A method and system for centralized speed control of unmanned vehicles

Abstract

The present application discloses a method and system for centralized speed control of unmanned vehicles, which relates to the field of unmanned driving, including: obtaining environmental data at an intersection through a roadside sensing device; performing multi-sensor data fusion and calculation through a roadside computing unit to obtain a multi-vehicle speed control plan; broadcasting the multi-vehicle speed control plan to unmanned vehicles at the intersection through a roadside communication RSU; the on-board controller of the unmanned vehicle controls the unmanned vehicle according to the multi-vehicle speed control plan, combined with environmental data within a preset range of the unmanned vehicle obtained by the on-board sensing device. In view of the lack of coordinated control of unmanned vehicles at intersections in the prior art, the present application utilizes a roadside computing unit to perform multi-sensor data fusion, combined with trajectory prediction and model predictive control algorithms, to generate a multi-vehicle speed control plan, and realizes centralized speed control of multiple vehicles at intersections.

Description

Technical Field

[0001] The present application relates to the field of unmanned driving, and in particular to a method and system for centralized speed control of an unmanned vehicle. Background Art

[0002] With the rapid development of technologies such as artificial intelligence and automatic control, driverless technology has become a hot topic and future development direction in the automotive industry. Driverless vehicles sense the surrounding environment through on-board sensors, and autonomously plan driving paths and control vehicle movement based on the sensed information, which is expected to greatly improve traffic efficiency and driving safety, thereby alleviating current problems such as traffic congestion and frequent accidents.

[0003] However, in actual road traffic scenarios, there are often a large number of unmanned vehicles and manned vehicles driving together, especially at urban intersections with dense traffic flow. There is a lack of effective coordinated control between multiple unmanned vehicles, which is prone to traffic conflicts and potential accidents. The general speed control method for unmanned vehicles mainly relies on on-board sensors to obtain information about the surrounding environment, such as lidar, cameras, etc. However, due to the limitations of sensor detection distance and field of view, only the speed of a single vehicle can be controlled, and it is difficult to take into account the interaction between vehicles. Multiple unmanned vehicles plan their speeds independently, which often leads to inconsistent decisions and the inability to form a unified and coordinated speed control, exacerbating the risk of vehicle conflicts.

[0004] In response to the above problems, a centralized speed control method based on vehicle-road collaboration has become a promising solution. This method utilizes the deployment of roadside equipment and V2X vehicle-road communication technology to obtain global information about vehicles and the traffic environment, and on this basis, optimizes the speed of all unmanned vehicles in the area as a whole. This centralized control architecture can coordinate the movement of multiple vehicles through the calculation and scheduling of roadside equipment, effectively reducing the risk caused by mutual interference between vehicles. In addition, V2X communication provides a low-latency, highly reliable information exchange channel between vehicles and roads, which helps roadside equipment to grasp the status and intentions of vehicles in real time, thereby making more accurate and coordinated control decisions. V2X communication can also expand the perception range of unmanned vehicles, make up for the blind spots of single-vehicle sensors, and enable them to obtain beyond-line-of-sight environmental perception capabilities, which is crucial to ensuring driving safety. Summary of the invention

[0005] In response to the problem of lack of coordinated control of unmanned vehicles at intersections in the prior art, the present application provides a method and system for centralized speed control of unmanned vehicles. By acquiring environmental data at the intersection, using a roadside computing unit to perform multi-sensor data fusion, and combining trajectory prediction and model predictive control algorithms, a multi-vehicle speed control scheme is generated to achieve centralized speed control of multiple vehicles at the intersection.

[0006] The purpose of this application is achieved through the following technical solutions.

[0007] One aspect of the present application provides a method for centralized speed control of unmanned vehicles, including: S1, obtaining environmental data at an intersection through a roadside sensing device, wherein the roadside sensing device includes a laser radar, a camera, and a millimeter-wave radar; the environmental data includes the position, speed, and posture of the vehicle; S2, based on the environmental data, multi-sensor data fusion and calculation are performed by a roadside computing unit to obtain a multi-vehicle speed control plan; S3, the multi-vehicle speed control plan is broadcast to the unmanned vehicles at the intersection through a roadside communication RSU; S4, the on-board communication OBU of the unmanned vehicle receives the multi-vehicle speed control plan broadcast by the RSU, and the on-board controller of the unmanned vehicle controls the unmanned vehicle according to the multi-vehicle speed control plan and the environmental data within a preset range of the unmanned vehicle obtained by the on-board sensing device.

[0008] Furthermore, S2, multi-sensor data is fused and calculated by a roadside computing unit to obtain a multi-vehicle speed control scheme, including: S21, time-space alignment and coordinate conversion of environmental data by a roadside computing unit to obtain multi-sensor data in the same coordinate system, wherein the multi-sensor data includes lidar point cloud data, camera image data and millimeter-wave radar data; S22, multi-sensor data in the same coordinate system is fused by a Kalman filter algorithm to obtain fused vehicle data, wherein the vehicle data includes vehicle position, speed and posture; S23, based on the fused vehicle data, the trajectory of the unmanned vehicle at the intersection is predicted by a trajectory prediction algorithm; S24, based on the predicted trajectory, the optimal speed control curve of each unmanned vehicle at the intersection is calculated by a model predictive control algorithm as a multi-vehicle speed control scheme.

[0009] Further, S22, multi-sensor data in the same coordinate system are fused through the Kalman filter algorithm to obtain fused vehicle data, including: taking the lidar point cloud data, camera image data and millimeter-wave radar data in the same coordinate system as input, and using the Kalman filter algorithm to establish a vehicle state equation and an observation equation; the vehicle state equation describes the predicted values ​​of the vehicle position, speed and posture, and the observation equation describes the observed values ​​of the vehicle state by the lidar, camera and millimeter-wave radar; the Kalman filter algorithm is used to predict and update the vehicle state to obtain fused vehicle data.

[0010] Further, S23, based on the fused vehicle data, the trajectory of the unmanned vehicle at the intersection is predicted by a trajectory prediction algorithm, including: taking the fused vehicle data as input, using the constructed trajectory prediction model, and predicting the trajectory of the vehicle in the future based on the historical trajectory and current state of the vehicle; using the B-spline curve fitting method to smooth the predicted trajectory as the trajectory of the unmanned vehicle at the intersection.

[0011] Furthermore, the constructed trajectory prediction model is used to predict the trajectory of the vehicle in the future based on the historical trajectory and current state of the vehicle, including: establishing a vehicle kinematic model to describe the changing relationship between the vehicle's position, speed and posture; wherein the vehicle kinematic model includes state equations of the vehicle's position, speed and posture; using the vehicle kinematic model as the state equation of the Kalman filter algorithm, and using the fused vehicle position, speed and posture as the observation equation to construct a trajectory prediction model; using the trajectory prediction model to predict the trajectory of the vehicle in the future based on the vehicle's historical trajectory and current state; wherein, through the prediction and update process of the Kalman filter algorithm, the predicted value of the vehicle trajectory is recursively calculated.

