A safe collaborative distribution control method, system and device for unmanned flow vehicles

Abstract

The application discloses a safe collaborative distribution control method, system and device of an unmanned logistics vehicle, relates to the field of logistics distribution, and comprises the following steps: determining a task state set and a task scheduling result; constructing a dynamic risk field; constructing a local traffic interaction state set, determining a reference path, and then obtaining a predicted state sequence; generating a candidate collaborative strategy based on the reference path, the dynamic risk field, the predicted state sequence and the traffic strategy of other unmanned logistics vehicles; reserving the candidate collaborative strategy corresponding to the safety constraint value less than or equal to zero; determining a target traffic strategy based on the global unified collaborative constraint boundary and the reserved candidate collaborative strategy; generating a plurality of groups of candidate control input sequences, determining a predicted candidate trajectory, then determining a safe trajectory, and obtaining a final control input; determining whether to trigger a safety intervention mechanism during the completion of the distribution task of the unmanned logistics vehicle; and generating an emergency control instruction when the safety intervention mechanism is triggered. The application can realize safe collaborative distribution of the unmanned logistics vehicle.

Description

Technical Field

[0001] This application relates to the field of logistics and distribution, and in particular to a safe collaborative distribution control method, system and device for unmanned logistics vehicles. Background Technology

[0002] With the rapid development of smart logistics and intelligent driving technologies, unmanned logistics vehicles, as an important carrier for urban last-mile delivery, are gradually expanding from low-complexity scenarios such as closed parks and campuses to public urban road environments. Compared to closed scenarios, public urban road environments are characterized by a complex mix of traffic participants, frequent dynamic changes, strict traffic rules, and high safety requirements, placing higher demands on the environmental perception, collaborative decision-making, and real-time safety control capabilities of unmanned logistics vehicles. Currently, existing research and industrial practices mainly focus on single-vehicle autonomous navigation, local path planning, and basic obstacle avoidance control, and have achieved some progress. However, achieving multi-vehicle collaboration and safe, efficient delivery under public road conditions still faces many technical bottlenecks.

[0003] I. Insufficient Multi-Vehicle Collaborative Decision-Making Capabilities. Currently, most unmanned logistics vehicle systems primarily adopt a single-vehicle independent decision-making mode, where each vehicle independently completes path planning, obstacle avoidance, and motion control based on environmental information obtained from its own sensors. This mode can meet basic operational needs when the number of vehicles is small and the traffic environment is relatively simple. However, in urban public road environments, multiple unmanned logistics vehicles often operate simultaneously and share road resources, inevitably leading to interactions and conflicts between vehicles.

[0004] For example, at unsignalized intersections, multiple unmanned logistics vehicles may arrive at the conflict zone simultaneously. Without a unified collaborative decision-making mechanism, problems such as mutual waiting, decision-making deadlock, and even conflicts over right-of-way can easily arise, severely impacting delivery efficiency and operational safety. Similarly, in scenarios such as vehicles meeting on narrow side roads, unmanned logistics vehicles merging into main roads, and multiple vehicles making U-turns near loading and unloading points, traditional single-vehicle independent decision-making methods struggle to achieve globally optimal coordination, easily leading to traffic congestion or local deadlocks. Some existing research attempts to uniformly allocate paths or right-of-way through centralized scheduling platforms, but these methods heavily rely on stable communication and central computing resources. As the vehicle scale increases, computational delays and scheduling bottlenecks can easily occur, making it difficult to meet the real-time collaborative needs of public roads.

[0005] Second, safety control methods lack long-term time-domain assurance capabilities. Current unmanned logistics vehicles typically employ classic control methods such as Model Predictive Control (MPC), Proportional-Integral-Derivative (PID) control, or Pure Pursuit control. While these methods can optimize vehicle trajectory and satisfy certain motion constraints within a short prediction window, their safety assurance capabilities are usually limited to a finite prediction time domain.

[0006] Specifically, traditional model predictive control typically considers only the obstacle positions and vehicle states within the next few seconds, ensuring that the vehicle does not collide in the short term. However, in complex dynamic environments on public roads, such as when pedestrians suddenly cross the road, other vehicles change lanes, or road construction areas are temporarily closed, vehicles may remain safe in the short term, but gradually enter an irreversible dangerous state over a longer timescale, such as entering narrow blind spots, being surrounded by dynamic obstacles, or getting stuck in areas that cannot be bypassed. Furthermore, traditional control methods are usually based on local optima, making them susceptible to non-convex optimization problems, leading to trajectory oscillations, frequent sudden braking, or even getting stuck in local optima, making it difficult to guarantee continuous and stable operation during delivery on public roads.

[0007] III. Insufficient Prediction and Risk Assessment Capabilities for Dynamic Traffic Participants. Urban public road environments contain a large number of dynamic traffic participants, including pedestrians, bicycles, electric vehicles, motor vehicles, and other unmanned logistics vehicles. These targets are characterized by diverse motion patterns and strong randomness in behavior, posing a significant challenge to the real-time risk perception and prediction capabilities of unmanned logistics vehicles. Existing technologies typically employ Kalman filtering, extended Kalman filtering, or simple trajectory extrapolation models for state estimation and short-term prediction of dynamic targets. These methods are effective under conditions of uniform target motion, but their prediction accuracy is low for targets with sudden turns, accelerations / decelerations, or complex interactive behaviors, making it difficult to reflect potential collision risks in a timely manner.

[0008] For example, near pedestrian crossings in cities, pedestrians may change direction at any time; at intersections of non-motorized vehicle lanes, electric vehicles may quickly cut into the path of vehicles. If risk assessment is insufficient, unmanned logistics vehicles are prone to problems such as braking too late, failing to avoid obstacles, or parking too conservatively, thereby affecting delivery safety and traffic efficiency.

[0009] Fourth, the delivery task scheduling and vehicle motion control are disconnected. Existing unmanned logistics delivery systems typically separate task scheduling and vehicle control into two independent levels. The upper-level scheduling system is responsible for order allocation, delivery route generation, and task priority management, while the lower-level vehicle control system is responsible for local navigation and obstacle avoidance. Because these two systems lack real-time information exchange and collaborative optimization mechanisms, overall delivery efficiency is easily reduced.

[0010] When a delivery platform temporarily inserts a high-priority order, existing systems often can only adjust by replanning the global route, failing to dynamically optimize by comprehensively considering current road traffic conditions, multi-vehicle collaboration relationships, and vehicle operational risks. For example, when an unmanned delivery vehicle slows down due to road congestion, other vehicles cannot adjust their collaboration strategies in time, easily leading to an imbalance in delivery resource allocation. This scheduling-control separation architecture limits the system's responsiveness to dynamic delivery tasks, making it difficult to meet the complex and ever-changing business needs of urban public road terminal delivery.

