Thesis title: Multi-Robot Trajectory OptimisationSupervisors: Steven SCHEDING
Mining haul truck technology has improved over the years by increasing load capacities, improving reliability, and using computer-based dispatching systems. Autonomous trucks raise the opportunity to achieve optimal behaviour based on additional criteria such as energy. Typical dispatching systems control trucks using heuristics that are based on criteria such as cycle times, but are myopic and unproven to be optimal or give reproducible results. Alternative solutions may exist that are more efficient, which could be achieved by considering the paths and trajectories of the vehicles.
The research proposed here can be broken down into several separate questions. The first is how to incorporate criteria such as energy to a multi-robot path and trajectory optimisation, in the context of a system that aims to maximise production. In a multi-vehicle system, the actions and performance of one vehicle are coupled to that of the others. The second question is how to account for interdependencies in the optimisation model. The third is how to account for the dynamic nature of the environment, and adapt to unforeseen disruptions during operation, such as when equipment fails, or when unscheduled events occur. The last question is how to manage the uncertainty that is present in the estimate of the system’s state, which increases with the prediction horizon. Accounting for uncertainty will mean improving the ability to analyse and make decisions.
The first focus is on individual vehicle trajectory optimisation, followed by an extension to multi-vehicle cases, by combine individual trajectories using prioritisation and decoupling methods. Three broad methods have been identified as relevant to solving interaction points and addressing dynamic problems, including sampling algorithms, metaheuristics, and black-box global optimisation. Techniques such as FMT* and GAs show particular promise as they are able to handle large and complex search spaces.