[0012] Further, S24, through the model predictive control algorithm, the optimal speed control curve of each unmanned vehicle at the intersection is calculated as a multi-vehicle speed control scheme, including: according to the predicted trajectory, a kinematic model of each unmanned vehicle is established, and a global optimization objective function is constructed, wherein the global optimization objective function includes minimizing the speed change, minimizing the acceleration change of each vehicle and the preset safety distance between the vehicles; for each unmanned vehicle, the speed and acceleration are used as control variables, the position is used as a state variable, the corresponding kinematic model is used as the state equation and constraint condition, and the global optimization objective function is used as the performance indicator function to construct the model predictive control problem of the corresponding vehicle; using the model predictive control algorithm, through a cyclic prediction method, in each control cycle, the model predictive control problem of the vehicle is solved to obtain the optimal speed control sequence of the corresponding vehicle in the next N time steps; the optimal speed control sequence of the next N time steps obtained by each unmanned vehicle in each control cycle is connected to form the optimal speed control curve of each vehicle; the optimal speed control curves of all unmanned vehicles are combined to generate a multi-vehicle speed control scheme at the intersection.

[0013] Furthermore, the model predictive control problem of the vehicle is solved to obtain the optimal speed control sequence of the corresponding vehicle in the next N time steps, including: in each control cycle, for each unmanned vehicle, according to the corresponding kinematic model, predicting the state change of the corresponding vehicle in the next N time steps, wherein the state change includes changes in position, speed and acceleration; using the predicted state change in the next N time steps as the initial condition, the speed and acceleration of the corresponding vehicle as the control variables, and the global optimization objective function as the performance indicator function, the model predictive control problem of the corresponding vehicle is converted into a sequential quadratic programming problem; using the sequential quadratic programming algorithm to solve the sequential quadratic programming problem, the optimal speed control sequence and the optimal acceleration control sequence of the corresponding vehicle in the next N time steps are obtained.

[0014] Further, S3, broadcasting the multi-vehicle speed control scheme to the unmanned vehicles at the intersection through the roadside communication RSU, including: S31, the roadside communication RSU extracts the identification information and the corresponding optimal speed control curve of each unmanned vehicle in the multi-vehicle speed control scheme; S32, the roadside communication RSU associates the corresponding optimal speed control curve with the communication address of the vehicle according to the identification information of each unmanned vehicle, and generates a speed control instruction for the vehicle; S33, the roadside communication RSU broadcasts the speed control instruction of each unmanned vehicle to the corresponding unmanned vehicle at the intersection through the communication protocol.

[0015] Further, S4, the on-board controller of the unmanned vehicle controls the unmanned vehicle according to the multi-vehicle speed control scheme and the environmental data within the preset range of the unmanned vehicle obtained by the on-board sensing device, including: S41, the on-board communication OBU of the unmanned vehicle receives the multi-vehicle speed control scheme broadcast by the RSU, and transmits the multi-vehicle speed control scheme to the on-board controller; S42, the on-board controller extracts the optimal speed control curve from the multi-vehicle speed control scheme; S43, the on-board controller obtains the environmental data within the preset range of the unmanned vehicle collected by the on-board sensing device through the CAN bus; S44, the on-board controller uses the extracted optimal speed control curve as the global constraint of the vehicle speed planning, combines the collected environmental data, and generates the optimal speed control instruction of the unmanned vehicle in the current scenario through the local speed planning algorithm; S45, controls the driving of the unmanned vehicle according to the optimal speed control instruction until it passes the intersection.

[0016] Another aspect of the present application also provides an unmanned vehicle centralized speed control system for executing an unmanned vehicle centralized speed control method of the present application.

[0017] Compared with the prior art, the advantages of this application are:

[0018] This application obtains environmental data at the intersection through roadside sensing equipment, uses roadside computing units to perform multi-sensor data fusion and trajectory prediction, and then combines model predictive control algorithms to generate multi-vehicle speed control plans, thereby realizing centralized control of the speed of unmanned vehicles at intersections. Compared with existing decentralized control methods, centralized control fully considers the interaction between vehicles, can better coordinate vehicle movement, reduce unnecessary acceleration and deceleration and parking and waiting, thereby effectively improving traffic efficiency and alleviating traffic congestion. At the same time, by predicting and planning vehicle trajectories in advance, potential conflicts caused by the intersection of vehicle trajectories can be prevented, greatly reducing the probability of accidents and ensuring driving safety.

[0019] The method of the present application integrates the environmental data acquired by the roadside sensing device and the vehicle surrounding environment data collected by the vehicle-mounted sensing device. On the one hand, the roadside device has a broader field of view and more powerful computing power, which can obtain global information and conduct macro analysis and prediction of traffic conditions at intersections; on the other hand, the vehicle-mounted sensing device can obtain local fine information around the vehicle to compensate for the perception blind spots of the roadside device. Through the complementary fusion of roadside and vehicle-mounted information, the ability to perceive the environment and the accuracy of trajectory prediction and speed planning can be significantly improved, making control decisions more intelligent and reliable.

[0020] This application adopts a hierarchical control architecture of "centralized planning + local control". At the upper level, the roadside computing unit comprehensively plans and optimizes the speed of each vehicle according to the traffic conditions at the intersection, and generates a multi-vehicle speed control plan that takes into account efficiency and safety; at the lower level, each unmanned vehicle generates a practical speed control instruction through a local speed planning algorithm based on the received speed plan and the real-time local environmental information perceived by the vehicle. This hierarchical control allows each vehicle to flexibly adjust according to the dynamic environment around it while ensuring the global optimum, greatly improving the robustness and adaptability of the plan.

[0021] This application uses a model predictive control algorithm to optimize and solve the vehicle's speed in the future in advance based on the vehicle's kinematic model and predicted trajectory, making the control forward-looking and able to prevent and respond to possible problems in advance. At the same time, model predictive control is a rolling optimization method that can adjust the optimization target in time within each control cycle based on real-time updated environmental information and vehicle status, so that the control decision is synchronized with environmental changes, has strong real-time performance, and can adapt to the needs of dynamic traffic scenarios.

[0022] This application transforms the model predictive control problem into a sequential quadratic programming problem through mathematical modeling, and then uses a mature and efficient sequential quadratic programming algorithm to solve it. While ensuring the solution speed, the obtained control sequence is globally optimal. This method of combining optimization control with convex optimization can not only quickly generate feasible solutions that meet various constraints, but also achieve multi-objective balanced optimization in a global scope, so that the control scheme achieves the optimal balance in safety, efficiency, comfort and other aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The present application will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0024] Figure 1 is a hardware system architecture diagram according to some embodiments of the present application;

[0025] Figure 2 is a layout diagram of a drive test device according to some embodiments of the present application;

[0026] Figure 3 This is a layout diagram of the vehicle-side hardware device shown in some embodiments of the present application. DETAILED DESCRIPTION

[0027] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0028] like Figure 1 As shown in the figure, the hardware system architecture and the equipment included in the centralized safety control system and device of unmanned vehicle speed based on V2X communication are as follows. In detail, the road test computing unit MEC: MEC (Multi-access Edge Computing) is an edge computing device deployed on the roadside, which undertakes the tasks of data fusion and calculation. It receives the original environmental data collected from the roadside perception equipment (such as roadside cameras, lidar, millimeter wave radar, etc.), and unifies the heterogeneous data into the same time and space coordinate system through multi-sensor data fusion algorithms such as Kalman filtering, forming a unified description of the traffic environment at the intersection. On this basis, MEC uses environmental perception algorithms to detect and track obstacles and vehicles, and uses trajectory prediction, traffic flow prediction and other algorithms to predict the future state of traffic flow at the intersection. Finally, MEC solves the multi-vehicle speed optimization problem through the model predictive control algorithm and generates the speed control plan for all unmanned vehicles at the intersection.

