Intelligent ballistic control of a jointed leg

Abstract

In order to perform a wide range of locomotive tasks, a legged robot should be able to autonomously adjust its stride and properly move its legs so as to fit the current locomotion requirement. We develop an adaptive intelligent system that controls the movement of a jointed leg during a running stride. This paper focuses on the ballistic controller, which controls the movement of the leg from the point of takeoff from the ground to the point of landing. The controller learns from experience to move the leg along a specified series of configurations which provide a smooth transition from takeoff to landing. Learning is online, so the controller adapts to the requirements of the stride on the fly. It can handle strides of different sizes and can scale obstacles. Very accurate control is achieved even during the first learning attempt.