[0011] Fifth, there is a lack of multi-layered safety redundancy mechanisms for public roads. Urban public roads are open transportation systems, requiring unmanned logistics vehicles to have near-zero collision, zero violation, and zero loss of control safety capabilities. However, existing control systems mainly achieve passive safety protection through empirical rules, local obstacle detection, and simple emergency braking mechanisms, lacking a systematic active safety assurance framework.

[0012] Currently, some unmanned logistics vehicles only trigger braking when the distance to obstacles is below a threshold. This reactive strategy makes it difficult to identify potential risks in complex conflict scenarios in advance. Furthermore, for multi-vehicle collaborative delivery tasks, there is a lack of shared safety information and risk coordination mechanisms among the vehicles, hindering group-level safety optimization. Summary of the Invention

[0013] The purpose of this application is to provide a safe collaborative delivery control method, system and device for unmanned logistics vehicles, which solves the problems of insufficient collaboration, insufficient safety and insufficient dynamic response capability in the delivery process of unmanned logistics vehicles, and can realize safe collaborative delivery of unmanned logistics vehicles.

[0014] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a safe collaborative delivery control method for unmanned logistics vehicles, including: Obtain delivery tasks; determine a task status set based on the delivery tasks and the vehicle's own status information; use a dynamic task value function to determine the task scheduling result based on the task status set; Real-time acquisition of environmental information; construction of a dynamic risk field based on environmental information; A local traffic interaction state set is constructed based on the task state set and the dynamic risk field; a reference path is determined based on the task scheduling results; a predicted state sequence is obtained based on the local traffic interaction state set and the reference path; and candidate cooperative strategies are generated based on the reference path, the dynamic risk field, the predicted state sequence, and the traffic strategies of other unmanned logistics vehicles. A local security constraint value function is used to determine the security constraint value based on candidate cooperative strategies; candidate cooperative strategies corresponding to security constraint values ​​that are less than or equal to zero are retained. The target passage strategy is determined based on the globally unified collaborative constraint boundary and the retained candidate collaborative strategies; Multiple sets of candidate control input sequences are generated based on local traffic interaction states, target traffic strategies, vehicle dynamics constraints, and reference paths. The predicted candidate trajectory is obtained based on the candidate control input sequence using the vehicle dynamics state transition function. A control layer safety filtering function is used to filter the predicted candidate trajectories and determine the safe trajectory; A trajectory optimization objective function is adopted, and the final control input is obtained based on a safe trajectory; the unmanned logistics vehicle executes the final control input until the delivery task is completed. During the delivery process of the unmanned logistics vehicle, it is determined whether to trigger the safety intervention mechanism based on the vehicle's own status information and dynamic risk field; when the safety intervention mechanism is triggered, an emergency control command is generated; the unmanned logistics vehicle executes the emergency control command.

[0015] Secondly, this application provides a safe collaborative delivery control system for unmanned logistics vehicles, used to implement the aforementioned safe collaborative delivery control method for unmanned logistics vehicles, comprising: a delivery task scheduling module, an environmental perception and dynamic risk assessment module, a multi-vehicle collaborative decision-making module, and a safe trajectory optimization control module connected in sequence, as well as a safety supervision and redundancy control module; the safety supervision and redundancy control module is connected to the environmental perception and dynamic risk assessment module. The task scheduling module is used to acquire delivery tasks; the task scheduling module is also used to determine a task status set based on the delivery tasks and the vehicle's own status information; the task scheduling module is also used to determine the task scheduling result based on the task status set using a dynamic task value function; The environmental perception and dynamic risk assessment module is used to acquire environmental information in real time; the environmental perception and dynamic risk assessment module is also used to construct a dynamic risk field based on the environmental information. The multi-vehicle collaborative decision-making module is used to construct a local traffic interaction state set based on a task state set and a dynamic risk field; the multi-vehicle collaborative decision-making module is also used to determine a reference path based on task scheduling results; the multi-vehicle collaborative decision-making module is also used to obtain a predicted state sequence based on the local traffic interaction state set and the reference path; the multi-vehicle collaborative decision-making module is also used to generate candidate collaborative strategies based on the reference path, the dynamic risk field, the predicted state sequence, and the passage strategies of other unmanned logistics vehicles; the multi-vehicle collaborative decision-making module is also used to determine a safety constraint value based on the candidate collaborative strategies using a local safety constraint value function; the multi-vehicle collaborative decision-making module is also used to retain candidate collaborative strategies corresponding to safety constraint values ​​less than or equal to zero; the multi-vehicle collaborative decision-making module is also used to determine a target passage strategy based on a globally unified collaborative constraint boundary and the retained candidate collaborative strategies; The safe trajectory optimization control module is used to generate multiple sets of candidate control input sequences based on local traffic interaction states, target traffic strategies, vehicle dynamics constraints, and reference paths. It is also used to obtain predicted candidate trajectories based on the candidate control input sequences using a vehicle dynamics state transition function. Furthermore, it uses a control layer safety screening function to filter the predicted candidate trajectories and determine the safe trajectory. Finally, it uses a trajectory optimization objective function to obtain the final control input based on the safe trajectory. The unmanned logistics vehicle executes the final control input until the delivery task is completed. The safety monitoring and redundancy control module is used to determine whether to trigger a safety intervention mechanism based on the vehicle's own status information and dynamic risk field during the unmanned logistics vehicle's delivery task. When the safety intervention mechanism is triggered, the safety monitoring and redundancy control module generates an emergency control command, and the unmanned logistics vehicle executes the emergency control command.

[0016] Thirdly, this application provides a safe collaborative delivery control device for unmanned logistics vehicles, including a task acquisition and scheduling control center and a safe collaborative delivery control system for multiple unmanned logistics vehicles; The task acquisition and scheduling control center is deployed on the delivery platform side; the safe collaborative delivery control system of the unmanned logistics vehicle is installed on the unmanned logistics vehicle; the task acquisition and scheduling control center is connected to the safe collaborative delivery control systems of multiple unmanned logistics vehicles; The task acquisition and scheduling control center is used to acquire order task information and vehicle status information of multiple unmanned logistics vehicles, generate delivery tasks, and transmit the delivery tasks to the corresponding unmanned logistics vehicle's safe collaborative delivery control system in real time. The safe and collaborative delivery control system of the unmanned logistics vehicle is used to complete the delivery task.

