Sensor-driven neural controller for self-adaptive collision-free behavior of a snake-like robot

Abstract

Biologically inspired control approaches based on the central pattern generator (CPG) have been studied to apply to a snake-like robot. One of the important problems is to determine how to construct a sensor-driven neural system in order to control the robot for adaptive locomotion. To solve this problem, a sensor-based neural network is presented in this paper. To realize collision-free behavior of the snake-like robot, three IR range sensors were used to obtain the obstacle information. By analyzing the motion strategies for the snake-like robot, a signal feedback network was constructed based on the neuron model. The sensory signals were used as the adjusted values for the input of CPG oscillators. By changing the driving input of the extensor neurons or flexor neurons in the CPG network, the snake-like robot could perform the desired turning motion to avoid the obstacles. The performance of the proposed sensor-driven neural controller was verified by conducting an experiment on a snake robot in an environment with obstacles.