Quantum-Assisted Optimization Guided by Machine-Learned Risk Maps for Aerial Surveillance Routing

Abstract

Analog quantum computing with neutral atom computers is rapidly evolving into a promising way to perform combinatorial optimization tasks at scale. As technology matures, exploring real use cases leveraging such unique information processors becomes warranted; one example is logistics management where vehicles are assigned routes to perform a given task. Autonomous aerial vehicles (drones) are increasingly used for surveillance over a given territory. In this paper, we propose an innovative framework integrating classical machine learning forecasts and quantum annealing to find the best combination of tours for a given surveillance criterion. The risk maps produced by machine learning are used to find possible surveillance tours, while a quantum-assisted heuristic chooses the best combination, producing a full surveillance plan. By modelling our decision problem for drone paths as a Team Orienteering Problem (TOP) with additional constraints, we construct a new heuristic using quantum annealing to find good solutions. We apply this framework to wildfire prevention, where finding tours most likely to detect a starting fire has clear benefits. We study the effect of larger quantum resources on heuristic performance and compare those results to the use of classical resources.

Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.