For safe and feasible pathplanning in real time of autonomous parking system, a parking path planning algorithm based on constrained reinforcement learning with a hybrid action space is proposed in this paper. Specifically, the proposed algorithm employs a hybrid action space reinforcement learning framework that integrates discrete actions with continuous parameters to achieve parameterized trajectory planning, thereby enhancing the executability of planned paths. On this basis, a constrained reinforcement learning algorithm within the hybrid action space is designed to optimize safe policy execution, ensuring the safety of parking paths. Moreover, a curriculum learning mechanism is introduced during model training to guide exploration progressively, improving training stability and convergence speed. Finally, extensive comparative and ablation experiments are conducted on both perpendicular and parallel parking scenarios. The experimental results show that the proposed parking path planning algorithm outperforms existing stateoftheart methods in terms of success rate, safety, and realtime performance, exhibiting superior overall effectiveness.

In a mixed traffic ecosystem, accurately predicting the trajectories of surrounding vehicles is crucial for the safety of autonomous vehicles. However, existing technologies still face issues of accuracy and computational complexity in longterm prediction. A spatiotemporal interactive sparse attention model combined with intention probability is proposed in this paper, which predicts trajectories through an efficient encoderdecoder structure. The position mask matrix is first constructed to extract positional information from historical trajectories, and key features are selected using the sparse attention mechanism. The intention behavior analysis module is utilized to improve the accuracy of intention recognition. Finally, spatiotemporal features, positional features, and intention features are fused and input into the decoder, and the model is trained using a multitask learning approach. The experimental results show that, compared to the optimal algorithm on the HighD and NGSIM datasets, the proposed model achieves a notable reduction in root mean square error (RMSE) in longterm prediction of 3 to 5 seconds, significantly enhancing prediction accuracy. In addition, the model's performance in realworld scenarios is validated through road tests, further demonstrating its application potential in complex traffic environment.

For the problem that the spatiotemporal separation trajectory planning method used in autonomous vehicles is prone to insufficient vehicle flexibility, and even cannot generate feasible trajectories under complex working conditions, while the existing spatiotemporal unified trajectory planning method is difficult to meet the requirements of structured road application, a spatiotemporal unified planning method based on dynamic programming and numerical optimization algorithm is proposed. Firstly, the spatiotemporal unified coarse trajectory is generated by dynamic programming algorithm in Frenet coordinate system. In the process, deterministic sampling method is used to expand the child nodes. Then, taking the coarse trajectory as reference, the feasible spatiotemporal corridor is constructed in Cartesian coordinate system, and the NMPC optimization model is established to generate the final trajectory. Finally, the algorithm is verified by simulation. The results show that the proposed algorism has good adaptability to structured road, and can better balance the requirements of traffic efficiency, trajectory comfort and time consumption than other spatiotemporal unified algorithms.

To meet the demand of efficient and accurate perception in autonomous driving systems, relying solely on cameras makes it challenging to achieve highprecision and robust 3D object detection. An effective solution to address this issue is to combine cameras with costeffective millimeterwave radar sensors, enabling more reliable multimodal 3D object detection. An effective approach to address this problem is to combine cameras with costeffective millimeterwave radar sensors, enabling more reliable multimodal 3D object detection, which not only improves the accuracy of environmental perception but also enhances the system's robustness and safety. In this paper, an autonomous driving perception algorithm based on the fusion of millimeterwave radar and cameras, named HPRDet (historical pillar of ray cameraradar fusion bird's eye view for 3D object detection) is proposed. Specifically, a radar BEV (bird's eye view) feature extraction module called RadarPRANet (radar point RCS attention net) is designed firstly. It comprises a dualstream radar backbone that extracts radar features with two representations, and an RCSaware BEV encoder that distributes radar features into the BEV space based on radarspecific RCS characteristics. Secondly, Historical radar of Object Prediction paradigm is adopted, designing both longterm and shortterm decoders that operate only during training, thus avoiding additional inference overhead. Due to the sparsity of the input data in this network, multimodal historical multiframe input is introduced to facilitate more accurate BEV feature learning. Lastly, the millimeterwaveoptimized ray denoising method is proposed, which utilizes the information from the current frame's millimeterwave radar point cloud as prior knowledge to assist in proposal generation, thereby enhancing the query feature representation for the camera. The proposed algorithm is trained and validated on the largescale public dataset nuScenes, with the NDS reaching 56.7% on the backbone of Resnet50.

The power system and energy consumption characteristics of electric buses significantly differ from those of traditional buses with internal combustion engines, and conventional ecodriving strategies cannot fully adapt to electric buses. An energy consumption predictionbased deep reinforcement learning model is proposed for ecodriving of connected electric buses, taking into account of signal timing, information from preceding vehicles, energy consumption characteristics and comfort of passengers. Firstly, natural driving data from battery electric buses is collected, and a basic energy consumption model is established using vehicle dynamics, considering the regenerative braking characteristics of electric buses. A system identification model is then constructed to identify and estimate the unknown parameters in the basic energy consumption model. Next, the impact of different signal phases on speed patterns when entering and exiting signalized intersections is analyzed, and state variables that accurately describe traffic environment information are determined. Based on the constructed energy consumption model, a reward function is developed, considering safety, efficiency, energy conservation, and comfort. An optimization model for ecodriving strategies at signalized intersections for electric buses is established using the SAC (soft actor critic) algorithm. Finally, the proposed strategy is compared with the classic intersection passage strategy GLOSA. The results show that the proposed ecodriving strategy ensures vehicle safety across the four defined traffic scenarios. Despite an average increase in travel time of only 7.29%, the strategy enhances comfort by an average of 21.96% and reduces energy consumption by an average of 24.47%.

