Intelligent control of quadruped gallops

In this paper, a new intelligent control approach for high-speed quadruped bounding and galloping gaits is presented. The controller is capable of learning the leg touchdown angles and leg thrusts required to track the desired running height and velocity of a quadruped in only one stride. Training of the controller is accomplished not with a mathematical model, but with simple rules based on a heuristic knowledge of the quadruped mechanics. The result is a controller that produces better velocity and height tracking characteristics than a Raibert-based controller and is robust to modeling errors. Additionally, by making use of the natural dynamics of the system, gait characteristics comparable to biological quadrupeds result. The status of a legged machine being constructed for demonstration of the control approach and further study of the characteristics of galloping is also presented.