[0017] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a safe collaborative delivery control method, system, and device for unmanned logistics vehicles. By combining real-time acquired delivery tasks with the vehicle's own state information, a task state set is determined. A dynamic task value function is then used to determine the task scheduling result based on the task state set, enabling the vehicle to dynamically adjust its control strategy according to the delivery task and its own state, achieving unified optimization of delivery efficiency and operational safety. By acquiring environmental information in real-time and constructing a dynamic risk field, risk quantification analysis is performed on dynamic traffic participants around the unmanned logistics vehicle, improving the vehicle's perception and proactive risk avoidance capabilities in complex dynamic environments. By introducing the traffic strategies of other unmanned logistics vehicles during the decision-making process, candidate collaborative strategies are generated and subjected to local safety constraints and globally unified collaborative constraints to determine the target traffic strategy. This achieves state sharing and collaborative behavior optimization among unmanned logistics vehicles, as well as collaborative traffic control of multiple unmanned logistics vehicles. Under the premise of ensuring vehicle safety, optimal delivery efficiency is achieved, reducing path conflicts, vehicle waiting, and local congestion. By generating multiple candidate control input sequences based on local traffic interaction states, target traffic strategies, vehicle dynamics constraints, and reference paths, and determining the corresponding predicted candidate trajectories, a control-layer safety screening function is used to screen the predicted candidate trajectories to determine safe trajectories. Furthermore, a trajectory optimization objective function is employed to obtain the final control input based on the safe trajectory, achieving trajectory optimization control with long-time-domain safety constraints. This provides continuous safety assessment and dynamic constraints for the vehicle's future operation, enhancing its proactive safety capabilities. During the delivery process of the unmanned logistics vehicle, the system determines whether to trigger a safety intervention mechanism based on the vehicle's own state information and dynamic risk field. When a safety intervention mechanism is triggered, an emergency control command is generated, improving the robustness and reliability of the unmanned logistics vehicle in the face of sudden traffic events. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a safe collaborative delivery control method for unmanned logistics vehicles in one embodiment of this application; Figure 2 This is a schematic diagram of a safe collaborative delivery control system for an unmanned logistics vehicle, provided as an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The existing unmanned logistics delivery technology on public urban roads has the following main shortcomings: 1. Insufficient multi-vehicle coordination capability. Most existing unmanned logistics vehicles adopt an independent decision-making approach for each vehicle, lacking a collaborative control mechanism between vehicles. In scenarios such as intersections without traffic signals, narrow roads where vehicles meet each other, and merging into main roads, problems such as traffic conflicts, mutual waiting, and local congestion are likely to occur, affecting delivery efficiency.

[0022] 2. Limited safety control range. Traditional model predictive control methods can usually only guarantee safety within a limited prediction time domain, and it is difficult to constrain potential risks outside the prediction range. Vehicles are prone to entering an irreversible dangerous state in complex dynamic environments, affecting operational continuity and reliability.

[0023] 3. Insufficient dynamic risk identification capability. There are a large number of dynamic traffic participants such as pedestrians, non-motorized vehicles and social vehicles on urban public roads. Existing methods have limited ability to predict complex behavioral changes, making it difficult to identify potential collision risks in a timely manner, which can easily lead to delayed obstacle avoidance response or overly conservative control.

[0024] 4. Lack of linkage between delivery task scheduling and vehicle control. Existing systems typically handle order scheduling and vehicle motion control separately, making it difficult to dynamically adjust vehicle coordination behavior based on changes in delivery tasks and real-time traffic conditions, thus affecting overall delivery efficiency.

[0025] 5. Insufficient safety redundancy mechanisms. Existing unmanned logistics vehicles mainly rely on passive obstacle avoidance and emergency braking strategies, lacking the ability to predict risks throughout the entire process and actively control safety, making it difficult to meet the high safety requirements of open road terminal delivery.

[0026] Therefore, there is an urgent need for a comprehensive control system that integrates multi-vehicle collaborative decision-making, dynamic risk prediction, and safety trajectory optimization to improve the overall safety and reliability of unmanned logistics delivery systems on urban public roads.

[0027] To make the objectives, features and advantages of this application more apparent and understandable, the application will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0028] This application proposes a safe and collaborative delivery control method for unmanned logistics vehicles in urban open road terminal delivery scenarios. It is applicable to open delivery environments characterized by multiple vehicles operating simultaneously, dense dynamic traffic participants, complex road structures, and real-time task changes. The method revolves around the overall concept of task scheduling, environmental awareness, collaborative decision-making, safety control, and risk monitoring. It constructs a hierarchical control system with task response capabilities, autonomous collaborative capabilities, dynamic risk response capabilities, and safety redundancy assurance capabilities, enabling unmanned logistics vehicles to achieve continuous autonomous delivery in open road environments while ensuring traffic safety and task reliability.

[0029] In all embodiments provided in this application, the term "first" is used. An unmanned logistics vehicle (i.e., an unmanned logistics vehicle) This example illustrates the execution process of the safe collaborative delivery control method for unmanned logistics vehicles.

[0030] In one exemplary embodiment, such as Figure 1 As shown, a safe collaborative delivery control method for unmanned logistics vehicles is provided, including: Step S1: Obtain delivery tasks. Determine a task state set based on the delivery tasks and the vehicle's own status information. Use a dynamic task value function to determine the task scheduling result based on the task state set.

[0031] In step S1, delivery demand analysis and task-driven control are managed in a unified manner. The core objective is to transform the traditional static order scheduling into a dynamic task decision-making process that is deeply coupled with the vehicle's motion state, so that the vehicle can adaptively adjust its delivery behavior according to the real-time road environment and its own state.

[0032] After receiving a delivery task, the delivery task is integrated with the vehicle's (i.e., the unmanned logistics vehicle's) real-time status information to construct the unmanned logistics vehicle. The set of task states is represented as: ; in, Indicates unmanned logistics vehicle The set of task states; Indicates unmanned logistics vehicle The current status information, including current position coordinates, speed, acceleration and heading angle, is obtained in real time by the vehicle's own positioning unit, integrated navigation unit and vehicle status sensors; Indicates unmanned logistics vehicle The location of the target delivery point in the obtained delivery task; Indicates unmanned logistics vehicle The task time constraints in the acquired delivery task are used to describe the maximum allowed time to complete the order delivery requirements; Indicates unmanned logistics vehicle The urgency level of the acquired delivery tasks is used to reflect the priority of the delivery tasks; Indicates unmanned logistics vehicle The vehicle's current resource status, including remaining battery power, cargo capacity, and vehicle health status, is obtained in real time by the unmanned logistics vehicle's own battery management system, vehicle controller, and status monitoring unit.

[0033] To enable the task allocation process to adapt to the uncertainties of the open road environment, a dynamic task value function is constructed: ; in, This represents the overall cost function value of the delivery task; This indicates the estimated time to complete the task (i.e., time cost). This indicates the remaining distance (i.e., path length) from the vehicle's current location to the target delivery point. This indicates the energy consumption (i.e., energy cost) expected to be required to complete the current delivery task. This represents the overall risk coefficient (i.e., risk cost) corresponding to the task path. , , , These represent the weighting coefficients for time cost, path length, energy cost, and risk cost, respectively, used to adjust the importance of different items.

