Characterizing and Predicting Submovements during Human Three-Dimensional Arm Reaches

Abstract

We have demonstrated that 3D target-oriented human arm reaches can be represented as linear combinations of discrete submovements, where the submovements are a set of minimum-jerk basis functions for the reaches. We have also demonstrated the ability of deterministic feed-forward Artificial Neural Networks (ANNs) to predict the parameters of the submovements. ANNs were trained using kinematic data obtained experimentally from five human participants making target-directed movements that were decomposed offline into minimum-jerk submovements using an optimization algorithm. Under cross-validation, the ANNs were able to accurately predict the parameters (initiation-time, amplitude, and duration) of the individual submovements. We also demonstrated that the ANNs can together form a closed-loop model of human reaching capable of predicting 3D trajectories with VAF >95.9% and RMSE ≤4.32 cm relative to the actual recorded trajectories. This closed-loop model is a step towards a practical arm trajectory generator based on submovements, and should be useful for the development of future arm prosthetic devices that are controlled by brain computer interfaces or other user interfaces.

Signal-independent noise in intracortical brain–computer interfaces causes movement time properties inconsistent with Fitts’ law

“…In figure 9A we illustrate the size of the decoding noise for T6 and T8 compared to that of the able-bodied motor system in various motor tasks (hand force impulses, wrist rotations, stylus movements, and arm movements) (Schmidt et al 1979; Meyer et al 1988; Young, Pratt, and Chau 2009; Liao and Kirsch 2014). We defined SNR as the size of the signal divided by the standard deviation of the noise.…”

Velocity neurons improve performance more than goal or position neurons do in a simulated closed-loop BCI arm-reaching task

“…The three main components were the Multiple Submovement Controller (MSC), the Neural Encoder, and the Decoder. The MSC (Liao and Kirsch, 2014 ) is a submovement-based model of error corrections during human movements that provided movement commands in the form of position, velocity, and goal, as a function of the target position, start position, and the decoded position, velocity, acceleration, and goals that were fed back to the MSC from the Decoder. The Neural Encoder generated firing rates based on the movement commands computed by the MSC, the modulation properties of cortical neurons as extracted from previous studies, and spiking noise (see below).…”

“…The MSC (Liao and Kirsch, 2014 ) generates realistic reaching trajectories and consists of three Artificial Neural Networks (ANNs) that were pre-trained using experimentally recorded human reaching data to generate a command trajectory by linearly summing a discrete number of minimum-jerk submovements. This controller is based on a theory of human movement that represents movements as a set of overlapping-in-time submovements, each representing an error correction to the overall trajectory made in order to sustain progression to the target.…”

“…Three separate ANNs predicted these parameters as functions of kinematic features of the movement. For instance, based on the start position, target position, decoded position, decoded velocity, decoded acceleration, and the predicted position at the end of the current submovements, an ANN predicted the next amplitude d i to generate a submovement to bring the decoded position toward the target (see Supplementary Materials and (Liao and Kirsch, 2014 ) for a more detailed description).…”

“…The purpose of the current study was to provide some insight into this issue by developing a simulation of a closed-loop BCI for controlling arm movements. This simulation included a submovement-based movement controller (Liao and Kirsch, 2014 ) that was trained on experimental movements, an encoder of neural firing rate modulation and noise properties based on literature reports, and an Extended Kalman Filter decoder that had the same structure as the encoded neural properties (i.e., it was assumed that the ideal decoding structure was known a priori ). We then used this simulator to predict the relative contributions of various combinations of neurons modulated by position, velocity, and movement goal to overall BCI performance, as well as the impact of the number of neurons of each type on performance.…”