Velocity Control of a Hybrid Quadruped Bounding Robot

This paper addresses the issue of implementing an intelligent velocity controller on the Platform for Ambulating Wheels (PAW). The PAW robot is a hybrid quadrupedal wheeled-legged robot that can bound, gallop, roll and brake at high speeds and perform inclined turning. The goal of implementing intelligent control is to increase the robot's versatility and autonomy in order to traverse various terrain types and complete tasks. A Levenberg-Marquardt learning algorithm is executed at the top of flight instant in a stride and computes the forward and rear foot touchdown and liftoff positions. This enables the robot to track desired velocity in a Madab-Adams co-simulation model. Initial steps are also taken to implement this learning algorithm on the physical robot by way of developing an Extended Kalman Filter (EKF) to estimate the forward center of mass velocity. Additionally, a discussion on the future steps towards autonomous control of the PAW robot is presented.