[0029] Road test communication unit RSU: RSU (Road Side Unit) is a wireless communication device deployed on the roadside, supporting vehicle-road communication protocols such as DSRC or C-V2X. The main function of RSU is to interact with MEC and OBU. On the one hand, RSU obtains the multi-vehicle speed control scheme after fusion calculation from MEC, and broadcasts the scheme to all OBUs within the intersection range through a dedicated short-range communication signal in the 5.8GHz frequency band; on the other hand, RSU can also receive vehicle status information sent by OBU, such as position, speed, heading angle, etc., and transmit this information to MEC for more accurate perception of intersection traffic conditions and improvement of speed control schemes.

[0030] On-Board Unit (OBU): OBU (On-Board Unit) is a communication device installed on unmanned vehicles and used in conjunction with RSU. OBU regularly broadcasts the vehicle's status information through a dedicated short-range communication signal in the 5.8GHz frequency band for reception by RSU and OBU of other vehicles. At the same time, OBU is also responsible for receiving the multi-vehicle speed control scheme broadcast by RSU and transmitting the scheme to the on-board controller to guide the speed control of the vehicle. OBU is a key communication node for realizing vehicle-road collaboration and centralized speed control of multiple vehicles.

[0031] Visual sensor - camera: A camera is a visual sensor that can obtain image information of the environment around the unmanned vehicle. The image data collected by the camera is analyzed through image processing algorithms, which can detect and identify key environmental elements such as lane lines, traffic signs, signal lights, pedestrians, other vehicles, etc., and provide visual feature information for the perception module. The camera can also cooperate with other sensor data to achieve semantic understanding of the environment, such as judging the status of traffic lights, vehicle intentions, etc., providing an important basis for unmanned vehicles to perceive the surrounding environment and avoid risks.

[0032] Laser sensor - LiDAR: LiDAR uses laser beams to scan the surrounding environment, calculates the distance and angle of obstacles through the flight time of laser reflection, and obtains 3D point cloud data of the environment. It can accurately measure the distance between the vehicle and the obstacle and draw the 3D outline of the surrounding environment. The point cloud data of LiDAR can be used for perception tasks such as positioning, map construction, obstacle detection and tracking of unmanned vehicles. It is one of the indispensable sensors for unmanned driving systems and can effectively make up for the problem of insufficient perception ability of cameras in distance errors, night and bad weather conditions.

[0033] Millimeter wave sensor - millimeter wave radar: millimeter wave radar is a radar sensor based on millimeter waves that can measure the distance, speed and angle of objects. Compared with lidar, millimeter wave radar is less affected by bad weather and lighting conditions and has a longer detection distance. Millimeter wave radar is mainly used for mobile obstacle detection and tracking in the perception module, and is particularly suitable for obtaining speed information of other vehicles. Millimeter wave radar can accurately estimate the relative speed between the vehicle and surrounding vehicles, determine the risk of collision, and provide an important reference for speed planning and control of unmanned vehicles.

[0034] Autonomous driving vehicle-side controller: The vehicle-side controller is the core execution unit of the unmanned driving system, responsible for summarizing and processing the data collected by the on-board sensing equipment, and finally outputting the vehicle control instructions. The controller receives multi-source heterogeneous data from on-board sensors such as cameras, lidars, and millimeter-wave radars, and fuses these data through perception algorithm software to form a unified understanding of the vehicle's surrounding environment. On this basis, the controller receives the multi-vehicle speed control scheme broadcast by the RSU from the OBU, and uses this scheme as a global constraint. Combined with the vehicle's own driving status and local road conditions, the optimal unmanned vehicle motion control instructions are generated through decision-making algorithms, trajectory planning, and speed planning processes, including throttle, brake, steering and other control quantities, to enable the vehicle to travel along the specified path and speed. The controller sends control instructions to the actuator through the CAN bus, and finally realizes the autonomous driving control of the unmanned vehicle.

[0035] In the V2X communication-based unmanned vehicle speed centralized safety control system provided in this application, the deployment and functions of the roadside equipment are as follows: Figure 2 As shown. The roadside equipment mainly includes the roadside communication unit RSU, the roadside computing unit MEC and a series of roadside perception equipment, including lidar, camera and millimeter wave radar: The roadside communication unit RSU1 is installed on the roadside, and communicates wirelessly with the roadside computing unit MEC2 and the on-board communication unit OBU on the unmanned vehicle through standards such as dedicated short-range communication (DSRC) or cellular vehicle-to-everything (C-V2X) in the 5.8GHz frequency band. The main function of RSU is to broadcast the multi-vehicle speed control information after MEC fusion calculation, and at the same time receive the vehicle status information sent by OBU to realize information interaction between vehicles and roads.

[0036] The roadside computing unit MEC2 is also installed on the roadside and is an edge computing device. It is connected to the RSU and various roadside sensing devices through a wired network to receive and process the massive heterogeneous data collected by them. The core function of MEC is to perform fusion calculations on multi-sensor data, perceive the traffic status of the intersection, and generate speed control plans for all unmanned vehicles within the intersection. Various types of roadside sensing devices are deployed at different locations of the intersection to achieve all-round three-dimensional perception of the intersection environment. Specifically, a laser radar 5 is installed at the center of the intersection formed by the intersection of roads A, B, and C to perform a 360-degree scan of the road environment in three directions, obtain three-dimensional geometric information of the intersection, and construct a point cloud map.

[0037] Cameras and millimeter-wave radars are deployed at each entrance of the intersection, which can carefully perceive the traffic conditions in all directions. Among them, on the road in the C direction, camera 3 and millimeter-wave radar 4 are installed to collect image information and speed information of vehicles coming from the C direction. Similarly, camera 8 and millimeter-wave radar 9 are installed in the A direction, and camera 6 and millimeter-wave radar 7 are installed in the B direction. In addition, traffic lights 10 are hung at the intersection to indicate and manage the traffic status of vehicles by displaying red, yellow and green lights. The status information of traffic lights can be broadcast to unmanned vehicles through RSU to help them make more reasonable speed decisions.

[0038] When working, LiDAR, camera and millimeter wave radar work together to collect environmental data at the intersection, and obtain information such as the position, speed, trajectory and other information of targets such as vehicles, pedestrians and obstacles. These data are transmitted to MEC in real time through the wired network for fusion calculation. MEC performs spatiotemporal registration of multi-sensor data, tracks the target's motion trajectory, predicts the future state of traffic flow, and finally generates a speed control plan for multiple vehicles. This plan is broadcast to all unmanned vehicles through RSU using V2X communication technology.

