Human and animal behavior exhibits an immense behavioral flexibility, which cannot be achieved by state-of-the-art robotic systems.
Particularly, redundant behavioral alternatives are flexibly and efficiently employed to satisfy current goals considering bodily and environmental circumstances. Neuroscientific studies show that this flexibility is realized by the means of modular, sensorimotor bodyspace representations. In the COBOSLAB (COgnitive BOdySpaces: Learning And
Behavior) laboratory we study how such bodyspaces can be learned and adapted in ways maximally suitable for the realization of flexible, goal-directed behavioral control.
Studies in robot arm control with redundant degrees of freedom show that distributed population codes can yield highly effective, flexibly adjustable, kinematic behavior. Also dynamic plants can be controlled when coupling the resulting kinematic control routines with adaptive PD mechanisms. Similar principles apply for autonomous robot vehicle control. Finally, we show how motivations can be easily included, so that the resulting system is able to self-activate the currently most relevant goal-directed behavior in a self-motivated manner.