In this paper we outline a framework that would enable us to program complex dynamic sensorimotor behavior in such characters. Since the emphasis is on dynamic behaviors, it is necessary to deal with and reason about dynamic variables such as velocities and accelerations; hence a dynamical systems approach is adopted. Our approach is to demonstrate that apparently complex patterns of dynamic coordination (such as obstacle avoidance, pole-balancing, jumping and hopping) can be formulated using a dynamic constraint-generation and constraint satisfaction framework. The lowest level motor controller behavior is altered by constraints generated by the supervisory layer on top. The solution to the resulting constrained problem, which is accomplished in real-time, automatically generates a desired behavior pattern. A central theme of the proposed research is the dynamic generation of sub-goals and constraints to satisfy each individual behavior.