AN ANALOG CMOS CHIP IMPLEMENTING A CNN-BASED LOCOMOTION CONTROLLER FOR QUADRUPED WALKING ROBOTS

摘要

We propose an analog CMOS circuit that implements a class of cellular neural networks (CNNs) for locomotion control in robotics. Our circuit is constructed using multiple-input floating-gate MOS (FGMOS) FETs aiming at the voltage-mode operation, and it can be expected to reduce power consumption. Furthermore, we fabricated a prototype chip using a standard CMOS process. From experimental results, we have confirmed that the proposed circuit has capability of generating stable rhythmic patterns for robot locomotion control under an actual environment.