Goal distance-based UAV path planning approach, path optimization and learning-based path estimation: GDRRT*, PSO-GDRRT* and BiLSTM-PSO-GDRRT*

Abstract

The basic conditions for mobile robots to be autonomous are that the mobile robot localizes itself in the environment and knows the geometric structure of the environment (map). After these conditions are met, this mobile robot is given a specific task, but how the robot will navigate for this task is an important issue. Especially for Unmanned Aerial Vehicles (UAV), whose application has increased recently, path planning in a three-dimensional (3D) environment is a common problem. This study performs three experimental applications to discover the most suitable path for UAV in 3D environments with large and many obstacles. Inspired by Rapidly Random-Exploring Tree Star (RRT*), the first implementation develops the Goal Distance-based RRT* (GDRRT*) approach, which performs intelligent sampling taking into account the goal distance. In the second implementation, the path discovered by GDRRT* is shortened using Particle Swarm Optimization (PSO) (PSO-GDRRT*). In the final application, a network with a Bidirectional Long/Short Term Memory (BiLSTM) layer is designed for fast estimation of optimal paths found by PSO-GDRRT* (BiLSTM-PSO-GDRRT*). As a result of these applications, this study provides important novelties: GDRRT* converges to the goal faster than RRT* in large and obstacle-containing 3D environments. To generate groundtruth paths for training the learning-based network, PSO-GDRRT* finds the shortest paths relatively quickly. Finally, BiLSTM-PSO-GDRRT* provides extremely fast path planning for real-time UAV applications. This work is valuable for real-time autonomous UAV applications in a complex and large environment, as the new methods it offers have fast path planning capability.