Exploring Sensory-Motor Control Through Virtual Object Manipulation
Background
Brain-computer interfaces (BCIs) aim to restore motor control to individuals with motor impairments by using recordings of brain activity to generate an action.
Most BCIs trained on simple, highly reproducible tasks such as the Center-out Task.
Natural movements are often continuous and complex. For example, a bicycle requires continuous sensory feedback for balance.
Center-out Task: subject moves hand in direction cursor on screen moves.
Introduction
The “Critical Stability Task” (CST) probes sensory-driven motor control through virtual object manipulation.
[Quick et al. 2018]
Developed by the Batista Lab, the CST requires continuous sensory feedback to complete the task
The CST is an inherently unstable system akin to balancing an upside down broom in the palm of one's hand.
Critical Stability Task (CST): The task begins with the cursor in the center of the screen, as the task starts, the cursor will drift toward the left or right edge of the screen. The user can imagine tying a string around the cursor and "pulling" the cursor back to the center of the screen.
Exploratory Study
We propose that developing modifications to the CST will provide further insight and provide another tool in the toolbox for learning the language of the brain
Modifying the underlying equations of the CST in theory will change how the system responds (i.e. how fast/slow the onscreen cursor moves) therefore requiring the user to adjust their strategy and learn how to perform the modified CST.
The above figure shows how the subject moves their hand in the opposite direction of the cursor and imagines "pulling" the cursor back to center
Analysis
After modifying the existing CST paradigm, it was desired to quantify any differences between the original and modified CST.
To quantify a difference we trained a neural network to perform the original CST in place of a live subject. If the neural network could successfully perform the original CST but not the modified CST then that is an indication of a fundamental difference between the two paradigms.
The above figure shows how the trained neural network would be performing CST in place of a live subject.
Analysis: Simulations
When a live subject performs the CST, the subject has spatial awareness of where their hand is, where the cursor is on screen, as well as how fast the cursor is moving. The subject then decides where to move their hand based on the sensory information.
We created a neural network to be given the current cursor position and velocity as well as the simulated "hand" position of the neural network and asked it to generate the next hand position.
When training the neural network, we optimized the input information (i.e. for 05. Cpva Hp the network was given the Cursor Position (Cp), Cursor Velocity (Cv), and Cursor Acceleration (Ca) as well as the Hand Position (Hp) and asked to generate the next hand position required to control the onscreen cursor and complete the task. The darker the color on the graph, the better the neural network performed.
Simulation Results
The neural network trained to perform the original CST was able to successfully complete the original CST, but unsuccessful when attempting to perform the modified CST. This result provides a first step in quantifying the differences present in the modified CST
The neural network trained to perform the original CST was successfully able to complete the original CST trials in place of a live subject.
The neural network trained to perform the original CST was unable to perform the modified CST in place of a live subject. The red line reaching the edge of the figure indicates the cursor drifted off the screen and the CST trial was a failure.