Optimal Path Planning for Intelligent UAVs Using Graph Convolution Networks

Unmanned Aerial Vehicles (UAVs) are in use for surveillance services in the geographic areas, that are very hard and sometimes not reachable by humans. Nowadays, UAVs are being used as substitutions to manned operations in various applications. The intensive utilization of autonomous UAVs has given rise to many new challenges. One of the vital problems that arise while deploying UAVs in surveillance applications is the Coverage Path Planning(CPP) problem. Given a geographic area, the problem is to find an optimal path/tour for the UAV such that it covers the entire area of interest with minimal tour length. A graph can be constructed from the map of the area under surveillance, using computational geometric techniques. In this work, the Coverage Path Planning problem is posed as a Travelling Salesperson Problem(TSP) on these graphs. The graphs obtained are large in number of vertices and edges and the real-time applications require good computation speed. Hence a model is built using Graph Convolution Network (GCN). The model is effectively trained with different problem instances such as TSP20, TSP50, and TSP100. Results obtained from the Concorde Benchmark Dataset were used to analyze the optimality of the predicted tour length by the GCN. The model is also evaluated against the performance of evolutionary algorithms on several self-constructed graphs. Particle Swarm Optimization, Ant Colony Optimization, and Firefly Algorithm are used to find optimal tours and are compared with GCN. It is found that the proposed GCN framework outperforms these evolutionary algorithms in optimal tour length and also the computation time.