Three-dimensional (3D) Dynamic Obstacle Perception in a Detect-and-Avoid Framework for Unmanned Aerial Vehicles

Abstract

In this paper, a 3D dynamic obstacle perception is developed in a detect-and-avoid (DAA) framework for unmanned aerial vehicles (UAVs) or drones. The framework requires only an end point coordinate for collision-free path-planning and execution in an environment with dynamic obstacles. The sense portion of the DAA framework takes data from an mmWave sensor and a depth camera while the detect portion of the framework updates a probabilistic octree when static and dynamic obstacles are sensed. Perception of dynamic obstacle was achieved by implementing an algorithm that clears the sensor’s field of vision before computing the occupied voxels and populating the probabilistic octree. The avoidance portion of the framework is based on rapidly-exploring random tree (RRT) but the framework is flexible to allow other types of planners. This work develops the DAA framework for a UAV in a dynamic 3D environment by modifying the MoveIt framework. The framework is implemented on a UAV platform equipped with an on-board computational unit. The simulation and indoor experiments were conducted, which show that the modified DAA framework with dynamic 3D obstacle perception can successfully sense, detect and avoid obstacle. Additionally, the proposed perception method reduced the path re-plan time.