Neural implementation of behavior control

Abstract

The dynamical systems approach and recurrent neural control provides a rich foundation for the generation of natural behaviors on autonomous robots because the environment, the robot, and control systems are regarded as a single dynamical system. Robot behaviors can thus be shaped as attractors of this dynamical system. Within this framework, sensorimotor loops for walking and keeping balance have been realized on the Myon robot. Different behaviors can be shaped as co-existing attractors which allows for smooth and reliable switching between them. We introduce the concept of Cognitive Sensorimotor Loops (CSLs) as well as the use of quadrics and discuss their benefits for behavior control. The presentation of every technique is accompanied by a real world example using humanoid robots. Finally, a grasping motion is developed using the same methods.