HIGHSTAR: High-Speed and Efficient Online Autonomous UAV Exploration

Unmanned aerial vehicles (UAVs) are widely used in autonomous exploration, but their motion speed is underutilized due to inaccurate motion time cost evaluation and high computational cost. Existing methods either fail to consider UAV’s motion tendency and environment simultaneously or can’t ensure real-time planning in large 3-D environments. This paper presents a consistent, high-speed, and efficient online autonomous UAV exploration method. First, a motion primitive activated graph search method is proposed to fully take advantage of the UAV’s current velocity and acceleration. It improves motion time cost evaluation by simulating short-term motion tendencies with motion primitives and reduces the computational cost by searching on a voxel graph with a dynamic upper bound. Then, a minimum time trajectory to the optimal viewpoint with a non-zero terminal velocity constraint in a convex hull is optimized. Finally, an SE(3) coverage trajectory for unknown space around the exploration path is further optimized. Simulations in various environments with different speed settings show that the proposed method’s average UAV velocity is $18.6\%-116.1\%$ faster and its exploration efficiency is $13.8\%-49.5\%$ higher than state-of-the-art methods. Real-world tests verify its effectiveness. The source code of our method will be released at: https://github.com/NKU-MobFly-Robotics/HighStar Note to Practitioners—This paper proposes a real-time UAV planning method for high-speed exploration of large unknown 3-D environments. The key challenge the proposed method addresses is how to exploit the high-speed potential of UAVs in exploration. The proposed method uses a motion primitive activated graph search method to take advantage of the current UAV speed and acceleration in exploration target planning and plans an informative coverage trajectory. The proposed method achieves real-time and high-speed exploration planning in simulations. The experiments conducted in structured underground parking and large wild outdoor environments show the effectiveness of the proposed method.