[0039] After receiving the broadcast from the RSU, the OBU on the unmanned vehicle transmits the speed control plan to the onboard controller. Under the constraints of the plan, the controller plans a safe, smooth and efficient driving trajectory based on the perception and positioning information of the vehicle, and controls the vehicle to pass through the intersection at the specified speed and trajectory, ultimately achieving unified coordinated control of all unmanned vehicles at the intersection.

[0040] In the V2X communication-based unmanned vehicle speed centralized safety control system provided in this application, the deployment and functions of the vehicle-side equipment are as follows: Figure 3As shown. The vehicle-side equipment mainly includes on-board sensing equipment, on-board communication unit OBU and autonomous driving vehicle-side controller. The on-board sensing equipment is installed at different positions of the unmanned vehicle to collect environmental information around the vehicle and provide real-time sensing data for the autonomous driving system. A millimeter-wave radar 15 is installed at the front of the vehicle, which is mainly used to detect the position and speed of obstacles in front, especially dynamic traffic participants such as other vehicles and pedestrians. Millimeter-wave radar has a long detection distance and strong penetration, and can work normally under severe weather conditions.

[0041] A camera 16 is installed on the inside of the front windshield of the vehicle cab. The camera can obtain video image information in front of the vehicle and identify important road elements such as lane lines, traffic signs, and signal lights, as well as other vehicles, pedestrians, obstacles, etc. through computer vision algorithms. It is an important visual sensor for the perception system of unmanned vehicles.

[0042] A laser radar 17 is also installed on the top of the vehicle, which acquires three-dimensional point cloud data of the vehicle's surroundings by emitting and receiving laser beams. The laser radar can accurately measure the distance and angle information of the target and build a three-dimensional space model around the vehicle. The laser point cloud data can be used to identify road boundaries, obstacle locations, etc., and is an important source of information for the autonomous driving system to perform positioning, perception and planning.

[0043] The heterogeneous perception data collected by millimeter-wave radar, camera and lidar are transmitted in real time to the autonomous driving vehicle controller through the vehicle network such as CAN bus19. The controller runs multi-sensor fusion algorithms such as Kalman filter and particle filter to align and fuse the data of different sensors in time and space to obtain a unified representation of the vehicle's surrounding environment, including the type, position, speed, trajectory and other information of the target, which is used to support the decision-making and control of autonomous driving.

[0044] The vehicle is also equipped with an on-board communication unit OBU18, which supports wireless communication protocols for the Internet of Vehicles such as DSRC or C-V2X. An important function of the OBU is to receive information broadcast by the roadside RSU, especially the multi-vehicle speed control scheme generated by the roadside computing unit MEC. At the same time, the OBU can also send the vehicle's status information, such as position, speed, acceleration, etc., to the RSU, and share it with roadside equipment and other vehicles.

[0045] The autonomous driving vehicle controller is the core computing unit on the unmanned vehicle. It comprehensively processes information from the on-board sensing equipment and OBU, and makes vehicle control decisions based on the predetermined driving strategy. On the one hand, the controller integrates the on-board sensor data, perceives the road environment around the vehicle, and performs local path planning in the vehicle coordinate system. On the other hand, the controller receives the RSU broadcast information forwarded by the OBU to obtain the overall state of the intersection traffic and the multi-vehicle speed control plan. The controller uses this plan as a global constraint to plan and optimize the speed of the local path, so that the vehicle can pass the intersection smoothly, safely and efficiently under the premise of coordinating with other vehicles. Finally, the controller converts the planning results into specific control instructions such as throttle, brake, and steering, and realizes the lateral and longitudinal motion control of the vehicle through actuators such as electronic throttle and electronic stability system, completing the autonomous driving process.

[0046] The roadside equipment at the intersection and the vehicle-side equipment on the vehicle respectively perceive and integrate the traffic environment. Among them, 14 is a vehicle going straight from direction A to direction B, and 11, 12, and 13 are a group of vehicles turning right from direction C to direction B. Due to the obstruction of buildings between vehicles 11 and 14, they cannot perceive each other through on-board sensors, and there is a potential risk of collision. At this time, the roadside perception equipment is required to collect the position, speed, and posture information of vehicles 11 and 14, and send these data to the roadside computing unit MEC through the roadside communication unit RSU. Based on the received data, MEC generates the speed control plan for vehicles 11, 12, 13, and 14 through the following steps:

[0047] S21: MEC performs spatiotemporal alignment and coordinate conversion on the heterogeneous environment data collected by the roadside perception equipment. For the lidar point cloud data, MEC first converts the point cloud from the lidar coordinate system to the global coordinate system of the intersection according to the installation position and posture of the lidar. The translation vector and rotation matrix of the lidar need to be considered during the conversion process.

[0048] For camera image data, MEC uses the camera's internal and external parameter matrices to convert image pixel coordinates into spatial coordinates in the intersection coordinate system. The external parameter matrix describes the position relationship between the camera coordinate system and the intersection coordinate system, and the internal parameter matrix describes the projection relationship between the pixel coordinates and the camera coordinate system.

[0049] For millimeter-wave radar data, MEC converts the target distance and angle information in its polar coordinate system into Cartesian coordinates in the intersection coordinate system based on the installation position and angle of the radar.

[0050] On the basis of coordinate conversion, MEC also needs to perform time synchronization. Since the data frequencies and timestamps of different sensing devices may be inconsistent, MEC unifies heterogeneous data to the same time base through methods such as linear interpolation to ensure the time consistency of data fusion. After spatiotemporal alignment and coordinate conversion, MEC obtains lidar point cloud, camera image, and millimeter wave radar target data under a unified coordinate system and time base, providing a data basis for subsequent multi-sensor fusion.

[0051] S22: MEC uses the Kalman filter algorithm to fuse multi-sensor data in a unified coordinate system. MEC establishes a state equation that describes the vehicle state based on the vehicle kinematic model. The state vector includes the vehicle's position, velocity, and attitude angle, and the input vector is the vehicle's acceleration and angular velocity. The state equation predicts the prior estimate of the vehicle state at the current moment through the state vector at the previous moment and the input vector at the current moment.

[0052] MEC establishes observation equations for the vehicle state observations of laser radar, camera, and millimeter-wave radar. For laser radar, the observation vector is the coordinates of the vehicle position in the point cloud; for camera, the observation vector is the pixel coordinates of the vehicle in the image; for millimeter-wave radar, the observation vector is the distance and angle of the vehicle relative to the radar. The observation equation maps the vehicle state vector to the observation space of each sensor.

[0053] At each time step, MEC first calculates the prior estimate of the vehicle state and its covariance matrix through the state equation based on the state estimate of the previous moment and the input of the current moment. Then, MEC extracts the observation vector of the vehicle at the current moment from the lidar, camera and millimeter wave radar, and calculates the predicted value of the observation value and its covariance matrix through the corresponding observation equation. MEC uses the Kalman gain formula to weight the error fusion of the prior estimate and the observation value, and updates the posterior estimate of the state vector and its covariance matrix. The Kalman gain determines the relative degree of trust in the prior estimate and the observation value based on the uncertainty of the two.