[0034] Using the aforementioned dynamic value function, the value of the unmanned logistics vehicle is calculated based on the task state set. Total cost of performing the current delivery task This cost is used to characterize the overall cost of a vehicle performing its current delivery task. The higher the value, the longer the vehicle takes to complete the current delivery task, the farther the distance, the higher the energy consumption, or the greater the route risk. The smaller the value, the lower the overall cost of the vehicle performing the corresponding delivery task and the higher the task matching degree.

[0035] according to The vehicle tasks are dynamically updated, and task scheduling results are generated. . Indicates unmanned logistics vehicle The task priority of the acquired delivery tasks is calculated using a weighted average based on the task's urgency, time constraints, and overall cost, and is used to characterize the task's execution priority.

[0036] Step S2: Acquire environmental information in real time. Construct a dynamic risk field based on the environmental information. The environmental information includes observation information on road structure, obstacle boundaries, and dynamic traffic participants. Based on this, the process of constructing a dynamic risk field based on environmental information includes: mapping the observation information to a global reference coordinate system and constructing the target state vector of the dynamic traffic participants. Using a risk field function, determine the current comprehensive risk value of the unmanned logistics vehicle based on the target state vector. Construct the dynamic risk field based on the current comprehensive risk value of the unmanned logistics vehicle.

[0037] Specifically, step S2 is used to establish a real-time dynamic traffic scene around the vehicle, which is the foundation for subsequent collaborative decision-making and safety control. It mainly relies on onboard LiDAR, integrated navigation unit, and vehicle status sensors to acquire environmental information. The onboard LiDAR continuously scans the surrounding road structure, obstacle boundaries, and dynamic traffic participants, generating a local point cloud map (i.e., observation information). Combined with the vehicle positioning results (determined by the integrated navigation unit), the observation information is uniformly mapped to the global reference coordinate system.

[0038] For the detected dynamic traffic participants, construct the target state vector: ; in, Indicates the dynamic traffic participants at the current moment of control. The target state vector; Represents the lateral and longitudinal position coordinates of dynamic traffic participants in the global reference coordinate system; Indicates the speed of dynamic traffic participants; The heading angle represents the heading angle of a dynamic traffic participant, used to describe the direction of movement of the dynamic traffic participant; It represents the acceleration of dynamic traffic participants and is used to characterize the speed change trend of dynamic traffic participants.

[0039] Unlike traditional methods that rely solely on instantaneous distance to assess risk, this embodiment introduces a continuous time-domain risk evolution modeling mechanism to predict the short-term trajectories of dynamic traffic participants and construct a risk field function: ; In the formula, Indicates the vehicle's current control status. The overall risk value; Indicates the vehicle's current control status. Predicted minimum spatial distance between dynamic traffic participants; Indicates the current control time. The estimated time to collision (Time To Collision) is used to assess the remaining time before a potential collision occurs. Indicates the current control moment The probability of dangerous behavior intentions of dynamic traffic participants is used to describe the likelihood of dynamic traffic participants engaging in dangerous behaviors such as sudden crossing, lane changing, or acceleration. , , These represent the distance to the risk item. Time risk items and behavioral risk items The weighting parameters are used to control the degree of influence of each risk factor on the total risk value assessment.

[0040] The calculated comprehensive risk value is embedded in the vehicle's surrounding environment in the form of spatial heat, forming a dynamic risk field (e.g., a dynamic risk map). The dynamic risk field can not only reflect the current level of danger, but also reveal potential risk trends in advance, such as complex behaviors like pedestrians suddenly crossing the road, bicycles changing lanes, or other vehicles cutting in.

[0041] By continuously updating the dynamic risk field based on real-time environmental information, vehicles can decelerate, avoid danger, or adjust their routes in advance before a real danger occurs, thereby significantly improving proactive safety capabilities in open road delivery.

[0042] Step S3: Construct a local traffic interaction state set based on the task state set and the dynamic risk field. Determine a reference path based on the task scheduling results. Obtain a predicted state sequence based on the local traffic interaction state set and the reference path. Generate candidate cooperative strategies based on the reference path, the dynamic risk field, the predicted state sequence, and the traffic strategies of other unmanned logistics vehicles. Use a local safety constraint value function to determine safety constraint values ​​based on the candidate cooperative strategies. Retain candidate cooperative strategies corresponding to safety constraint values ​​less than or equal to zero. Determine the target traffic strategy based on the globally unified cooperative constraint boundary and the retained candidate cooperative strategies.

[0043] In one embodiment, in order to solve the problem of coordinated traffic control of multiple unmanned logistics vehicles in typical conflict scenarios such as intersections, narrow road encounters, merging lanes, and temporary parking to avoid collisions in urban public road environments, this embodiment introduces a distributed safety decision-making algorithm based on zero-constraint violation collaborative optimization on the basis of the traditional vehicle priority scheduling mechanism, so that multiple unmanned logistics vehicles can achieve optimal overall delivery efficiency under the premise of satisfying absolute safety constraints.

[0044] In this embodiment, a set of task states, task scheduling results, and a dynamic risk field are received. The target delivery point is used to determine whether there are intersections or conflicts in the vehicle's planned routes. Task time constraints, task urgency, overall task cost, and task priority are used to assist in determining the vehicle passage order when multiple unmanned logistics vehicles arrive at the conflict area simultaneously. Under the premise of meeting safety constraints, yielding decisions are made based on the distance between vehicles and the estimated arrival time at the conflict point, and a collaborative strategy that better meets delivery efficiency requirements is generated by combining task-level priorities.

[0045] First, a local traffic interaction state set is constructed based on the task state set and the dynamic risk field, represented as: ; in, Indicates unmanned logistics vehicle At the current control moment A set of local traffic interaction states; Indicates unmanned logistics vehicle At the current control moment Location; Indicates unmanned logistics vehicle At the current control moment speed; Indicates unmanned logistics vehicle At the current control moment The acceleration; Indicates unmanned logistics vehicle At the current control moment The heading angle; Indicates unmanned logistics vehicle At the current control moment The estimated time to reach the conflict zone is calculated based on the vehicle's current position, speed, acceleration, and reference path. Indicates unmanned logistics vehicle At the current control moment The dynamic risk assessment value is determined based on the dynamic risk field. The reference route is determined based on the vehicle's current status information and the location of the target delivery point.

