Robotic plan execution in dynamic and unpredictable environments

Abstract

To be useful in the real world, robots need to be able to move safely in unstructured environments and achieve their given tasks despite unexpected environmental changes or failures of some of their sensors. The variability of the world makes it impractical to develop very detailed plans of actions prior to execution since the world might change before execution begins, and thus invalidate the plan. The transformation from the high level description of the task to the primitive actions should be performed primarily at execution time. This allows access to up-to-date information about the environment through sensors as well as integration of sensed information in deciding how to achieve the task. The Logical Sensor/Actuator (LSA) theory was developed to specify the types of knowledge and processes necessary for a mobile robot to execute a plan. The LSA theory is based on three premises: (1) plan execution requires relevant details (hidden in the plan's abstraction) to be made explicit; (2) knowledge is integral to the plan execution process, and thus proper application of knowledge increases the robustness of plan execution; and (3) plan execution is an information-based process where determining which information is relevant constitutes a great deal of the process. This research also addresses the issue of what knowledge needs to be available about sensors, actuators and processes in order to be able to integrate their usage, and control them during execution. The methods proposed are applicable to any sensor/actuator existing on the robot when given such knowledge. This research yielded an object-oriented homogeneous architecture for plan execution, called the Logical Sensor/Actuator Testbed (LSAT). LSAT was used to develop laboratory experiments on a real mobile robot, providing data in support of the theory. A large number of laboratory experiments were conducted on the implementation. The experiments covered various capabilities of the implementation across a number of environmental scenarios. The results not only support the theory, but also illustrate how the proper application of knowledge in the integration and utilization of sensors and actuators increases the robustness of plan execution.