An activation-based behavior control architecture for walking machines

The high complexity of the mechanical system and the challenging task of walking itself makes the task of designing the control for legged robots a difficult one. Even if the implementation of parts of the desired functionality, such as Posture control or basic swing/stance movement, can be solved by the use of classical engineering approaches, the control of the overall system tends to be very inflexible. In this paper we introduce a new method to combine aspects of classical robot control and behavior-based control. Inspired by the activation patterns in the brain and the spinal cord of animals, we propose a behavior network architecture using special signals such as activity or target rating to influence and coordinate the behaviors. We, describe the general concept of a single behavior as well as their interaction within the network. This architecture is tested on the four-legged walking machine, BISAM, and experimental results are presented.