Logic Programming for Robot Control

This paper proposes logic programs as a specification for robot control. These provide a formal specification of what an agent should do depending on what it senses, and its previous sensory inputs and actions. We show how to axiomatise reactive agents, events as an interface between continuous and discrete time, and persistence, as well as axiomatising integration and differentiation over time (in terms of the limit of sums and differences). This specification need not be evaluated as a Prolog program; we use can the fact that it will be evaluated in time to get a more efficient agent. We give a detailed example of a nonholonomic maze travelling robot, where we use the same language to model both the agent and the environment. One of the main motivations for this work is that there is a clean interface between the logic programs here and the model of uncertainty embedded in probabilistic Horn abduction. This is one step towards building a decisiontheoretic planning system where the output of the planner is a plan suitable for actually controlling a robot.