Integrating ASP-based Planning and Diagnosis with POMDPs for Knowledge Representation and Reasoning on Mobile Robots
Mobile robots operating in real-world domains frequently encounter challenges due to an uncertain and dynamic environment. In order for a robot to successfully accomplish a given task, it must not only generate an effective plan, but also deal with unforeseen changes in the working environment and action outcomes that may be non-deterministic. This project adds navigational planning and diagnostic capabilities to an existing architecture that integrates high-level logical inference with low-level probabilistic decision making. Answer Set Programming (ASP), a non-monotonic logic programming paradigm, is used to represent and reason about domain knowledge, while Partially Observable Markov Decision Processes (POMDPs) are used to probabilistically model the uncertainty in sensing and acting on robots. The modified architecture enables robots to represent and reason with incomplete domain knowledge, adapting sensing and acting to the tasks at hand, and revising existing knowledge based on information extracted from sensors and humans. This architecture is evaluated in simulation and implemented on a wheeled robot in an indoor domain.
This project also investigates the design and use of a mobile robot (and the architecture described above) in the high-throughput phenotyping domain. To support precise navigation of the robot and the measurement of characteristics of individual plants in the field, sensors such as an RTK GPS and LIDAR are explored.