A Scalable Intelligent Takeoff Controller for a Simulated Running Jointed Leg

Running with jointed legs poses a difficult control problem in robotics. Neural controllers are attractive because they allow the robot to adapt to changing environmental conditions. However, scalability is an issue with many neural controllers. This paper describes the development of a scalable neurofuzzy controller for the takeoff phase of the running stride. Scalability is achieved by selecting a controller whose size does not grow with the dimensionality of the problem. Empirical results show that with proper design the takeoff controller scales from a leg with a single movable link to one with three movable links without a corresponding growth in size and without a loss of accuracy.