This article presents a novel approach to learning-based distributed model predictive control (LDMPC) for nonconvex optimization problems which aims to enhance the distributed system’s consensus and avoid collision. Selecting the objective function of a distributed model predictive control (DMPC) system over a finite horizon to maximize performance and ensure safety is a challenging problem. The current work of this article is to introduce a function approximator that integrates DMPC and reinforcement learning (RL) through policy iteration (PI) to reconstruct the terminal cost function and reformulate the finite time nonconvex optimization problem. This work decouples the constraints and objective functions in the optimization process between multiple agents and introduces an improved alternating direction multiplier method (ADMM) as an consensus optimization solution of LDMPC. Moreover, the convergence, feasibility, and stability properties of our algorithm are proved in this article. The numerical example shows that the method updates can be performed distributively without inconsistency and demonstrates the effectiveness and safety of the LDMPC.

The platoon technology is gradually applied in the intelligent transportation systems to improve the safety, efficiency, energy saving, and emission reduction of vehicles during driving. However, in contrast to the traditional ramp merging scenario that consists of individual vehicles, the existing platoons from the upstream bring new challenges to the merging problem on ramps. Without an effective coordination strategy, the unnecessary congestion and disintegration of the existing platoons will lead to an increase in driving costs and safety risk. Therefore, this paper optimizes the merging strategy of vehicle sequences involving platoons on highway ramps. A merging strategy, based on an adaptive headway threshold, is first proposed, where the comprehensive economic benefit is considered as its optimization objective while taking into account the decrements of the time, fuel, and carbon emission costs. The vehicle sequence merging is then modeled as a Markov decision process to derive the optimal threshold, where the action set includes three actions: accelerating, decelerating, and maintaining the constant cruising speed. Afterwards, the constraints are brought to the decision-making to ensure the safety of the merging process. The calculation of the action value, combining the derived optimal threshold and constraint, is then detailed. Finally, the results of the simulation are evaluated, and a comparison between the proposed strategy and other existing methods is conducted, which demonstrates that the proposed approach improves the comprehensive economic benefits.

Bilevel optimization has attracted substantial attentions in recent years due to its wide applications in machine learning. However, most existing algorithms either primarily developed under centralized setting or suffer expensive inner-loop updates for hypergradient estimation. What's worse, the projection operator in constrained scenario may demand prohibitively computational cost, which further necessitates efficient projection-free bilevel optimization algorithms over networks. To fill this gap, we propose a novel single-loop distributed Frank-Wolfe algorithm DBO-FW for constrained bilevel optimization problems by simultaneously leveraging a nested approximation technique and a gradient tracking mechanism to locally estimate the global hypergradient. Moreover, we provide the convergence guarantee for the proposed DBO-FW. Numerical results also validate the efficiency of our algorithm.

This study proposes a cooperative platoon formation scheme in a multi-lane traffic environment with connected and autonomous vehicles (CAVs). It coordinates the lane-changing decisions and longitudinal trajectories of CAVs to form platoons based on each vehicle’s target lane, aiming to reduce the negative impact of lane-changing maneuvers on traffic flow and improve platoon formation efficiency. Mathematically, a vehicle model for platoon formation is developed which couples the lateral lane-changing and longitudinal car-following behaviors of vehicles. A model predictive control-based mixed integer linear programming (MILP) problem is formulated, which optimizes all vehicles’ lateral and longitudinal trajectories and improves maneuverability and efficiency. In contrast to the majority of existing studies, the configurations of platoons to be formed are not predefined and can be jointly optimized to further improve flexibility. Moreover, the framework that integrates configuration decisions, trajectory planning and control is executed dynamically with a rolling horizon based on real-time traffic states to enhance reliability. To validate the proposed scheme, we conduct simulation experiments in SUMO to implement the cooperative platoon formation in a typical three-lane traffic flow with different traffic demand levels. The extensive comparison results indicate the superiority of the proposed method in improving traffic flow speed and platoon formation efficiency.

Positioning and posture of the robotic joints and end effector could probably introduce random initial errors. Those errors could exponentially deteriorate with compounded of common noise perturbation to cause the final failure of repetitive tracking control. To better improve the tolerance of those complex errors, a novel reciprocal of the exponential varying-parameter recurrent neural network (RE-VP-RNN) is proposed in this article to consider superimposed noise interference including the initial position deviation and noise perturbation together. Theoretical analysis further proves the convergence of the proposed method. The effectiveness, accuracy, and robustness of the proposed RE-VP-RNN solver are verified by simulation and physical experiments on three representative redundant and hype-redundant manipulators. The proposed model could be widely used in robot control for high-precision machining scenarios such as medical, industry, and aviation.

Both road users and road administrators are keen to know traffic flow of fine-grained vehicle type. Successful prediction on the traffic flow of heavy, medium and small vehicle could contribute to the improvement of travel safety and efficiency. However, the classification on vehicle type is always not accurate enough using in practice. It could cost a lot to identify from the additional video cameras to cover the full-length of large-scale freeway with high-resolution to capture vehicles clearly. In this paper, empirical data are cleaned, normalized, compensated, filled, decoded and filtered with help of the fusion of vehicle detector data, remote microwave sensors data and toll collection data. The traffic flows of fine-grained heavy, medium and small vehicles are successfully reconstructed. Improved deep belief network (DBN) are then proposed to forecast traffic flow of different types of vehicles in 30-, 60- and 120-minutes time interval. Random-selected road segments on a ring way around a city are trained with data accumulated three months and predict data in the next month. According to prediction error analysis, the proposed method performs better in estimation and forecasting, with respect to the existing methods, especially for longer time prediction and heavy vehicle prediction. It would benefit traffic control to prevent freeway congestion escalation, protect the traffic infrastructure via heavy vehicle control, reduce the road risk, prompt quick emergency response and eventually contributes to more applications for intelligent transportation system (ITS).

Allocation and dispatching strategies of rescue vehicles in emergency situation on the freeway are investigated in this paper. Three steps are proposed for rescue vehicles to arrive traffic accident point as fast as possible. Firstly, traffic accident level are cataloged based on severity of accident and other influencing factors. Then, which types and how many rescue vehicles are required for the specific traffic accident according to corresponding accident level. Finally, three dispatching methods are proposed and verified in a case study. This paper selected a traffic accident randomly happened on Hanning Freeway in Shaanxi Province as a case study. The result shows the application of the three methods in case study and contains optimal rescue vehicle dispatching method. This paper provides suggestions for emergency handling of traffic accidents on the freeway and gives an intelligent way to improve the management of the freeway.

The dynamic load on the bridge is normally generated by the traffic flow. It is therefore the acquisition of spatiotemporal information for vehicles driving on a bridge is of great significance to assess bridge structures. In this paper, we proposed an improved deep learning network, which is inspired from YOLOv4 (You only look once) network structure. The transfer learning method is applied in training, and designated nine sets of anchor values are obtained by the K-means clustering. The Soft-NMS (Non-Maximum Suppression) algorithm is fused to improve the detection effect of overlapping targets. At the same time, the SENet (Squeeze-and- Excitation Networks) is used to assign weights to the features of each channel to learn the correlation between different channels. Data set are established with video clips cut from the surveillance system currently used on a long span cable-stayed bridge in China. The proposed method is compared with SSD (Single Shot MultiBox Detector), YOLOv3 and YOLOv4 algorithms. Experimental proves that the proposed method can detect vehicle information more accurately, efficiently and stably. The result could extend to other bridge monitoring applications in future.