Accurate and reliable vehicle pose estimation is a critical input for intelligent vehicle decision, planning and motion control modules. In this paper, a positioning algorithm that integrates realtime slip ratio estimation and compensation for intelligent vehicles is proposed, which significantly enhances the fusion positioning accuracy of the Inertial Navigation System (INS) and Wheel Speed Sensor (WSS) during Global Navigation Satellite System (GNSS) interruption. Firstly, a realtime slip ratio estimation algorithm is proposed to correct the wheel speed information for different driving conditions, which uses vehicle acceleration and wheel speed data. Then, based on errorstate Kalman filter (ESKF), the corrected wheel speed data is fused with GNSS and Inertial Measurement Unit (IMU) information to achieve accurate and reliable vehicle pose estimation. The results of the realvehicle experiments show that during GNSS interruption, the Root Mean Square Error (RMSE) of velocity improves by up to 30% and the average horizontal position error mileage ratio reaches 1.68%.

The development of hydrogen fuel cell vehicle is one of the important measures to realize the "Double carbon" strategic goal in our country. As the main power source of fuel cell vehicle, proton exchange membrane fuel cell (PEMFC) system has nonlinear, strong coupling and timedelay characteristics. Those characteristics make PEMFC system have many difficulties when it is faced with complex power demand under various conditions like vehicle acceleration and climbing, especially in terms of precise control of gas supply and dynamic regulation of system response. The flow rate and pressure of gas supply play a decisive role in the output performance of PEMFC. Improper gas supply can lead to low efficiency of the stack and even damage or failure of the stack, and then affect the overall performance and service life of the system. Therefore, accurate gas supply system by optimizing the gas supply system is the key to improve the performance and extend the service life of PEMFC. Based on the establishment of a gas supply system model for PEMFC, in this paper the influence of key operating parameters such as oxygen excess ratio, gas pressure and gas pressure difference on the output performance of the system is analyzed. The synergetic control of oxygen excess ratio, cathode pressure and bipolar gas pressure difference in PEMFC system using nonlinear active disturbance rejection control (ADRC) algorithm is researched, which is then compared with those under the proportional integral derivative (PID) controller. Under PID control, the maximum overshoot of the oxygen excess ratio can reach 1, while under ADRC control, the overshoot only around 0.2, and the time to reach steady state is approximately 0.1 seconds, compared to around 1 seconds under PID control. After a sudden change in load current, the overshoot of the cathode gas pressure under the PID control algorithm is around 0.08 with large fluctuations, reaching a stable value within 2 seconds. Under the ADRC control algorithm, the cathode gas pressure can reach stable value within 0.8 seconds, with an overshoot much smaller than the PID control algorithm. Under PID control, the overshoot of the twostage gas difference can reach up to 0.15 with large fluctuations and longer time to reach stability, but under the ADRC controller, it can quickly and stably reach the set value of 0.2 bar with smaller fluctuations. The results show that the ADRC controller has better decoupling, robustness and stability under the disturbance factors of load current and hydrogen displacement action.

A speed loop optimization strategy based on cascaded extended state observer (ESO) is proposed to address the insufficient transient response of permanent magnet synchronous motor (PMSM) for steering power oil pump application in pure electric commercial vehicles. An extended Kalman filter (EKF) is designed as the basis of position sensorless control, with the adaptive design to avoid the problems of complicated parameter tuning and slow convergence. The antidisturbance and tracking ability of the speed loop is improved by the cascaded observation of internal and external disturbances, and the use of the linear state error feedback control rate (LSEFC) for replacement of the traditional PI controller. The bench tests show that the sensorless control scheme proposed in this paper significantly reduces the position estimation error under dynamic and steadystate conditions, with a steadystate error of only 1.4°. The optimized speed loop control effectively improves the system's performance of disturbance rejection and transient response. The reliability test shows that the steering power motor controller operates stably without performance failure.

An adaptive antidisturbance angle control strategy is proposed to address the nonlinear disturbances such as system parameter uncertainty, tire return torque obstruction, and coupling of electromagnetic characteristics of steering motors, which are faced by the active steering of SteerbyWire (SBW) system. A radial basis function neural network and robust sliding mode theory are used to design the outerloop cornering controller to adaptively compensate for the SBW system parameter uncertainty and tire return torque obstruction. Linear selfimmunity control is introduced into the innerloop current controller to cope with the coupling problem of electromagnetic characteristics of the steering actuator motor so as to improve the dynamic response performance of the SBW system. The simulation and hardwareintheloop test results show that the designed control strategy can help the SBW maintain the cornering steady state following error within 1.5° under various operating conditions.

The dualmotor coupled drive is a common configuration for the Electromechanical Transmission (EMT) system in tracked vehicles, which is characterized by inputoutput coupling, high power transmission efficiency, and variable load conditions. However, most existing torsional vibration control strategies for EMT are designed for symmetric excitation conditions on both sides, which do not align well with realworld operating scenarios. To improve the torsional vibration under asymmetric excitation, an EMT torsional vibration model is first established to investigate the vibration energy coupling effect between the two sides of the EMT under asymmetric excitation and its influence mechanism on the system's dynamic behavior. Based on these findings, a disturbance compensation method based on dualloop feedback is proposed, and a torsional vibration suppression strategy tailored for EMT under asymmetric excitation is developed. Verification results show that this strategy can effectively suppress torsional vibration of the EMT system under such excitation conditions.