[0054] Finally, MEC uses the updated state vector as the optimal estimate of the vehicle's position, speed, and attitude at the current moment, and uses it for iterative calculations in the next time step. Through the Kalman filter algorithm, MEC integrates the information of multiple sensors, overcomes the limitations of a single sensor, and improves the accuracy and robustness of vehicle state estimation. The fused vehicle data provides a reliable input for subsequent trajectory prediction and multi-vehicle speed planning. At the same time, as a recursive algorithm, the computational efficiency of the Kalman filter can meet the needs of real-time processing and ensure the timeliness of the control system.

[0055] S23: Based on the fused vehicle data, MEC uses the trajectory prediction algorithm to predict the driving trajectory of each unmanned vehicle in the future. MEC establishes the state transfer equation based on the vehicle kinematic model. The state vector includes the position, velocity and acceleration of the vehicle, and the input vector is the acceleration change rate (i.e., jerk) of the vehicle. The state transfer equation describes the dynamic evolution relationship of the vehicle state between two consecutive time steps.

[0056] MEC uses the fused vehicle position, velocity and attitude as the initial state, the recent acceleration change rate as input, and recursively calculates the predicted value of the vehicle state in the next N time steps through the state transfer equation. This process can be regarded as an open-loop prediction based on the vehicle dynamics model.

[0057] In order to improve the prediction accuracy and take into account the observation information, MEC models the trajectory prediction problem as a hidden Markov process and uses the Kalman filter algorithm for closed-loop prediction. MEC sets the state transfer equation as the state prediction equation, takes the fused vehicle state as the observation value, and establishes the observation equation. At each time step, MEC first calculates the prior estimate of the state and its covariance matrix through the state prediction equation based on the state estimate at the previous moment and the input at the current moment. Then, MEC takes the fused vehicle state at the current moment as the observation value, and calculates the predicted value of the observation value and its covariance matrix through the observation equation.

[0058] MEC uses the Kalman gain formula to weight the error fusion of the prior estimate and the observed value, and updates the posterior estimate of the state vector and its covariance matrix. This step uses real-time observation information to correct the predicted value, improving the accuracy of the prediction. MEC uses the updated state vector as the optimal estimate of the vehicle state at the current moment and uses it for iterative prediction of the next time step. Steps 4-7 are executed repeatedly until the vehicle state sequence for the next N time steps is predicted.

[0059] Since directly fitting the state sequence at discrete time points may produce a jagged trajectory, MEC uses cubic spline curves to smooth the predicted position sequence. Spline curves generate continuous and smooth trajectory curves by constructing low-order polynomials between discrete points, which is closer to the actual motion characteristics of the vehicle.

[0060] Finally, MEC obtains the smooth trajectory curve of each unmanned vehicle in the next N time steps as the predicted trajectory of the vehicle at the intersection. MEC makes comprehensive use of vehicle kinematic models, Kalman filtering, spline curve fitting and other methods to accurately and smoothly predict the future trajectory of each unmanned vehicle at the intersection. Among them, the vehicle kinematic model provides prior knowledge and describes the dynamic evolution law of the vehicle state; the Kalman filter uses real-time observation information to correct the predicted value, improving the dynamic adaptability of the prediction; the spline curve fitting ensures the geometric continuity and smoothness of the trajectory, generating a more realistic driving path. The predicted vehicle trajectory provides a basis for subsequent multi-vehicle speed planning, which helps to improve the safety and efficiency of intersection traffic.

[0061] S24: Based on the predicted vehicle trajectory, MEC uses the model predictive control algorithm to solve the optimal speed control curve of each unmanned vehicle as a multi-vehicle speed control solution. For each unmanned vehicle, MEC constructs a state space equation based on the vehicle kinematic model as the state prediction model of model predictive control. The state vector includes the position, velocity and acceleration of the vehicle, and the input vector is the acceleration of the vehicle (i.e., the following vehicle acceleration). The state equation describes the dynamic evolution of the vehicle state under a given input.

[0062] MEC sets optimization goals and constraints based on road traffic rules, vehicle dynamics characteristics and safe driving requirements, and constructs a global optimization objective function for multi-vehicle speed control. The objective function usually includes the following three aspects: a. Minimize the change in speed of each vehicle: By penalizing drastic changes in speed, ensure the smoothness of vehicle movement and ride comfort. b. Minimize the change in acceleration of each vehicle: By penalizing drastic changes in acceleration, reduce frequent acceleration and deceleration operations and improve energy efficiency. c. Maintain a safe distance between vehicles: By setting vehicle spacing constraints, ensure that adjacent vehicles always maintain a safe interval distance to avoid collision risks. MEC constructs the state equation and global optimization objective function of each unmanned vehicle, as well as the constraints of intersection traffic flow (such as vehicle speed limit, traffic light timing, etc.) into an optimization problem for multi-vehicle speed control. The decision variable of this problem is the acceleration sequence of all unmanned vehicles in the next N time steps.

[0063] MEC adopts the cyclic prediction strategy in model predictive control to solve the optimization problem in each control cycle, including: Initialization: At the beginning of the first control cycle, MEC constructs the optimization problem of multi-vehicle speed control based on the initial state (position, speed and acceleration) of each unmanned vehicle and traffic environment information (such as road speed limit, initial distance between vehicles, etc.). The decision variable of this problem is the acceleration sequence of all unmanned vehicles in the next N time steps, and the objective function and constraints are in the form described above.

[0064] Loop prediction: In each control cycle, MEC performs the following steps: Update initial conditions: MEC uses the state of each vehicle (position, speed and acceleration) obtained by optimization in the previous cycle as the initial conditions of the optimization problem in the current cycle. In this way, the control effect of the previous cycle can be transferred to the current cycle, ensuring the continuity of control. Update constraints: MEC updates the constraints of the optimization problem based on the latest information of the current traffic environment, such as changes in road speed limits, the appearance of new obstacles, etc. This allows control decisions to adapt to dynamically changing traffic conditions. Construct a sequential quadratic programming problem: MEC converts the updated optimization problem into a sequential quadratic programming problem. Specifically, MEC represents the objective function as a quadratic form with respect to the decision variable (i.e., the acceleration sequence for the next N steps), and its general form is: , where X is the decision variable vector, Q is the quadratic term matrix (positive definite), c is the linear term vector, and T represents the transpose. At the same time, MEC expresses constraints such as state equations, speed limits, acceleration limits, and vehicle spacing limits as linear inequalities or equations about decision variables, and its general form is: , (inequality constraint), (equality constraint), where A and is the constraint matrix, b and is a constraint vector, and <= indicates that the components are not equal. Solving the sequential quadratic programming problem: MEC uses the sequential quadratic programming algorithm to solve the above problem and obtain the optimal acceleration sequence of each unmanned vehicle in the current control cycle. Commonly used algorithms include the interior point method and the effective set method.

[0065] The interior point method constructs a barrier function ,in, is the barrier parameter, is the row of matrix A; the inequality constraints are converted into equality constraints, and then the optimal solution is iteratively solved using Newton's method and other methods.

