Non-linear dynamical system approach to behavior modeling

We present a dynamic systems approach to modeling and generating low-level behaviors for autonomous agents. Such behaviors include real-time target tracking and obstacle avoidance in time-varying environments. The novelty of the method lies on the integration of distinct non-linear dynamic systems to model the agent's interaction with the environment. An angular velocity control dynamic system guides the agent's direction angle, while another dynamic system selects the environmental input that will be used in the control system. The agent interacts with the environment through its knowledge of the position of stationary and moving objects. In our system agents automatically avoid stationary and moving obstacles to reach the desired target(s). This approach allows us to prove the stability conditions that result in a principled methodology for the computation of the system's dynamic parameters. We present a variety of real-time simulations that illustrate the power of our approach.