Path Planning of Unmanned Aerial Vehicle Based on Supervised Learning

Abstract

Traditional modular-based path planning methods have certain limitations, they have high latency and computational complexity, and cannot work in unknown environments. Current learning-based path planning methods show significant advantages in solving different planning problems in high-dimensional spaces and complex environments. Learning path planning from demonstrations is a hot research topic. We propose an end-to-end framework based on supervised learning. The end-to-end mapping of unmanned aerial vehicle (UAV) from perception to collision-free trajectories is achieved by learning demonstrations generated based on global path planning methods in a simulated environment. Replacing the traditional modular approach with neural networks can reduce computational latency and prevent error accumulation. Experimental results show that our method has a better performance than traditional path planner.