
• Existing BCI shared controllers often overlook individual differences in users' brain control capabilities.

Purpose of the Study:

• To propose a novel brain state evaluation network (BSE-NET) for online assessment of users' brain control ability in BCI systems.
• To develop a dynamic shared controller that adjusts control allocation based on evaluated user ability.

Main Methods:

• Developed BSE-NET using quantized attention-gated kernel reinforcement learning to assess brain control ability online.
• Integrated BSE-NET output (confidence score) into a shared controller for adaptive control weight adjustment.
• Validated the system's performance with human subjects exhibiting varying EEG decoding accuracy.

Main Results:

• Subjects achieved a high and stable experimental success rate of approximately 90%.
• BSE-NET generated accurate confidence scores reflecting individual brain control abilities.
• The adaptive shared control system effectively balanced control weights in real-time.

Conclusions:

• The proposed BSE-NET enables online evaluation of brain control ability in BCI users.
• This method facilitates self-adaptive shared control, optimizing the balance between user commands and robot autonomy.
• The system shows significant promise for enhancing BCI applications requiring adaptive human-robot interaction.