Abstract—Usage of multiple unmanned aerial vehicles (UAV) in a certain mission makes flight route planning more complicated and slower. In order to obtain better performance, in the literature, most of the researchers propose using evolutionary algorithms and artificial intelligence approaches based on heuristics as optimization techniques. In addition to this, parallel programming approaches increase the computation performance. Therefore, this study focuses to discuss and solve the route planning problem for multi-UAV systems by using optimization techniques based on an evolutionary algorithm: simulated annealing. The travel cost and execution time are downsized in this work by optimization on algorithm and code. We implemented CPU based parallel solution to compare results with the GPU-accelerated one. The efficiency and the effectiveness of our parallelized and optimized solution is demonstrated through simulations under different scenarios. The results show that our optimized GPU based parallel solution for route planning problem is up to 1.6 times faster than serial and parallel CPU solutions. Moreover, our optimized GPU solution is better on cost than other solutions. It is shown that our GPU based approach is the fastest one and increases performance thanks to the massive parallelization capabilities of GPUs. Index Terms—GPU programming, parallel programming, route planning, simulated annealing. I. INTRODUCTION Unmanned aerial vehicles (UAV) have various usage areas from delivery of goods to battlefield use. As the UAVs’ cost decreased with current technological developments, multiple minimized UAVs can be used for better performance instead of using a single large UAV. This kind of multiple usage of UAVs makes the flight route planning problem for these systems more complicated. In multi-UAV systems, total travel distance should be divided wisely among UAVs. Each UAV have to travel at minimum cost and their cost have to close each other if they have equal resources. In literature, Simulated Annealing (SA) algorithm is used many times for shortest path problem. In Manuscript received February 8, 2019; revised January 11, 2020. Seval Capraz is with Ante Grup Bilisim Ticaret A.S., Ankara, Turkey. She is also with the Department of Computer Engineering, Hacettepe University, Ankara, Turkey (e-mail: seval.capraz@antegrup.com.tr). Halil Azyikmis and Adnan Ozsoy are with the Department of Computer Engineering, Hacettepe University, Ankara, Turkey (e-mail: hazyikmis@hacettepe.edu.tr, adnan.ozsoy@hacettepe.edu.tr). our case, multiple UAVs are traversing the predefined target locations -waypoints- with total minimum cost which means total distance traversed by each UAV required to be minimum. It is a kind of NP-hard problem. This kind of problems are hard to solve and take a lot of time and energy. The problem is to find an acceptable solution that is near-optimal solution rather than the best solution. In order to simulate the problem in best way we have used Traveling Salesman Problem Library (TSPLIB [1]) which is created and presented by University of Heidelberg, Germany. We used this library as a dataset because it is used in many study in literature as well as in Turker et al.(2016) [2]. Therefore we compare our results with other studies. Dealing with these kind of complex algorithms is hard. The SA algorithm is difficult to implement and very slow. It is one of the best algorithms to find best route with best cost among other algorithms for many UAVs. For this reason, we focus on this algorithm and how to implement it faster. The modern supercomputing shows that GPUs (Graphics Processing Units) are very good accelerators speeding up all sorts of tasks from very hard problems to these kind of algorithms. Why is CPU not enough for it? Because GPUs offer many benefits. Architecturally, the CPU is composed of just a few cores with lots of cache memory. This cache memory can handle a few software threads at a time. In contrast, a GPU is composed of hundreds of cores that can handle thousands of threads simultaneously. It is ideal for algorithms which do a lot of little jobs like comparison. GPU-based route planning of multi-UAV system speeds the overall calculation. In this study, firstly, we implemented SA algorithm in CPU. It works only on CPU and it is slow. There are many proposed solutions in literature which uses only CPU and they are all slow. We need to minimize the calculation time to find best rouse plan for many UAVs. The best route plan means less in cost. It is so difficult to find best solution so we only want to find an acceptable one. There are also a few studies which uses GPU to run SA. It helps to minimize the run time of the algorithm. We realized the implemented solutions have lack of optimization. GPU has its own infrastructure. If we optimized the algorithm to fit well in GPU, we can gain more time. The optimization is so important and it improves both cost and run time in good manner. This paper is organized as follows. In the second section, related works are discussed. In the third section, background information of the proposed solution algorithm with simulated annealing (SA) are defined and algorithms for serial and parallel run are given. In the fourth section, serial An Optimized GPU-Accelerated Route Planning of Multi-UAV Systems Using Simulated Annealing Seval Capraz, Halil Azyikmis, and Adnan Ozsoy International Journal of Machine Learning and Computing, Vol. 10, No. 3, May 2020 471 doi: 10.18178/ijmlc.2020.10.3.959