[0066] The active set method performs an iterative search on active constraints (constraints where equality holds) and continuously updates the working set (the set of active constraints) until the optimality condition is met. The optimal acceleration sequence obtained by solving the problem minimizes the objective function while satisfying the constraints, that is, it strikes a balance between vehicle stability, energy consumption, and safety.

[0067] Execution control: MEC uses the first element of the optimal acceleration sequence as the control instruction of each unmanned vehicle at the current moment and sends it to the on-board controller for execution. The vehicle adjusts its speed according to the received acceleration instruction to achieve collaborative control. Rolling optimization: When the next control cycle arrives, MEC uses the remaining elements of the optimal acceleration sequence as the initial guess value for the new round of optimization and repeats. In this way, the control decision can be updated and optimized in real time based on the latest state feedback and environmental information, forming a closed-loop control process based on the rolling time domain.

[0068] Through the cyclic prediction strategy and sequential quadratic programming algorithm, MEC can quickly generate the optimal multi-vehicle cooperative control plan in each control cycle and dynamically adjust the plan based on real-time feedback. This method comprehensively considers multiple performance indicators of vehicle movement, comprehensively optimizes the behavior of vehicles in the future, and improves the comprehensive benefits of intersection traffic. At the same time, cyclic prediction enables control decisions to adapt to the dynamic changes of the traffic environment, and the efficient solution of the sequential quadratic programming algorithm ensures the real-time nature of control instructions, thereby constructing a robust, efficient, and intelligent intersection multi-vehicle cooperative control system.

[0069] After obtaining the optimal control sequence in each control cycle, MEC applies it to vehicle control and repeats the optimization process in the next cycle, finally obtaining a complete multi-vehicle speed control solution. Specifically, the issuance and execution of the control instructions of the current cycle: MEC extracts the first element from the optimal acceleration sequence of the current cycle as the acceleration control instruction of each unmanned vehicle at the current moment. MEC sends the control instructions to the corresponding unmanned vehicles through V2X communication. The instruction content includes vehicle ID, timestamp and acceleration value. After receiving the instruction, the on-board controller of each unmanned vehicle inputs the acceleration value to the underlying actuator (such as throttle, brake, etc.) to adjust the actual speed of the vehicle. The vehicle updates its own status (position, speed, acceleration) according to the results of the instruction execution, and feeds back the status to MEC through V2X communication for optimization in the next cycle.

[0070] Construction of the optimization problem for the next cycle: MEC uses the remaining elements of the optimal acceleration sequence of the current cycle as the initial guess value of the optimization problem for the next cycle. This is equivalent to providing a good starting point for the optimization solution, which helps to speed up the convergence. MEC collects the latest status information fed back by all unmanned vehicles, as well as real-time data on the traffic environment at the intersection (such as road conditions, traffic flow, etc.), and updates the initial conditions and constraints of the optimization problem. MEC preprocesses the updated optimization problem and converts it into a standard sequential quadratic programming form to facilitate subsequent algorithm solutions.

[0071] Solving the optimal control sequence for the next cycle: MEC calls the sequential quadratic programming algorithm to solve the updated optimization problem and obtain the optimal acceleration sequence for the next cycle. The sequential quadratic programming algorithm uses iterative optimization to continuously reduce the value of the objective function while satisfying the constraints until the optimal solution is reached or the termination condition is met. During the solution process, the algorithm uses the optimal solution of the previous cycle as the initial value and performs local search on this basis, which improves the optimization efficiency and numerical stability.

[0072] Implementation of rolling horizon control: MEC repeats steps 1-3 to implement model predictive control based on rolling horizon. In each control cycle, MEC updates the optimization problem based on the latest state feedback and environmental information, solves a new optimal control sequence, and applies it to vehicle control. Through continuous rolling updates and optimizations, control decisions can adapt to the dynamic changes of the traffic environment, realizing real-time, closed-loop multi-vehicle collaborative control.

[0073] Generation of speed control curve: After multiple control cycles, MEC obtains the optimal acceleration sequence of each unmanned vehicle during the entire process of passing through the intersection. MEC numerically integrates the optimal acceleration sequence of each vehicle with its initial speed to obtain the speed value of the vehicle at different times and form a speed control curve. The speed control curve reflects the speed change law of the unmanned vehicle in the intersection, reflecting different stages such as acceleration, deceleration and constant speed driving.

[0074] Integration of multi-vehicle speed control schemes: The speed control curve of each unmanned vehicle is represented in the form of a time series, that is, the target speed value of the vehicle is recorded at discrete time points. The speed control curve can be represented by a structured data object, which contains attributes such as vehicle ID, timestamp sequence, and speed sequence. MEC uses a unified data format and encoding scheme to store and transmit speed control curves to ensure data consistency and interoperability.

[0075] Spatiotemporal coordinated optimization of multi-vehicle speed control curves: When integrating multi-vehicle speed control solutions, MEC needs to consider the mutual relationships and constraints of vehicles in the time and space dimensions. In the time dimension, MEC optimizes the speed distribution of vehicles at different times so that they can form a smooth and continuous speed trajectory in the intersection area, avoiding uncomfortable behaviors such as sudden acceleration and deceleration. In the spatial dimension, MEC constrains the safe distance and conflict area between vehicles to ensure that vehicles will not collide or deadlock when passing through intersections. MEC uses multi-objective optimization algorithms, such as weighted sum method and Pareto optimality, to weigh multiple performance indicators such as vehicle motion stability, energy consumption and traffic efficiency, and generate a globally optimal speed control solution.

[0076] Rolling time domain optimization and real-time feedback control: In view of the dynamic variability of the traffic environment at intersections, MEC adopts a rolling time domain optimization strategy to regularly update the multi-vehicle speed control scheme to adapt to real-time traffic conditions. In each control cycle, MEC recalculates the optimal speed control curve for a period of time in the future based on the current state of the vehicle and environmental perception data, and sends it to the vehicle for execution. Based on the received speed control curve, the vehicle combines its own local perception and planning to generate real-time speed control instructions, and executes them through the underlying controller. While executing the speed control instructions, the vehicle continuously feeds back its own status and execution results to MEC, forming a closed-loop control system. MEC evaluates the control effect based on this and makes optimization adjustments in the next cycle.

[0077] Vehicle-road collaborative control based on V2X communication: V2X communication technology is used to realize real-time and reliable data interaction between MEC, roadside unit RSU and on-board unit OBU. RSU distributes the multi-vehicle speed control scheme generated by MEC to all unmanned vehicles in the intersection area by broadcasting or multicasting, avoiding the communication overhead of unicast one by one. After receiving the speed control scheme, OBU extracts the speed control curve related to itself and passes it to the on-board controller to guide the movement of the vehicle. OBU can also report the real-time status of the vehicle, environmental perception data, etc. to RSU and MEC for use by the optimization algorithm to realize two-way information interaction between the vehicle and the road.

