A CNN-based chip for robot locomotion control

Abstract

In this paper a VLSI chip for real-time locomotion control in legged robots is introduced. The control is based on the biological paradigm of Central Pattern Generator (CPG) and is implemented by a Cellular Neural Network (CNN). The gait generation is accomplished by the CNN and is fully analog, while a digital controller modulates the behavior of the CNN-based CPG to allow the locomotion system to adapt to sensory feedback. The chip is designed with a switched-capacitor technique, fundamental to address the speed control issue. Experimental results on the first prototype are illustrated. These results confirm the suitability of the approach and open the way to the design of a fully autonomous bio-inspired micro-robot.