Map Uncertainty Reduction for a Team of Autonomous Drones Using Simulated Annealing and Bayesian Optimization

This research focuses on the problem of reducing the uncertainty rate of an environment in the context of surveillance. A human operator designates a set of locations to be checked by a team of autonomous quadcopters. The goal of this work is to minimize the uncertainty rate of the environment while penalizing solutions whose the total travelled distance is large. To cope with this issue, the A* algorithm is employed to plan the shortest path between each pair of points. Then, a simulated annealing algorithm is used to allocate tasks among the team of drones. This paper discusses three different objective functions to solve the problem whose cost-efficient and feasible solutions can be obtained after a few minutes. The presented work also deals with optimization of the simulated algorithms parameters by using Bayesian optimization. It is currently the state-of-the-art approach for the problem of hyper-parameters search, for expensive to evaluate functions, since it allows to save computation time by modeling the cost function. The Bayesian optimizer returns the best parameters within one day, while the use of grid search methods required weeks of computations.