[0046] In order to obtain unmanned logistics vehicles Predicting the motion state in the future time domain, based on the current control time. Local traffic interaction state set Short-term predictions are made based on reference paths. Specifically, this applies to unmanned logistics vehicles. At the current control moment Using position, velocity, acceleration, heading angle, and reference path as inputs, and according to a set prediction step size and prediction time domain length, the position, velocity, and heading status of the vehicle at several predicted future moments are recursively estimated to obtain the unmanned logistics vehicle. At the current control moment Predicted state sequence During the prediction process, the current position is used to determine the vehicle's prediction starting point, velocity and acceleration are used to estimate the vehicle's displacement changes over a short period, and the heading angle and reference path are used to determine the vehicle's future direction of motion. For example, at the current control moment... First, the vehicle's current position is matched to the current reference path to obtain the vehicle's current position on the reference path. In the prediction time domain, it is assumed that the vehicle continues to travel along the reference path, and the speed is recursively calculated based on the current acceleration. Then, the distance traveled by the vehicle along the reference path is calculated based on the predicted speed. Finally, the corresponding position coordinates and heading angle are obtained from the reference path, thereby forming a sequence of position, speed and heading states for several future prediction times.

[0047] Therefore, it is possible to calculate the relative relationships between a vehicle and other vehicles, pedestrians, road boundaries, and conflict zones in the future prediction time domain, and further obtain the estimated time to reach the conflict zone. Information such as predicted distances between vehicles and estimated collision times is collected. This predicted state sequence serves as input to the subsequent local safety constraint value function, used to determine whether vehicles may experience risks such as insufficient inter-vehicle distance, crossing road boundaries, or excessively short collision times within the future prediction period.

[0048] Furthermore, unmanned logistics vehicles The passage strategies of other unmanned logistics vehicles within a dynamic risk field are obtained through a communication link. Candidate cooperative strategies are generated by combining the reference path, dynamic risk field, predicted state sequence, and passage strategies of other unmanned logistics vehicles. These candidate cooperative strategies are generated using a finite strategy enumeration method. First, based on the vehicle's current reference path and dynamic risk field, it is determined whether there are path intersections with other unmanned logistics vehicles or dynamic traffic participants. Then, based on the predicted state sequence and passage strategies of other unmanned logistics vehicles, the estimated arrival time of each vehicle at the conflict area is calculated. If the unmanned logistics vehicle... If the time period overlaps with that of other vehicles entering the conflict area, a traffic conflict is determined. Candidate coordination strategies are pre-combined into a finite number of traffic schemes based on the number of conflicting vehicles and the conflict relationships, according to preset behavioral templates. For example, for unmanned logistics vehicles... When there is a path conflict with another unmanned logistics vehicle, at least the following candidate strategies should be generated: unmanned logistics vehicle Priority passage; unmanned logistics vehicles Slow down and yield; if adjusting speed alone cannot resolve the conflict, generate a lane-changing detour strategy; if no feasible passage strategy exists, generate a temporary parking strategy. Task time limit, task urgency, and task priority are used to rank the above candidate passage strategies. When the unmanned logistics vehicle... Vehicles with higher task priority or shorter remaining time limits will be given priority passage and will be transferred to unmanned logistics vehicles with path conflicts. .

[0049] Since candidate cooperative strategies may alter the speed, acceleration, or heading angle of the unmanned logistics vehicle, for each generated candidate cooperative strategy, the unmanned logistics vehicle... At the current control moment Using position, velocity, acceleration, heading angle, reference path, and candidate cooperative strategies as inputs, and according to a set prediction step size and prediction time domain length, the position, velocity, and heading status of the vehicle at several future prediction moments are recursively estimated to obtain the unmanned logistics vehicle. At the current control moment Candidate predicted state sequence .

[0050] Based on this, a local safety constraint value function is constructed to assess the maximum risk violation that a vehicle may commit within the future prediction period. The local safety constraint value function is used to determine safety constraint values ​​based on candidate cooperative strategies. Specifically, the formula is used... The security constraint value is determined based on the candidate collaborative strategy. Where, Indicates unmanned logistics vehicle The safety constraint value, Indicates the length of the prediction time domain. Indicates unmanned logistics vehicle At the current control moment Based on vehicle's own status information and candidate collaborative strategies The predicted first The vehicle state at the nth prediction step (i.e., the nth vehicle in the candidate prediction state sequence) (number of elements) Indicates the first The decision-level security constraint value for each prediction step. ; Indicates unmanned logistics vehicle Vehicle-to-vehicle safety distance constraints Indicates unmanned logistics vehicle Road boundary constraints, Indicates unmanned logistics vehicle Collision time constraints.

[0051] Among them, the safety distance constraint between vehicles is used to determine whether there is a risk of spatial conflict between vehicles in the future prediction time domain. ; This indicates the preset minimum safe following distance threshold (usually 2m~5m). Indicates vehicle and neighboring vehicles In the The Euclidean distance for each prediction step. Road boundary constraints are used to determine whether a vehicle's candidate predicted state sequence might exceed the currently allowed passage area, and are defined as: ; Indicates the vehicle is in Lateral offset of each prediction step; This indicates the maximum permissible lateral offset, typically set to half the lane width minus a safety margin. Collision time constraints are used to assess the potential timing of conflict between vehicles and other dynamic road users, and are defined as: ; Indicates unmanned logistics vehicle In the The predicted collision time for each prediction step; This represents the minimum permissible collision warning time threshold, typically set to 2 to 3 seconds.

[0052] when At that time, it indicated an unmanned logistics vehicle. In the Each prediction step simultaneously meets the requirements for vehicle distance, road boundary, and collision time, indicating a safe state. When At that time, it indicated an unmanned logistics vehicle. If the candidate cooperative strategy does not enter the collision hazard zone or violate traffic safety boundaries throughout the entire prediction period, it is deemed safe and feasible (i.e., the candidate cooperative passage strategy is retained). This achieves the goal of zero-constraint violation control. At that time, it indicated an unmanned logistics vehicle. If, during the entire prediction period, the vehicle state of the candidate cooperative strategy violates the safety distance, road boundary, or collision time constraints at least once, the candidate system strategy is deemed infeasible (i.e., eliminated) and does not participate in subsequent constraint boundary determination. The aforementioned constraints are primarily used in the multi-vehicle cooperative decision optimization process to adjust vehicle passage order and behavior strategies in advance (i.e., generate candidate cooperative strategies), and do not directly trigger underlying vehicle control actions.

[0053] In one embodiment, the process of determining the target passage strategy based on the globally unified cooperative constraint boundary and retained candidate cooperative strategies may include: determining the cooperative passage cost of the unmanned logistics vehicle; obtaining the local cost of other vehicles, including neighboring vehicles that have an interaction relationship with the unmanned logistics vehicle; determining the globally unified cooperative constraint boundary based on the cooperative passage cost and the local cost; determining whether the retained candidate cooperative strategies satisfy the globally unified cooperative constraint boundary; if the globally unified cooperative constraint boundary is satisfied, the retained candidate cooperative strategy is adopted as the target passage strategy; if the globally unified cooperative constraint boundary is not satisfied, the candidate cooperative strategy is adjusted until the globally unified cooperative constraint boundary is satisfied; if a candidate cooperative strategy that satisfies the globally unified cooperative constraint boundary cannot be generated, a parking strategy is generated as the target passage strategy.

