Obstacle avoidance path planning algorithm for multi-rotor UAVs

Unmanned Aerial Vehicles (UAVs) have been commonly used for the plant protection in modern agriculture. Autonomous operation is a heated issue of UAVs development, while obstacle avoidance is one of essential abilities. If obstacles are not effectively avoided during automatic operation, the security of UAVs will be inevitably at risk. This study proposed a novel method of collision-free trajectory planning for multi-rotor UAVs, which used modified dynamic optimisation, thereby to deal with autonomous obstacle detection and avoidance. A quad-rotor UAV, Carto F4, equipped with the workload of 5 kg, was selected as the flight platform. A LIDAR, Rplidar S1, and a PIXHAWK flight controller were used on the UAV. Meanwhile, a high-speed computing module, NVIDIA TX2, was used for complex computation. The specific method of trajectory planning consisted of three procedures during optimisation. First, a probability grid map was established to serve as the environment map, using the binary Bayesian probability. Then, an optimal bi-directional Rapidly-exploring Random Trees (RRT) was developed to search a complete and low-cost path for the UAV. Specifically, a systematic optimisation included the application of both centroid bias sampling and online-rolling optimization. The centroid bias sampling was used for the mutual guidance in node growing, while the online-rolling optimization was used for the avoidance of repeated growing of nodes. A more efficient path was established according to the two steps. Third, a dynamic optimisation of full trajectories was applied, where the dynamic optimization of minimizing the fourth derivative of displacement was utilised to make the path to be a trajectory that was more in line with dynamic performance, thereby to achieve stable avoidance of obstacles. A minimum snap was employed during optimisation, where three types of constraints were added, containing planning constraints, continuity constraints, and dynamic constraints. Meanwhile, a probability grid map with high and low expansions was developed to ensure that the full trajectory did not interfere with obstacle areas. In-depth simulation test results illustrated that the re-planning duration of obstacle avoidance could be reduced by up to 23.69%, compared with the non-improvements, indicating that the dynamic optimisation made the trajectory more feasible and smoother. Moreover, the duration of planning of effective trajectories for obstacle avoidance was less than 0.33s, and the average speed of trajectory tracking of obstacle avoidance was not lower than 1.16 m/s. In addition, the extension rate of trajectories for obstacle avoidance was up to 20.82%, indicating that the efficiency and effectiveness of trajectory planning were improved. The proposed method of obstacle-free trajectory planning for multi-rotor UAVs can provide a sound of theoretical scheme and technical reference for the autonomous operation and obstacle avoidance of multi-rotor UAVs.