Abstract
In this publication a new workflow of point cloud processing for LiDAR-based trajectory planning and collision avoidance is proposed, which enables autonomous vehicles such as UAV to navigate in semi-unknown environments. Semi-unknown means that no complete map of the area and of potential obstacles is available. To guarantee safe operation of the UAV, a low-level collision avoidance based on LiDAR-data is proposed. The approach consists of a voxel-grid-based downsampling, kd-tree-construction for efficient 3D range searching, a novel modification of the Euclidian clustering approach that is used for computationally inexpensive object detection, path planning based on artificial potential fields, and a final generation of a feasible trajectory by means of a model predictive controller (MPC).The proposed modification of the Euclidean clustering algorithm consists in choosing adaptive radii for the spherical search areas around the LiDAR points and yields superior clustering results for several test measurements in various environments than the existing algorithm that uses a fixed search radius while causing only negligible additional computational cost. Besides, the introduction of artificial potential fields to LiDAR measurements for trajectory planning and collision avoidance is proposed. By modelling point clouds as spatial distributions of electrical charges, an analytical expression for the potential field can be obtained. Subsequently, an arbitrarily smooth path to circumvent all obstacles is generated using gradient-descent based on which a feasible trajectory can be obtained by model predictive control. This novel approach for collision avoidance significantly reduces computational cost compared to graph-based state-of-the-art path planning solutions such as the A*-algorithm and its numerous derivatives, thus allowing near real-time solving of the 3D perception and cognition problem, i.e. obstacle detection and trajectory planning. Considering that the physical limits of payload capacity and power supply impose hard constraints for the available computing power on UAV, smaller computational cost offers a decisive advantage in order to enable low-latency on-boardprocessing of the sensor data. In order to evaluate the proposed approach, all algorithms were successfully applied to several static point clouds in a cluttered office environment. Additionally, near real-time capability was ensured by implementing the object detection and the trajectory planning on two asynchronous threads of an AscTec Mastermind board for UAV, equipped with an Intel Core i7-3517UE Dual-Core-CPU with 2 x 1.7 GHz. While the object detection thread was successfully running at variable refresh rates of approximately 1-10 Hz depending on the complexity of the scene, i.e. the size of the point cloud or the available free space, respectively, the trajectory planning including the MPC fulfilled the requirement of running independently on a separate thread at a fixed clock rate of 100 Hz.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.