 |
International Society of Science and Applied Technologies |
| Learning to Optimize Abort Thresholds for Multi-Attempt UAV Missions | ||||
| Author |
Yike Zhang
|
|||
| Co-Author(s) |
Qingan Qiu; Rongchi Sun; Rui Ye
|
|||
| Abstract | We consider multi-attempt mission-abort decisions for
a single UAV operating in an uncertain environment, where both
the mission completion rate and the external shock intensity are
unknown. The UAV is allowed to make at most N attempts to
accomplish the mission. A fundamental trade-off arises in each
attempt: continuing the mission increases the probability of completion
but also raises the risk of being shot down, whereas aborting
the mission reduces exposure risk but incurs a recall cost and
may leave the mission unfinished. We model the random mission
completion time and the stochastic shock arrival process by an
exponential distribution and a Poisson process, respectively, and
update the unknown parameters online through Bayesian learning
based on the observed mission duration and shock counts
from previous attempts. We formulate the problem as a finitehorizon
Markov decision process and characterize the optimal
abort threshold for each attempt. We further establish monotonicity
properties of the optimal value function.
|
|||
| Keywords | UAV; Mission Abort Decision-making; MDP; Bayesian Statistical Learning | |||
| Article #: RQD2026-187 | ||||
Proceedings of 31st ISSAT International Conference on Reliability & Quality in Design |