Probabilistic, Prediction-Based Schedule Debugging for Autonomous Robot Office Couriers

Abstract. Acting efficiently and meeting deadlines requires autonomous robots to schedule their activities. It also requires them to act flexibly: to exploit opportunities and avoid problems as they occur. Scheduling activities to meet these requirements is an important research problem in its own right. In addition, it provides us with a problem domain where modern symbolic AI planning techniques could considerably improve the robots’ behavior. This paper describes PPSD, a novel planning technique that enables autonomous robots to impose order constraints on concurrent percept-driven plans to increase the plans’ efficiency. The basic idea is to generate a schedule under simplified conditions and then to iteratively detect, diagnose, and eliminate behavior flaws caused by the schedule based on a small number of randomly sampled symbolic execution scenarios. The paper discusses the integration of PPSD into the controller of an autonomous robot office courier and gives an example of its use.