Path planning plays a crucial role in the execution of pursuit tasks for multiple unmanned ground vehicles (multi-UGVs). Although existing popular path-planning methods can achieve the pursuit goals, they suffer from some drawbacks such as long computation time and excessive path inflection points. To address these issues, this paper combines gradient descent and deep reinforcement learning (DRL) to solve the problem of excessive path inflection points from a path-smoothing perspective. In addition, the prioritized experience replay (PER) method is incorporated to enhance the learning efficiency of DRL. By doing so, the proposed model integrates PER, gradient descent, and a multiple-agent double deep Q-learning network (PER-GDMADDQN) to enable the path planning and obstacle avoidance capabilities of multi-UGVs. Experimental results demonstrate that the proposed PER-GDMADDQN yields superior performance in the pursuit problem of multi-UGVs, where the training speed and smoothness of the proposed method outperform other popular algorithms. As a result, the proposed method enables satisfactory path planning for multi-UGVs.
The registration of laser point clouds in complex conditions in wilderness scenes is an important aspect in the research field of autonomous vehicle navigation. It serves as the foundation for solving problems such as environment reconstruction, map construction, navigation and positioning, and pose estimation during the motion process of autonomous vehicles using laser radar sensors. Due to the sparse structured features, uneven point cloud density, and high noise levels in wilderness scenes, achieving reliable and accurate point cloud registration is challenging. In this paper, we propose a semantic-supervised sparse point cloud registration network (S3PCRNet) aiming to achieve effective registration of laser point clouds in wilderness large-scale scenes. Firstly, a local feature aggregation module is designed to extract the local structural features of the point cloud. Then, based on rotation position encoding, a randomly grouped self-attention mechanism is proposed to obtain the global features of the point cloud through learning. A semantic information weight matrix is calculated to filter out negligible points. Subsequently, a semantic fusion feature module is utilised to find reliable correspondences between point clouds. Finally, the proposed method is trained and evaluated on both the RELLIS-3D dataset and a self-made Off-road-3D dataset.
Most existing methods of active tracking focus mainly on slow-moving objects, resulting in limited adaptability to objects with variable speeds. To bridge this research gap, a novel spherical coordinate guided adaptive active tracking (SCAAT) approach based on the pan–tilt–zoom (PTZ) camera machine vision system is proposed in this study. For object detection and tracking, YOLOv5 and DeepSORT are employed in the PTZ vision system. The spherical coordinates and angular speeds of the moving object can be acquired under the spherical coordinate system. For practical application, the start-time and start-angle delay of the PTZ cameras are calibrated, and a speed control equation is conducted in the spherical coordinate system to reduce rapid location deviation between the PTZ and the moving object. To adapt different speeds of the object and avoid camera shaking under different zooms, an adaptive tracking window is designed to keep the object within the camera field of view. Experimental testing has been performed to evaluate the proposed SCAAT method. The results indicate that the SCATT can not only expand the effective following distance and zoom of the PTZ camera, but also effectively improve the accuracy and stability of active tracking for the moving object with large speed variations.
LiDAR point cloud-based place recognition (LPR) in unstructured natural environments remains an open challenge with limited existing research. To address the limitations of unstructured environments, such as sparse structural features, uneven point cloud density, and significant viewpoint variations, we present BSPR-Net, a dual-branch point cloud feature extraction approach for point cloud place recognition, which consists of a BEV - projection rotation - invariant convolution branch and a point cloud sparse convolution branch. This design enhances the representation capability of geometric structural features while aggregating rotation-invariant characteristics of point clouds, thereby better addressing the challenge of large viewpoint disparities in reverse-revisited unstructured environments. The proposed network was tested on multiple reverse-revisited sequences of the Wild-Places data set, a benchmark for unstructured natural environment place recognition. It achieved a maximum F1 score of 85.46 %, exceeding other classical methods by more than 4 %. The ablation experiments further confirmed the effectiveness of each module in improving place recognition performance.