Evaluating the performance of a lidar-based autonomous flying system for flying inside forest

Abstract. In recent years, flying autonomously with UAVs has been a widely researched topic. While most autonomous flying systems use the Global Navigation Satellite Systems (GNSS) for localization, using GNSS under the forest canopy is not applicable. Thus other sensors such as lidars are needed. The objective of this study was to evaluate the performance of an autonomous flying system for flying inside boreal forests. Two open-sourced algorithms, IPC (Integrated Planning and Control framework) path planner and control algorithm, and LTA-OM (Long-Term Association Lidar-Inertial Odometry and Mapping) simultaneous location and mapping (SLAM) algorithm, were chosen as the base for the custom-built quadrotor system. Livox Mid-360 lidar was used as the sensor. The system was evaluated with a set of extensive experiments. First, a set of simulation experiments were conducted with multiple flights flown in multiple forests with different tree densities and varying target flight velocities. Second, the localization accuracy of LTA-OM was evaluated by measuring the end-point drift from two manually flown real-world flights. Lastly, 33 autonomous real-world flights were performed in two different forest plots with varying levels of difficulty and tree density as well as with varying target flight velocities. Based on real-world experiments, the performance of the system was somewhat promising inside the medium-difficulty forest, but poor inside the dense difficult forest. The success rate of real-world flights was 10/15 inside the medium-difficulty forest and 6/15 inside the difficult forest with a target flight velocity of 1 m/s. The flight distance on the real-world flights was 60 meters.