Quadruped Gait Transitions from a Neural Pattern Generator with Arousal Modulated Interactions

Center for Adaptive Systems and Department of Cognitive and Neural Systems Boston University, 111 Cummington Street, Boston, MA 02215 A four-channel neural pattern generator is described in which both the frequency and the relative phase of oscillations are controlled by a scalar arousal or GO signal. The generator is used to simulate quadruped gaits; in particular, rapid transitions are simulated in the order walk, trot, pace, and gallop -that occurs in the cat. Precise switching control is achieved by using an arousal dependent modulation of the model's inhibitory interactions. This modulation generates a different functional connectivity in a single network at different arousal levels. 1 A Neural Pattern Generator This article describes a neural pattern generator that is capable of triggering the primary gaits and gait transitions that are observed in quadrupeds as emergent properties of network dynamics. Phase transitions from one gait to the next are triggered by a descending scalar control signal, called the GO signal, that realizes a form of network arousal. An analysis of quadruped gait transitions leads to propose that the GO signal modulates the functional connectivity of the controlling neural pattern generator. Task-specific modulation of the functional connectivity of neural pattern generators has been experimentally observed; for example, in the stomatogastric ganglion of the crab [a]. The present model predicts a. pattern of arousal-dependent modulation that permits only the naturally occurring quadruped galt transitions to be generated by a single model circuit, in some parameter ranges, as its GO signal is parametrically increased. We therefore call such a model a GO Gait Generator, or 0 3 model. The 0 3 model an extension of the Ellias·Grossberg nonlinear cooperativecompetitive feedback network (2]. This model extends to the quadruped case earlier modeling of these generators tha.t was summarized in Cohen, Grossberg and Pribe (1 J. The 0 3 model is defined by the equations where i; =-Ax;+ (Bx;)(f(x;) + Ii](C + x;)[2::;D;jh;j(I)g(yj)] y; = B[(l y;)[xi]+ yi], (w]+ = max(w, 0), •Supported in part by AFOSR (AFOSR 90-0128 <end AFOSR 1'49620-92-J-0225). (1)