Profiles of movement speed and positional variability based on extended LQG for various noises

Stochastic optimal control has been studied to explain the characteristics of human upper-arm reaching movements. The optimal movement based on an extended linear quadratic Gaussian (LQG) demonstrated that control-dependent noise is the essential factor of the speed-accuracy trade-off in the point-to-point reaching movement. Furthermore, the extended LQG reproduced the profiles of movement speed and positional variability. However, the expected value and variance were computed based on the Monte Carlo method in these studies, which is not considered efficient. In this study, I obtained update equations to efficiently compute the expected value and variance based on the extended LQG. Using the update equations, I computed the profiles of simulated movement speed and positional variability for various amplitudes of noises in a point-to-point reaching movement. The profiles of movement speed were basically bell-shaped for the noises. The speed peak was changed by the control-dependent noise and state-dependent observation noise. The positional variability changed for various noises, and the period during which the variability changed differed with the noise type. Efficient computation in stochastic optimal control based on the extended LQG would contribute to the elucidation of motor control under various noises.