[0054] To avoid the overall efficiency degradation caused by traditional priority rules relying solely on fixed yield relationships, this embodiment further constructs a vehicle cooperative cost function to determine the unmanned logistics vehicle's cost. The value of collaborative passage is uniformly quantified in terms of waiting time, additional energy consumption, and the impact of local conflicts: ; Indicates unmanned logistics vehicle The value of collaborative passage Indicates unmanned logistics vehicle The waiting time loss, Indicates unmanned logistics vehicle Increased energy consumption due to additional braking or detours Indicates unmanned logistics vehicle The extent of the impact of local conflicts; , , They are respectively , , The corresponding weighting coefficients are used to balance delivery efficiency and control stability. Among them, ; Indicates unmanned logistics vehicle Nearby unmanned logistics vehicles (i.e., unmanned logistics vehicles) The set of vehicles that have interactive relationships represents unmanned logistics vehicles. With unmanned logistics vehicles Euclidean distance, To prevent small positive numbers with a denominator of zero (such as 0.01).

[0055] Each unmanned logistics vehicle exchanges local cost values ​​(i.e., collaborative travel cost values) via a communication link. After acquiring the local cost values ​​of other unmanned logistics vehicles, a global unified collaborative constraint boundary is generated using a maximum consistency mechanism. (The one with the largest median value of the cooperative travel cost of this vehicle and the local costs of other unmanned logistics vehicles shall be used as the global unified cooperative constraint boundary.) ).

[0056] After generating the globally unified cooperative constraint boundary, the cooperative passage cost corresponding to each of the retained candidate cooperative strategies is calculated and compared with the globally unified cooperative constraint boundary. If a retained candidate cooperative strategy is less than or equal to the globally unified cooperative constraint boundary (the candidate cooperative strategy is determined to satisfy the globally unified cooperative constraint boundary), the corresponding candidate cooperative strategy is output as the target passage strategy (e.g., priority passage, yielding, or temporary stopping). If a retained candidate cooperative strategy is greater than the boundary (the candidate cooperative strategy is determined not to satisfy the globally unified cooperative constraint boundary), and if all retained candidate cooperative strategies do not satisfy the globally unified cooperative constraint boundary, the target speed or waiting time parameters in the candidate cooperative strategies are adjusted, candidate cooperative strategies are regenerated, and the safety constraint value and cooperative passage cost are recalculated until a candidate cooperative strategy less than or equal to the globally unified cooperative constraint boundary is generated. If a candidate cooperative strategy that satisfies the constraints cannot be generated within the current prediction period, the vehicle executes the minimum risk stopping strategy (i.e., the stopping strategy is used as the target passage strategy).

[0057] Based on this globally unified cooperative constraint boundary, the target traffic strategy of vehicles is dynamically adjusted, generating candidate cooperative strategies including deceleration and waiting, priority passage, lane changing and detour, or temporary stopping. Since all candidate cooperative strategies participating in the global cooperative selection have passed the local safety constraints of each vehicle, it can be guaranteed that the selected target traffic strategy meets the local safety requirements of the vehicles. The globally unified cooperative constraint boundary obtained by maximum aggregation is mainly used to further coordinate the cooperative traffic costs between different vehicles, avoiding excessively high cooperative costs for individual vehicles.

[0058] This approach differs from traditional collaborative control methods based on fixed rules or centralized scheduling. It does not rely on a single central control node; each unmanned logistics vehicle can independently complete local risk assessment and cost calculation, achieving globally consistent decision-making through distributed collaboration. This mechanism not only effectively avoids the amplification of conflicts caused by multi-vehicle competition but also enables rapid adaptive adjustments when delivery tasks frequently change, orders are temporarily inserted, or road conditions dynamically change, significantly improving traffic efficiency and operational stability in public road terminal delivery scenarios.

[0059] Step S4: Based on the local traffic interaction state, target traffic strategy, vehicle dynamics constraints, and reference path, multiple sets of candidate control input sequences are generated. A vehicle dynamics state transition function is used to obtain predicted candidate trajectories based on these sequences. A control layer safety screening function is used to filter the predicted candidate trajectories and determine the safe trajectory. A trajectory optimization objective function is used to obtain the final control input based on the safe trajectory. The unmanned logistics vehicle executes the final control input until the delivery task is completed.

[0060] To transform the target traffic strategy (such as yielding, priority passage, etc.) into control commands executable at the vehicle level, trajectory optimization calculation is further performed in step S4. Under conditions such as vehicle dynamics constraints, road boundary constraints, and local environmental risk constraints, an optimal control trajectory that satisfies safety, ride comfort, and execution feasibility is generated based on the local traffic interaction state, the target traffic strategy, and the reference path, and the final control input of the vehicle is output. Vehicle dynamics constraints include: speed not exceeding the vehicle's permissible range; acceleration and deceleration not exceeding the vehicle's maximum acceleration or deceleration; steering angle not exceeding the maximum steering angle; steering changes between adjacent control cycles not exceeding the vehicle's capacity; and the turning radius determined by the wheelbase and maximum steering angle not being less than the vehicle's minimum turning radius.

[0061] In step S4, the unmanned logistics vehicle is in the current vehicle state (i.e., the local traffic interaction state). At the current control moment The system takes position, heading angle, velocity and acceleration, local environmental risks (determined based on the dynamic risk field) and road boundary constraints as inputs to generate candidate control input sequences and perform trajectory optimization.

[0062] First, multiple candidate control input sequences are generated based on the current vehicle state, local environmental risks, road boundary constraints, target traffic strategy, vehicle dynamics constraints, and reference path. The candidate control input sequences are defined as follows: ; This represents the sequence dimension of the candidate control input sequence (i.e., the number of future control steps considered in trajectory planning, which is the same as the length of the prediction time domain). Indicates the prediction of the first time domain Each prediction step corresponds to a candidate control input.

[0063] Each control input Including vehicle steering angle control for adjusting the direction of travel. and longitudinal acceleration / deceleration control quantity for adjusting driving speed : Specifically, when the target traffic strategy is to slow down and yield, candidate control inputs for slowing down or stopping are generated first; when the target traffic strategy is to maintain traffic or prioritize passage, candidate control inputs are generated around the reference path and the target speed. Candidate control inputs can be obtained through discrete sampling or roll optimization within the allowable acceleration and steering angle ranges.

