Using Physics- and Sensor-based Simulation for High-Fidelity Temporal Projection of Realistic Robot Behavior

Abstract

Planning means deciding on the future course of action based on predictions of what will happen when an activity is carried out in one way or the other. As we apply action planning to autonomous, sensor-guided mobile robots with manipulators or even to humanoid robots we need very realistic and detailed predictions of the behavior generated by a plan in order to improve the robot's performance substantially.

In this paper we investigate the high-fidelity temporal projection of realistic robot behavior based on physics- and sensor-based simulation systems. We equip a simulator and interpreter with means to log simulated plan executions into a database. A logic-based query and inference mechanism then retrieves and reconstructs the necessary information from the database and translates the information into a first-order representation of robot plans and the behavior they generate.  The query language enables the robot planning system to infer the intentions, the beliefs, and the world state at any projected time.  It also allows the planning system to recognize, diagnose, and analyze various plan failures typical for performing everyday manipulation tasks.