[0078] S3, broadcast the multi-vehicle speed control scheme to the unmanned vehicles at the intersection through the roadside communication RSU, including: S31: The roadside communication RSU extracts the identification information of each unmanned vehicle in the multi-vehicle speed control scheme and the corresponding optimal speed control curve. The RSU receives the multi-vehicle speed control master plan from the MEC, which contains the speed control information of all unmanned vehicles at the intersection. The RSU parses the data structure of the master plan and extracts the unique identifier of each unmanned vehicle (such as vehicle VIN code, temporary ID, etc.) and the corresponding optimal speed control curve. The optimal speed control curve is represented in the form of a time series, and each time point corresponds to a speed value. The RSU discretizes the curve into a series of time-speed pairs to facilitate subsequent data processing and transmission.

[0079] S32: Roadside communication RSU associates the corresponding optimal speed control curve with the communication address of each unmanned vehicle according to the identification information of the vehicle, and generates a speed control instruction for the vehicle. RSU maintains a registry of unmanned vehicles at the intersection, which records the mapping relationship between the identifier and the communication address (such as IP address, MAC address, etc.) of each vehicle. RSU queries the corresponding communication address in the registry according to the vehicle identifier extracted in S31. RSU associates the optimal speed control curve of each vehicle with its communication address and generates a speed control instruction. The instruction content includes vehicle identifier, timestamp, speed sequence, etc. RSU encodes and packages the speed control instruction to generate a data frame format suitable for wireless transmission, such as WSMP (WAVE Short Message Protocol) frame of IEEE 802.11p.

[0080] S33: Roadside communication RSU broadcasts the speed control instructions of each unmanned vehicle to the corresponding unmanned vehicle at the intersection through the communication protocol. RSU uses vehicle networking communication technologies such as DSRC (Dedicated Short Range Communication) or C-V2X (Cellular Vehicle-to-Everything) to establish a wireless connection with the unmanned vehicles in the intersection. RSU encapsulates the speed control instructions of each unmanned vehicle into the corresponding data frame according to the communication address of each unmanned vehicle, and adds the broadcast destination address to indicate that the instruction is applicable to all unmanned vehicles. RSU sends the data frame of the speed control instruction to the unmanned vehicle at the intersection through the broadcast channel of the vehicle networking. The broadcast method can improve communication efficiency and avoid the overhead of unicast one by one. The on-board communication equipment of the unmanned vehicle (such as OBU) monitors the broadcast channel, and decodes and parses the speed control instruction frame sent by the RSU after receiving it. The unmanned vehicle searches for the speed control curve that matches itself in the received instruction based on its own identifier and extracts the corresponding time-speed pair sequence. The unmanned vehicle passes the extracted speed control curve to the vehicle control system as a speed reference trajectory for a period of time in the future to guide the longitudinal movement of the vehicle.

[0081] The roadside communication RSU acts as an information exchange bridge between MEC and unmanned vehicles, decomposing the centrally optimized multi-vehicle speed control scheme into speed control instructions for a single vehicle, and broadcasting it to the vehicle in real time through the communication protocol of the Internet of Vehicles. This vehicle-road collaborative control architecture can give full play to the computing power of MEC and the communication capability of RSU to achieve refined control of intersection traffic. At the same time, the broadcast communication method ensures the real-time and reliability of the control instructions, enabling all vehicles to synchronously execute the optimized speed curve, thereby ensuring the safety and traffic efficiency of multiple vehicles at the intersection.

[0082] S4, the on-board communication OBU of the unmanned vehicle receives the multi-vehicle speed control scheme broadcast by the RSU, and the on-board controller of the unmanned vehicle controls the unmanned vehicle according to the multi-vehicle speed control scheme and the environmental data within the preset range of the unmanned vehicle obtained by the on-board sensing device, including: S41, the on-board OBU receives the multi-vehicle speed control scheme data frame sent by the RSU by monitoring the V2X broadcast channel. The OBU decodes and performs integrity check on the received data frame to extract the complete multi-vehicle speed control scheme data packet. The OBU transmits the received speed control scheme data packet to the on-board controller ECU via the on-board Ethernet or other in-vehicle communication buses (such as CAN, FlexRay, etc.).

[0083] S42, after receiving the multi-vehicle speed control scheme data packet transmitted by the OBU, the on-board controller ECU parses it. The ECU searches for the optimal speed control curve corresponding to it in the scheme data according to its own vehicle identifier (such as VIN code). The ECU extracts the discretized speed sequence of the vehicle in the future as the global speed constraint condition.

[0084] S43, unmanned vehicles are equipped with a variety of on-board sensing devices, such as laser radar, millimeter-wave radar, cameras, etc., which are used to collect environmental information within a certain range around the vehicle. The sensing devices exchange data with the on-board controller ECU through the CAN bus. Each device occupies a specific CAN bus ID and data frame format according to its data characteristics. As the master node of the CAN bus, the ECU monitors and receives the environmental data frames sent by each sensing device by configuring the corresponding receiving filter. The ECU parses the received environmental data frames, extracts the location, speed, type and other information of the obstacles, and builds an environmental perception model around the vehicle.

[0085] In S44, the ECU uses the optimal speed control curve extracted in S42 as the global speed constraint condition to set the upper and lower bounds of the vehicle's speed in the future. The ECU uses the environmental perception data obtained in S43 as the local constraint condition, taking into account factors such as the safe distance between the vehicle and obstacles and the risk of collision. The ECU runs local speed planning algorithms, such as A* search, dynamic planning, etc., to generate the optimal speed control instruction sequence for the current scenario under the global speed constraint and local environmental constraint. The local speed planning algorithm searches and optimizes in the speed-time-position space to find a speed trajectory that meets the constraints, is smooth, safe, and has the smallest deviation from the global speed curve. The ECU sends the generated local optimal speed control instruction sequence to the lower-level actuators, such as the throttle, brake, steering, etc., to control the actual movement of the vehicle.

[0086] S45, after receiving the local optimal speed control instruction sequence sent by the ECU, the vehicle's actuator converts it into specific throttle, brake, steering and other control signals. The actuator adjusts the vehicle's acceleration, deceleration, steering angle and other motion parameters according to the instructions of the control signal to achieve precise control of the vehicle's speed and direction. During driving, the vehicle continues to receive the latest speed control instructions sent by the ECU and makes corresponding motion adjustments in a timely manner to adapt to changes in traffic conditions at the intersection.

[0087] The vehicle continuously executes local speed control instructions, and follows the global optimal speed curve as much as possible while meeting safety and comfort requirements until it passes the intersection smoothly. Unmanned vehicles can receive and execute multi-vehicle speed control plans issued by roadside facilities, and generate local optimal speed control instructions based on their own real-time perception data. This hierarchical control architecture that combines global path planning with local trajectory generation can not only follow the macro-dispatching of the traffic management center, but also flexibly respond to dynamic road conditions, thereby ensuring vehicle safety while improving the overall efficiency of intersection traffic. At the same time, through vehicle-road collaboration and multi-vehicle collaboration, unmanned vehicles can form real-time information interaction and decision-making coordination with surrounding vehicles and infrastructure.

