Task planning and the Connect-R: explainable AI for real world multi-robot system deployment

Abstract

The Connect-R is a novel Multi-Robot System that is intended to provide maintenance and servicing capabilities in nuclear environments. The Connect-R system is intended to address the problem of mitigating the effects of radioactive environments by acting as a preliminary robotic deployment that, through its deployment, maximise the useful time in the environment for secondary service robots. The stated mission goal of the Connect-R MRS is to provide structure in unstructured environments, which in reality provides a physical structure within the nuclear environment upon which service robots can traverse the nuclear environment, effectively increasing the efficiency with which they can spend their time, with respect to completing the mission. To realise the underlying ethos of the Connect-R, this work presents a total design of the robotic design whilst discussing the development of the AI system that allows for human-in-the-loop architecture and the tools to support human-system interaction. This thesis presents the principle ethos behind the Connect-R approach in Chapter 3, the Connect-R robotic design in Chapter 4, the AI system in Chapter 5 and the software developed to operate the Connect-R system in Chapter 6. The major contributions of the thesis are the definition and justification of the core approach of the Connect-R system, key design choices to facilitate robot development, the human-intuitive domain representation of the AI system and optimisation measures taken to improve accuracy, scalability and efficiency and also the development of a bespoke software tool that facilitates human-system interaction. This thesis addresses the key challenges in utilising classical AI techniques for full robotic deployment in 3D environments such that the AI system solutions are scalable, human-intuitive and consistently accurate.