Date of Award
Thesis - Open Access
Master of Science in Mechanical Engineering
Dr. Eric Coyle, Ph.D.
First Committee Member
Dr. Patrick Currier, Ph.D.
Second Committee Member
Dr. M. Ilhan Akbas, Ph.D.
The sector of maritime robotics has seen a boom in operations in areas such as surveying and mapping, clean-up, inspections, search and rescue, law enforcement, and national defense. As this sector has continued to grow, there has been an increased need for single unmanned systems to be able to undertake more complex and greater numbers of tasks. As the maritime domain can be particularly difficult for autonomous vehicles to operate in due to the partially defined nature of the environment, it is crucial that a method exists which is capable of dynamically accomplishing tasks within this operational domain. By considering the task allocation problem as a graph search problem, Minion Task, is not only capable of finding and executing tasks, but is also capable of optimizing costs across a range of parameters and of considering constraints on the order that tasks may be completed in. Minion task consists of four key phases that allow it to accomplish dynamic tasking in partially defined environments. These phases are a search space updater that is capable of evaluating the regions the vehicle has effectively perceived, a task evaluator that is capable of ascertaining which tasks in the mission set need to be searched for and which can be executed, a task allocation process that utilizes a modified version of the A* with Bounded Costs (ABC) algorithm to select the best ordering of task for execution based on an optimization routing, and, finally, a task executor that handles transiting to and executing tasks orders received from the task allocator. To evaluate Minion Task’s performance, the modified ABC algorithm used by the task allocator was compared to a greedy and a random allocation scheme. Additionally, to show the full capabilities of the system, a partial simulation of the 2018 Maritime RobotX competition was utilized to evaluate the performance of the Minion Task algorithm. Comparing the modified ABC algorithm to the greedy and random allocation algorithms, the ABC method was found to always achieve a score that was as good, if not better than the scores of the greedy and random allocation schemes. At best, ABC could achieve an up to 2 times improvement in the score achieved compared to the other two methods when the ranges for the score and execution times for each tasks in the task set as well as the space where these tasks could exists was sufficiently large. Finally, using two scenarios, it was shown that Minion Task was capable of completing missions in a dynamic environment. The first scenario showed that Minion Task was capable of handling dynamic switching between searching for and executing tasks. The second scenario showed the algorithm was capable of handling constraints on the ordering of the tasks despite the environment and arrangement of tasks not changing otherwise. This paper succeeded in proving a method, Minion Task, that is capable of performing missions in dynamic maritime environments.
Scholarly Commons Citation
Hendrickson, James, "Dynamic Task Allocation in Partially Defined Environments Using A* with Bounded Costs" (2021). Doctoral Dissertations and Master's Theses. 592.
Artificial Intelligence and Robotics Commons, Robotics Commons, Theory and Algorithms Commons