Claims

1. A method for centralized speed control of an unmanned vehicle, characterized in that: include: S1, obtaining environmental data at the intersection through a roadside sensing device, wherein the roadside sensing device includes a laser radar, a camera, and a millimeter-wave radar; the environmental data includes the position, speed, and posture of the vehicle; S2, based on the environmental data, multi-sensor data fusion and calculation are performed through the roadside computing unit to obtain the optimal speed control curve combination of all unmanned vehicles to generate a multi-vehicle speed control plan at the intersection: According to the predicted trajectory, a kinematic model of each unmanned vehicle is established, and a global optimization objective function is constructed, wherein the global optimization objective function includes minimizing the speed change, minimizing the acceleration change of each vehicle, and the preset safety distance between vehicles; For each unmanned vehicle, the speed and acceleration are used as control variables, the position is used as the state variable, the corresponding kinematic model is used as the state equation and constraint condition, and the global optimization objective function is used as the performance indicator function to construct the model predictive control problem of the corresponding vehicle; Using the model predictive control algorithm, through the cyclic prediction method, in each control cycle, the model predictive control problem of the vehicle is solved to obtain the optimal speed control sequence of the corresponding vehicle in the next N time steps: in each control cycle, for each unmanned vehicle, according to the corresponding kinematic model, the state change of the corresponding vehicle in the next N time steps is predicted, and the state change includes the change of position, speed and acceleration; using the predicted state change in the next N time steps as the initial condition, the speed and acceleration of the corresponding vehicle as the control variables, and the global optimization objective function as the performance indicator function, the model predictive control problem of the corresponding vehicle is converted into a sequential quadratic programming problem; using the sequential quadratic programming algorithm, the sequential quadratic programming problem is solved to obtain the optimal speed control sequence and the optimal acceleration control sequence of the corresponding vehicle in the next N time steps; The optimal speed control sequence of the future N time steps obtained by each unmanned vehicle in each control cycle is connected to form the optimal speed control curve of each vehicle; S3, broadcast the multi-vehicle speed control scheme to the unmanned vehicles at the intersection through the roadside communication RSU; S4, the on-board communication OBU of the unmanned vehicle receives the multi-vehicle speed control scheme broadcast by the RSU, and the on-board controller of the unmanned vehicle controls the unmanned vehicle according to the multi-vehicle speed control scheme combined with the environmental data within the preset range of the unmanned vehicle obtained by the on-board sensing device.

2. The method for centralized speed control of an unmanned vehicle according to claim 1, characterized in that: S2, through the roadside computing unit to perform multi-sensor data fusion and calculation, to obtain a multi-vehicle speed control scheme, including: S21, performing spatiotemporal alignment and coordinate conversion on the environmental data by a roadside computing unit to obtain multi-sensor data in the same coordinate system, wherein the multi-sensor data includes laser radar point cloud data, camera image data, and millimeter wave radar data; S22, fusing the multi-sensor data in the same coordinate system through a Kalman filter algorithm to obtain fused vehicle data, where the vehicle data includes vehicle position, speed, and posture; S23, predicting the trajectory of the unmanned vehicle at the intersection through a trajectory prediction algorithm based on the fused vehicle data; S24, according to the predicted trajectory, the optimal speed control curve of each unmanned vehicle at the intersection is calculated through the model predictive control algorithm as a multi-vehicle speed control solution.

3. The method for centralized speed control of an unmanned vehicle according to claim 2, characterized in that: S22, fusing the multi-sensor data in the same coordinate system through the Kalman filter algorithm to obtain fused vehicle data, including: The laser radar point cloud data, camera image data and millimeter wave radar data in the same coordinate system are used as input, and the vehicle state equation and observation equation are established by using the Kalman filter algorithm; the vehicle state equation describes the predicted values ​​of the vehicle position, speed and posture, and the observation equation describes the observation values ​​of the laser radar, camera and millimeter wave radar on the vehicle state; The Kalman filter algorithm is used to predict and update the vehicle status to obtain fused vehicle data.

4. The method for centralized speed control of an unmanned vehicle according to claim 2, characterized in that: S23, predicting the trajectory of the unmanned vehicle at the intersection through a trajectory prediction algorithm based on the fused vehicle data, including: The fused vehicle data is used as input, and the constructed trajectory prediction model is used to predict the trajectory of the vehicle in the future based on the historical trajectory and current status of the vehicle; The predicted trajectory is smoothed using the B-spline curve fitting method and used as the trajectory of the unmanned vehicle at the intersection.

5. The method for centralized speed control of an unmanned vehicle according to claim 4, characterized in that: The constructed trajectory prediction model is used to predict the trajectory of the vehicle in the future based on the historical trajectory and current status of the vehicle, including: Establish a vehicle kinematics model to describe the changing relationship between the vehicle's position, speed and posture; the vehicle kinematics model includes the state equations of the vehicle's position, speed and posture; The vehicle kinematic model is used as the state equation of the Kalman filter algorithm, and the fused vehicle position, speed and posture are used as the observation equation to construct a trajectory prediction model. The trajectory prediction model is used to predict the trajectory of the vehicle in the future based on the historical trajectory and current state of the vehicle. The predicted value of the vehicle trajectory is recursively calculated through the prediction and update process of the Kalman filter algorithm.

6. The method for centralized speed control of an unmanned vehicle according to claim 2, characterized in that: S3, broadcasting the multi-vehicle speed control scheme to the unmanned vehicles at the intersection through the roadside communication RSU, including: S31, the roadside communication RSU extracts the identification information of each unmanned vehicle in the multi-vehicle speed control scheme and the corresponding optimal speed control curve; S32, the roadside communication RSU associates the corresponding optimal speed control curve with the communication address of each unmanned vehicle according to the identification information of the vehicle, and generates a speed control instruction for the vehicle; S33, the roadside communication RSU broadcasts the speed control instruction of each unmanned vehicle to the corresponding unmanned vehicle at the intersection through the communication protocol.

7. The method for centralized speed control of an unmanned vehicle according to claim 6, characterized in that: S4, the on-board controller of the unmanned vehicle controls the unmanned vehicle according to the multi-vehicle speed control scheme and the environmental data within the preset range of the unmanned vehicle obtained by the on-board sensing device, including: S41, the on-board communication OBU of the unmanned vehicle receives the multi-vehicle speed control scheme broadcast by the RSU, and transmits the multi-vehicle speed control scheme to the on-board controller; S42, the vehicle controller extracts the optimal speed control curve from the multi-vehicle speed control scheme; S43, the vehicle controller obtains environmental data within a preset range of the unmanned vehicle collected by the vehicle sensing device through the CAN bus; S44, the vehicle controller uses the extracted optimal speed control curve as a global constraint for vehicle speed planning, combines the collected environmental data, and generates an optimal speed control instruction for the unmanned vehicle in the current scenario through a local speed planning algorithm; S45, controlling the unmanned vehicle to travel according to the optimal speed control instruction until the vehicle passes the intersection.

8. A centralized speed control system for unmanned vehicles, characterized in that: include: At least one processing unit; used to execute instructions to implement the unmanned vehicle speed centralized control method as described in any one of claims 1 to 7.

Images