Multi-Agent Persistent Task Performance
Abstract
A method to control a system of robots to persistently perform a task while operating under a constraint such as battery life is presented. Persistently performing a task is defined as continuously executing the task without a break or stopping due to low battery constraints or lack of capabilities of a particular agent. If an agent is no longer able to execute the task it must be replaced by one that can continue the execution of the task. This is achieved through the utilization of two distinctions of agent roles: workers and helpers. This method is focused on addressing problems that require task handoffs where a second robot physically replaces a robot that has run low on battery. The worker agents are assigned the tasks, and perform the tasks until the constraint prevents further performance. Once a worker agent has reached a low battery threshold a task handoff is performed with a helper agent. This method utilizes a proactive approach in performing these handoffs by predicting the time and place that a worker will reach a low battery threshold and need to perform a handoff. This decreases the time necessary to respond to a low battery in these problems compared to prior developed reactive methods. As a result the total time needed by the multi agent team to complete a set of tasks is decreased. In this paper, the method is demonstrated utilizing a physics based simulator to model the behavior of the multi agent team. Experiments are run over three standard problems requiring agent task handoffs: sentry, inspection, and coverage. These demonstrate the effectiveness of the method when compared against the existing reactive methods.
Citation
Motes, James Donald (2018). Multi-Agent Persistent Task Performance. Undergraduate Research Scholars Program. Available electronically from https : / /hdl .handle .net /1969 .1 /166538.