Semantic segmentation is a crucial component of autonomous driving. However, the segmentation performance in unstructured roads is challenging owing to the following reasons: (1) irregular shapes and varying sizes of road boundaries, (2) low contrast or blurred boundaries between the road and background, and (3) environmental factors such as changing light intensities and dust particles. To overcome these challenges, this study proposes SwinURNet, a Transformer-convoluted neural network (CNN) architecture for real-time point cloud segmentation in unstructured scenarios. First, the point cloud is projected onto a range image via spherical projection. Then, a lightweight ResNet34-based network is designed for encoding abstract features. A non-square Swin transformer is designed for decoding information and capturing high-resolution transverse features. A multidimensional information fusion module is introduced to balance the semantic differences between feature maps in CNNs and attention maps in transformers. A multitask loss function comprising boundary, weighted cross-entropy, and Lovász–Softmax losses are introduced at the network end to guide network training. Experimental data from autonomous mining trucks in the Baiyuneboite mining area is used to evaluate the performance on unstructured roads. The proposed architecture is also applied to the public unstructured dataset RELLIS-3D and the large structured dataset SemanticKITTI. The experimental results show performance gains of 74.2, 42.6, and 61.6% in accuracy and 8–19 FPS inference speed with the proposed architecture, surpassing those of the compared methods.

Accurate train positioning is crucial for train safety. In this paper, we propose a train positioning method which fuses vison and millimeter-wave radar data. The proposed method contains two parts: loop closure detection (LCD) and radar-based odometry. The loop closure detection part fuses the convolutional neural network (CNN) features and the line features to achieve accurate key location detection. The radar-based odometry part proposes a train speed measurement algorithm using millimeter-wave radar, and combines the results of loop closure detection to further realize train positioning. Experiments conducted on the Hong Kong metro Tsuen Wan line show that our proposed loop closure detection can achieve an efficient key location detection with 98.57% precision and 99.37% recall; the speed detection method fulfills the ETCS requirements; and the relative error of the proposed train positioning method is 0.45%. Besides, the proposed method has been applied on the Hong Kong Metro TSUEN WAN line.

Autonomous driving involves multi-timescale and multi-objective tasks coupled with long-term driving decision and short-term motion planning. However, existing studies tend to investigate them separately or assume they were synchronous and instantaneous, failing to deeply interpret their interconnections in autonomous driving. To address it, this study explored driving decision and motion planning in an integrated manner, and proposed a double-layer reinforcement learning (RL) framework to couple and optimize these two modules. Specifically, on the upper layer of the framework, a trajectory-level reward function associated with safety, efficiency, comfort and decision execution was proposed, and a decision-making model was established based on value function-based deep reinforcement learning (DRL) algorithms. Then, the reward function of the upper layer was taken as the objective function of the lower layer with relevant state constraints. The model predictive control (MPC) method was used to derive the optimal maneuver sequence guided by driving decisions, and the performance evaluation of motion planning was fed back to the decision-making for DRL parameter optimization. Accordingly, a closed-loop mechanism was developed consisting of forward decision-making output guidance and backward motion planning feedback optimization. Then, two-lane and three-lane interactive scenarios were set for framework training and testing. Last, comparative experiments were conducted using NGSIM dataset. The results demonstrated the effectiveness and the enhanced driving performance of the proposed framework in contrast to four benchmarks.

Trajectory planning for autonomous driving is challenging in unstructured scenarios such as mining sites. Existing studies mainly resort to a heuristic search-based planner to find a feasible trajectory. In the case of the narrow area, the heuristic function of the planner suffers from over-expansion problems, which may result in heavy computation burden and memory usage as well as potential failure. Therefore, a novel trajectory planning approach, namely learning and optimization-based trajectory planning (LOTP), is proposed, which is featured by a hierarchical structure consisting of two modules: (1) path searching, (2) speed profile generation. Firstly, a path searching method based on deep learning and Monte-Carlo tree search is proposed to generate a coarse path connecting starting and terminal points. Then, the path is smoothed using path optimization and provided to the speed-planning module as the reference. Next, a speed planning method based on quadratic optimization is developed, which allows to seek maximum driving comfort and energy saving. Last, extensive simulation experiments were conducted in the real environment of mining sites. The results verified that LOTP enhances the computational efficiency and success rate of path planning and helps generate an optimal speed profile. Furthermore, LOTP exhibits desirable potential for the practical application of autonomous driving at mining sites. Source implementation will be released as an open-source code.

Rail track segmentation is key to environmental perception of autonomous train. However, due to the complexity of railway track environment, critical issues such as the detection of rail tracks with different curvatures remain to be overcome. In this study, a novel architecture called FarNet is proposed for long-range railway track point cloud segmentation. The proposed FarNet is mainly divided into three parts, i.e., spherical projection, attention-aggregation network and results refinement. Specifically, spherical projection converts the LiDAR point cloud into a pseudo range image, and attention-aggregation network enables railway track detection using the pseudo range image. Furthermore, in the attention-aggregation network two components, i.e., spatial attention module and information aggregation module, are proposed to enhance the capability of rail track segmentation. Last, the results refinement helps further filter out the noise points after segmentation. Experimental results show that the proposed FarNet achieved 98.0% mean intersection-over-union (MIoU) and 98.9% mean pixel accuracy (MPA) for rail track segmentation.

Maneuver decision-making and trajectory planning play important roles in autonomous driving since a safe and flexible decision module is indispensable for navigation. Typical algorithms apply sampling methods to generate feasible trajectories. However, the fixed sampling distance and maneuver execution time in a sampling approach sacrifice the flexibility of algorithm. Moreover, since motion planning can be represented as a high-dimensional problem, it usually results in unnecessary samples that require additional resources to search the solution. Therefore, a semantic-level maneuver sampling and trajectory planning algorithm is proposed to solve the above problems. In the upper-level maneuver decision, the decision-making problem is formulated as a selection of the forward leading object. A semantic-level decision tree is built to sample long-term maneuver sequences, and the safety corridor of each maneuver sequence is calculated according to the surrounding environment. In the lower-level trajectory planning, the process is decoupled into longitudinal and lateral directions. First, a heuristic search method is proposed to generate longitudinal trajectory candidates for each maneuver sequence. Then, an exhaustive search algorithm is employed to synchronously generate the lateral trajectory within safety corridor. Among the generated trajectory candidates, the one with minimum cost will be chosen as the searching result. Furthermore, in order to improve the driving comfort, numerical optimization is adopted to refine the result by accounting for the constraints of kinematics and safety. Finally, the proposed method was evaluated through simulations of typical on-road dynamic scenarios, which help verify its performance with desirable computation efficiency of less than 32 ms.