Modeling of kinematic redundancy by incremental successive approximations

This paper presents a modification of the successive approximation method so the model can be updated with an each new learning example, without need to keep them all from the beginning. The incremental update of the model is achieved by recursive least mean square algorithm. Incremental successive approximation method enables optimization of the number of learning examples, duration of learning process, as well as adaptation of the model to varying parameters of both the robot itself and the given task. The efficiency of suggested method is shown on the example of multilegged locomotion robot.