This position paper for the Fall Symposium on the Control of the Physical World by Intelligent Agents discusses the need to consider the entire set of desired robot behaviors when designing a software architecture and knowledge representation. Integrating multiple representations and action mechanisms is necessary in an real robot and constraints on that integration come from a variety of areas including: control, planning, and natural language understanding. This paper uses robot navigation as an example to speculate on integrating vision, sonars, dynamic obstacle avoidance, and natural language understanding into the Animate Agent Architecture. The discussion brings up the need for a symbolic representation for obstacles that can directly drive the robot’s motors to react quickly to moving objects. Reconciling these demands leads to a variety of feedback control loops that can be influenced by symbolic planners in different ways.