How to plan the collision-free path effectively is crucial for autonomous ground vehicles to avoid collisions. The vehicle may lose its lateral dynamic stability due to high speed, time-varying speed and multifriction road during tracking the collision-free path. In this article, an advanced collision avoidance framework is proposed to avoid collision efficiently with road friction estimation and dynamic stability control. First, an improved A* algorithm is constructed to generate a desired trajectory for collision avoidance, which is capable of observing road regulations and overcoming the vehicle mechanical drawbacks. Next, a model predictive control-based path-tracking controller is established to solve the tracking task as a multiconstraint and multitarget optimization problem, where optimized steering angle and additional yaw moment can be calculated. Meanwhile, a novel long short-term memory based road friction coefficient estimator is built to observe road friction. Moreover, electric power steering system controls the steering motor to realize the desired steering angle. The additional yaw moment can be obtained by differential braking. Finally, hardware-in-the-loop platform tests are conducted to validate that the proposed controller can not only avoid collisions effectively, but also have a good performance on keeping the vehicle dynamic stability with accurate road friction estimation.

An advanced LKS (lane keeping system) for use on curving roads is presented to maintain autonomous vehicle driving within the target lane, without unintentional lane departure. There are the following two main objectives in designing this system: one is performing perfect lane keeping and the other is ensuring the dynamic stability of the vehicle, especially when driving on a curving and low-friction road with time-varying high speed. In this paper, a combined vehicle model, consisting of a lane keeping model and a vehicle lateral dynamic model, is firstly introduced. Then, a novel adaptive-weight predictive controller is used to calculate the desired steering angle and the additional yaw moment which provide coordinated control forlane keeping and dynamic stability control. Meanwhile, a square-root cubature Kalman filter-based vehicle sideslip angle observer, with a strong tracking theory modification (ST-SRCKF), is established to estimate the sideslip angle during the driving process. Finally, HIL (hardware-in-the-loop) tests and field tests are constructed, and the results show the effectiveness of our proposed LKS controller and ST-SRCKF sideslip angle estimation.

In the context of autonomous driving, it is both critical and challenging to locate 3D objects by using a calibrated RGB image. Current methods typically utilize heteroscedastic aleatoric uncertainty loss to regress the depth of objects, thereby reducing the impact of noisy input while also ensuring the reliability of depth predictions. However, experimentation reveals that uncertainty loss can also lead to serious overfitting issue and performance degradation. To address this issue, we propose MonoSample, an augmentation method that collects samples from the dataset and places them randomly during training. MonoSample takes into account the occlusion relationships and applies strict restrictions to ensure the verisimilitude of the enhanced scenes. Furthermore, MonoSample avoids the complex conversion process between 2D and 3D, thereby enabling the extraction of a large number of samples and efficient operation. Experiments on different models have verified its effectiveness. Leveraging MonoSample in DID-M3D, our model achieves state-of-the-art (SOTA) performance on the KITTI 3D object detection benchmark. The code will be published in https://github.com/EE615/MonoSample .

Lane detection is an essential task in autonomous driving. A good lane detection model should achieve many objectives, such as high accuracy, rapid detection, and low memory. In this article, a grid-based network (G-NET) is designed to realize the aforementioned goals. In G-NET, the traditional pixel-level semantic segmentation is replaced with the area-level grid segmentation to release the detection burden. Then, a position vector is introduced to indicate where lane key point is in the grid. Meanwhile, the novel rolling convolution layer following with the down-sampling and up-sampling convolution layer has been designed for good feature extraction, ensuring each feature grid perceives all other grid features in the feature map. Then, an adaptive hyperparameter branch is introduced to calculate the binary threshold effectively. Finally, the detected lane points are classified into different lanes by introducing distance-based quaternion. G-NET is extensively evaluated on three most widely datasets: TuSimple, CULane, and CurveLanes. The results show that G-NET has a state-of-the-art performance. Meanwhile, field tests are conducted.

The Maglev transportation mainly completes the propulsion, levitation and guidance functions, which generally require the application of linear motor. Based on this, a novel transverse flux linear synchronous motor is proposed in this paper which is integrated with propulsion, levitation and guidance. The whole structure and principle of the novel transverse flux linear synchronous motor are discussed. Then, the air gap magnetic flux density is calculated based on the equivalent current layer analysis method, and the propulsion and levitation are calculated with different operating times. Besides, the guidance force is calculated with different transverse displacements by using the Schwarz-Christoffel conformal mapping method. After the simulated of the finite element model, the analytical results of the forces are compared with the finite element method. Finally, all the calculated results of the propulsion, the levitation and the guidance are validated by an experimental prototype, which provide an electromagnetic theory analysis for electromagnetic suspension transportation that integrated with propulsion, levitation and guidance.

As a complex nonlinear system, modeling analysis is an important method to the research of the static and dynamic characters of the locomotive diesel engines. Also the model facing the control behavior verification can be used in the electronic control unit's HIL(hardware-in-the-loop) simulation which can lessen the cost and uncertain factors in the later platform experiments. With the mode of system identification, this paper builds the dynamic model of the diesel with neural network. Aiming at the nonlinear character of the diesel with large inertia, the paper uses NARMAX(Nonlinear Auto-Regressive Moving Average with eXogenous inputs) as the main structure and uses LM(Levenberg-Marquardt) algorithm to train the network. In order to overcome the redundancy of the network structure caused by the artificial experience, this paper puts forward the Time Cycle Comparison method to determine the ranks of the input signals and uses the Optimal Brain Surgeon strategy to optimize the network structure. The simulation results proves that this method can eliminate the redundancy and improve the generalization capability of the network commendably under the same output error scopes. Comparison between the train results and the measured results shows that the dynamic model has the good real-time performances and little output error. So the model can meet the need of the system character analysis and technology application.