Control-theoretic Synthesis of Hierarchical Dynamics for Embodied Cognition in Autonomous Robots
Nebu John Mathai, Takis Zourntos · 2007
We consider the design of a continuous hierarchical dynamic system architecture to endow cognition in an autonomous mobile robot with bounded resources. The hierarchy is decomposed on the basis of behavior with lower levels being concerned with motion control, and upper levels being concerned with ensuring overall progress. Our architecture consists of a composition of regulators, all of which have access to sensor data and which control problems relating to the agent's motion in the world (i.e., its velocity, position, and path); it hence realizes a physically embodied cognitive system. To synthesize the regulators, we employ a plant-controller co-design methodology; after constructing a plant that models the salient features of the environment relevant to the behavioral level under consideration, we apply rigorous nonlinear control-theoretic toolsets to synthesize a corresponding controller. An interesting side-effect of our synthesis is the emergence of normalized and unnormalized radial basis functions in our control law. Simulation results and animations illustrating the life-like behavior of the system are provided. We highlight the emergence of satisficing intelligent behavior by our cognitive architecture and show how this satisfies Bedau's definition of weak emergence