[0064] Wherein, the sequence dimension of the candidate control input sequence is the prediction time domain length. Based on this, the vehicle dynamics state transition function (in this embodiment, a bicycle model is used as the vehicle's discrete dynamics model, i.e., the vehicle dynamics state transition function, which is suitable for the kinematic description of low-speed unmanned logistics vehicles) is used to expand the candidate control input sequence and predict the vehicle's future trajectory state. The process of obtaining the predicted candidate trajectory based on the candidate control input sequence can include: using the formula Based on the candidate control input sequence, the predicted candidate trajectory is determined; where, Indicates unmanned logistics vehicle At the current control moment The predicted first The candidate vehicle states (including vehicle position coordinates, heading angle, and speed information) for each prediction step. This represents the vehicle dynamics state transition function (used to describe the change in vehicle state under the action of candidate control inputs). Indicates the first Each prediction step has candidate control inputs. By continuously expanding the candidate control input sequence, multiple sets of corresponding prediction candidate trajectories can be generated.

[0065] Furthermore, a local safety check is performed on each predicted candidate trajectory. To ensure that the vehicle remains within the safe and feasible region throughout execution, a control layer safety screening function is introduced. Specifically, the formula is used. The control layer safety value for the candidate vehicle state at each prediction step in the predicted candidate trajectory is determined. Where, Indicates unmanned logistics vehicle At the current control moment The predicted first The control layer safety value for the candidate vehicle state in each prediction step. Indicates unmanned logistics vehicle At the current control moment The predicted first The distance between the candidate vehicle state and dynamic obstacles in each prediction step. This indicates the setting of a safe interval distance. Indicates unmanned logistics vehicle At the current control moment The predicted first The remaining safe distance from the candidate vehicle state to the road boundary in each prediction step. Indicates unmanned logistics vehicle At the current control moment The predicted first The minimum braking safety margin of the candidate vehicle states in each prediction step (used to determine whether the vehicle has enough space to complete an emergency stop). When the control layer safety value of the candidate vehicle states in each prediction step of the predicted candidate trajectory is greater than zero (i.e., ... , When predicting a candidate trajectory, the predicted candidate trajectory is taken as a safe trajectory (that is, the vehicle state in the predicted candidate trajectory is always within the safe and feasible domain, meeting the requirements of collision avoidance and safe operation); otherwise, the predicted candidate trajectory is eliminated.

[0066] After completing the trajectory safety screening, all safe trajectories are comprehensively evaluated to select the execution scheme with the best overall performance. Specifically, the process of obtaining the final control input based on the safe trajectory using the trajectory optimization objective function includes: using the formula... Determine the comprehensive cost value of the safety trajectory; where, Indicates comprehensive value. Indicates the length of the prediction time domain. Indicates the first The trajectory tracking error cost per prediction step (used to measure the deviation between the vehicle's safe trajectory and the reference trajectory; the greater the trajectory deviation, the higher the trajectory tracking error cost). Indicates the first The comfort cost of each prediction step (used to constrain vehicle acceleration and steering rate of change to avoid drastic control; the more drastic the changes in steering angle and longitudinal acceleration, the higher the comfort cost). Indicates the first The energy cost of each prediction step Indicates the first The dynamic risk cost of each prediction step (used to penalize safe trajectories near high-risk areas and improve safety redundancy) is used. The safe trajectory corresponding to the minimum comprehensive cost is selected as the final execution trajectory. The first candidate control input in the candidate control input sequence corresponding to the final execution trajectory is extracted as the final control input.

[0067] in, ; Indicates the vehicle is in The trajectory position coordinates of each prediction step. This represents the coordinates of the corresponding point in the reference path. ; This represents the change in longitudinal acceleration at adjacent control moments (i.e., adjacent prediction steps). , Indicates the vehicle is in The longitudinal acceleration of each prediction step; This represents the change in rotation angle between adjacent control moments. , Indicates the vehicle is in The steering angle of each prediction step. The greater the absolute value of acceleration, the higher the vehicle's power output demand, and the greater the corresponding energy consumption cost. This indicates the real-time distance between the vehicle's current position and the nearest dynamic obstacle; when the vehicle approaches an obstacle, The risk decreases, the risk-value increases, and thus the penalty for high-risk trajectories is increased.

[0068] After the comprehensive cost value is calculated, the safe trajectory with the minimum comprehensive cost value is selected as the final execution trajectory. The candidate control input sequence corresponding to the final execution trajectory is determined, and the first candidate control input in the candidate control input sequence is extracted as the final control input.

[0069] After executing the final control input, the unmanned logistics vehicle reacquires the latest environmental information and enters the next round of rolling optimization to achieve continuous closed-loop trajectory adjustment until the delivery task is completed.

[0070] In step S4, a smooth mapping of multi-vehicle collaborative decision-making results (i.e., target traffic strategy) to vehicle execution layer control commands (i.e., final control input) is achieved through candidate control input sequence generation, predicted candidate trajectory generation, local safety verification, comprehensive evaluation, and rolling control output. Step S3 resolves the priority and conflict relationships in multi-vehicle interaction, while step S4 is responsible for transforming the collaborative results into specific, safe, and continuous vehicle motion behaviors. The two form a complete connection, jointly ensuring the stable operation of unmanned logistics vehicles on urban public roads in complex traffic environments.

[0071] Step S5: During the delivery process of the unmanned logistics vehicle, it is determined whether to trigger a safety intervention mechanism based on the vehicle's own state information and the dynamic risk field. When the safety intervention mechanism is triggered, an emergency control command is generated. The unmanned logistics vehicle executes the emergency control command. The process of determining whether to trigger the safety intervention mechanism based on the vehicle's own state information and the dynamic risk field includes: determining the vehicle's safety margin based on the vehicle's own state information and the dynamic risk field; determining the expected collision time based on the dynamic risk field; and determining that the safety intervention mechanism is triggered when the vehicle's safety margin is less than the set safety boundary or the expected collision time is less than the collision warning time threshold.

[0072] In one embodiment, to further improve the system's operational reliability on public roads, a safety monitoring mechanism independent of the main control link is established to monitor the final control input and vehicle operating status. The mechanism focuses on detecting the following risks: violations of traffic rules (such as driving against traffic or crossing lane lines), the presence of unidentified dynamic targets, insufficient minimum safety distance, and abnormal control execution (such as command deviation exceeding a threshold).

[0073] During vehicle operation, safety margins are continuously calculated based on the vehicle's own status information and dynamic risk field: . This indicates the safety margin of the vehicle under its current operating condition; This indicates the minimum distance between the vehicle and surrounding obstacles or dynamic traffic participants. This represents the preset safe distance threshold. When the condition is met... When the minimum distance between the vehicle and surrounding obstacles or dynamic traffic participants is less than the safe distance threshold, the vehicle is determined to have entered a dangerous boundary. At this point, a safety intervention mechanism is triggered.

[0074] Simultaneously calculate emergency time indicators . This indicates the estimated time before collision, which is the time remaining before the vehicle collides with an obstacle (or dynamic traffic participant) based on its current movement trend. When the following conditions are met... hour( This indicates the set collision warning time threshold, signifying that the vehicle is at risk of an emergency collision and triggering a safety intervention mechanism.

