Towards ensemble AI behaviours for satellite plan execution

The execution of plans onboard satellites is a challenging process due to the inherently non-deterministic and stochastic environment. We define an execution agent as the system responsible of dispatching and monitoring the actions of a plan while guaranteeing real-time response. In this abstract
we propose an execution agent that is able to execute the a number of plans in parallel, managing a dynamic priority between them. This method utilises the idea of ensemble AI, which takes input from multiple voices, such that each voice has its own goal specific interest, and with the help of a human control variable or function to manage priority. Finally an arbiter agent chooses an optimal action to take at each moment. Our results show that the ensemble execution agent is able to scale to much larger sets of mission objectives when
compared to a system that attempts to merge these objectives before planning and execution.