Identifying feedback control strategies of running cockroaches and humans.

摘要

In this dissertation, we model biological sensorimotor behaviors of two species, a cockroach following a wall and a human running on a split-belt treadmill, to elucidate the neural processing that underlie locomotor control in biological systems: (1) We model the horizontal musculoskeletal dynamics of antenna-based wall following for the American cockroach, Periplaneta americana, as a dynamic planar unicycle with an idealized antenna. Performing nonlinear regression on the transient responses of blinded cockroaches running along various wall perturbations, we show that the stabilizing neural feedback requires not only the distance-to-wall information but also the rate of approach to the wall. We corroborate this result using a robotic platform equipped with an artificial antenna, a numerical simulation of antenna-based lateral leg spring (LLS) model, and a comparison with a neurophysiological experiment. (2) For human running, we model the sagittal-plane feedback control strategies during early and late adaptation phases of split-belt treadmill running. For the early adaptation phase, we assume spring-loaded inverted pendulum (SLIP) body mechanics with compositions of one-step deadbeat feedback controllers; we show that the compositions of slow-belt feedback controllers best represented the steady-state human running data. We compare the eigenvalues of the linearized stride-to-stride return map during late adaptation with those during baseline tied-belt running. Our result suggests larger eigenvalues (i.e. slower recovery rate) during late adaptation, suggesting that adapted split-belt is not simply the dynamic composition of a fast steps and slow steps.