[0075] The emergency control commands generated after the safety intervention mechanism is triggered include operations such as active deceleration, emergency braking, or minimum risk stopping. After the safety intervention mechanism is triggered: if the vehicle still has a safe and feasible trajectory, a safe trajectory is regenerated first and a new control input is output; if the vehicle has an emergency collision risk, an active deceleration control command is generated; if there is no safe and feasible trajectory and the vehicle has an emergency collision risk, an emergency braking or minimum risk stopping control command is generated to control the vehicle to enter a minimum risk state.

[0076] Among them, the emergency control command generated after the safety intervention mechanism is triggered directly overrides the original control input. This step runs independently in a high-priority thread. Even if the main control thread consisting of steps S1 to S4 fails, the vehicle can still be kept in a minimum risk state, which significantly improves the overall safety level of the unmanned logistics vehicle.

[0077] Based on the same inventive concept, this application also provides a safe collaborative delivery control system for unmanned logistics vehicles to implement the aforementioned safe collaborative delivery control method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the safe collaborative delivery control system for unmanned logistics vehicles provided below can be found in the limitations of the safe collaborative delivery control method for unmanned logistics vehicles described above, and will not be repeated here.

[0078] In one exemplary embodiment, such as Figure 2 As shown, a safe collaborative delivery control system for unmanned logistics vehicles is provided, comprising: a delivery task scheduling module, an environmental perception and dynamic risk assessment module, a multi-vehicle collaborative decision-making module, and a safe trajectory optimization control module, all connected in sequence, as well as a safety monitoring and redundancy control module. The safety monitoring and redundancy control module is connected to the environmental perception and dynamic risk assessment module.

[0079] The task scheduling module is used to acquire delivery tasks. It also determines a set of task states based on the delivery tasks and the vehicle's own status information. Furthermore, it uses a dynamic task value function to determine the task scheduling result based on the task state set.

[0080] The environmental perception and dynamic risk assessment module is used to acquire environmental information in real time. This module is also used to construct a dynamic risk field based on environmental information.

[0081] The multi-vehicle cooperative decision-making module is used to construct a local traffic interaction state set based on the task state set and the dynamic risk field. It also determines a reference path based on task scheduling results. Furthermore, it obtains a predicted state sequence based on the local traffic interaction state set and the reference path. Finally, it generates candidate cooperative strategies based on the reference path, the dynamic risk field, the predicted state sequence, and the traffic strategies of other unmanned logistics vehicles. The module also determines safety constraint values ​​based on the candidate cooperative strategies using a local safety constraint value function. It retains candidate cooperative strategies corresponding to safety constraint values ​​less than or equal to zero. Finally, it determines the target traffic strategy based on the globally unified cooperative constraint boundary and the retained candidate cooperative strategies.

[0082] The safe trajectory optimization control module generates multiple candidate control input sequences based on local traffic interaction states, target traffic strategies, vehicle dynamics constraints, and reference paths. It also uses a vehicle dynamics state transition function to obtain predicted candidate trajectories based on these sequences. Furthermore, it employs a control-layer safety screening function to filter the predicted candidate trajectories and determine the safe trajectory. Finally, it uses a trajectory optimization objective function to obtain the final control input based on the safe trajectory. The unmanned logistics vehicle executes the final control input until the delivery task is completed.

[0083] The safety monitoring and redundancy control module is used to determine whether to trigger a safety intervention mechanism based on the vehicle's own status information and dynamic risk field during the delivery process of the unmanned logistics vehicle. When the safety intervention mechanism is triggered, the safety monitoring and redundancy control module generates an emergency control command. The unmanned logistics vehicle executes the emergency control command.

[0084] The entire system adopts a hierarchical and progressive structure, consisting of, from top to bottom, a delivery task scheduling layer (i.e., task scheduling module), a dynamic environment cognition layer (i.e., environment perception and dynamic risk assessment module), a multi-vehicle collaborative decision-making layer (i.e., multi-vehicle collaborative decision-making module), a safety trajectory optimization layer (i.e., safety trajectory optimization control module), and a safety supervision and execution layer (i.e., safety supervision and redundancy control module). Each layer interacts with the others via a unified state bus and forms a closed-loop control process through a cyclic feedback mechanism.

[0085] The delivery task scheduling layer is responsible for receiving delivery tasks from the delivery platform, parsing vehicle delivery targets, task deadlines, and task priorities, and forming task status information based on the current resource status of the vehicles. The dynamic environment cognition layer is responsible for integrating vehicle perception information to construct a real-time risk field (i.e., a dynamic risk field) around the vehicles. The multi-vehicle collaborative decision-making layer generates candidate collaborative strategies based on the shared status of vehicles and potential conflict relationships, and adopts a distributed decision-making mechanism based on zero-constraint violation collaborative optimization to achieve safe collaborative scheduling of multiple vehicles. The safe trajectory optimization layer combines vehicle dynamics constraints to perform safety screening and optimization of candidate control input sequences. The safety supervision and execution layer implements risk verification and redundancy protection for the final control input, and triggers minimum risk stopping or manual takeover when necessary.

[0086] In one exemplary embodiment, a safe collaborative delivery control device for unmanned logistics vehicles is provided, including a task acquisition and scheduling control center and a safe collaborative delivery control system for multiple unmanned logistics vehicles.

[0087] The task acquisition and scheduling control center is deployed on the delivery platform side. The safety collaborative delivery control system for the unmanned logistics vehicles is installed on the unmanned logistics vehicles. The task acquisition and scheduling control center is connected to the safety collaborative delivery control systems of multiple unmanned logistics vehicles.

[0088] The task acquisition and scheduling control center acquires order task information and vehicle status information from multiple unmanned logistics vehicles, generates delivery tasks, and transmits these tasks in real time to the corresponding unmanned logistics vehicle's safe collaborative delivery control system. The unmanned logistics vehicle's safe collaborative delivery control system is then used to complete the delivery tasks.

[0089] This safe and collaborative delivery control device for unmanned logistics vehicles is designed for urban public road delivery scenarios. It is suitable for open delivery environments characterized by multiple vehicles operating simultaneously, dense dynamic traffic participants, complex road structures, and real-time task changes. Delivery tasks are uniformly input and managed by a task acquisition and scheduling control center located on the delivery platform. This control center is independent of each unmanned logistics vehicle and receives delivery task requests from the order system, manual scheduling terminals, or business interfaces. It acquires order task information such as task number, delivery origin, delivery destination, task time limit, task priority, cargo status, and loading / unloading requirements. Simultaneously, the task acquisition and scheduling control center on the delivery platform receives vehicle status information from each unmanned logistics vehicle, including vehicle location, remaining battery power, cargo status, vehicle health status, and current task execution status, and generates corresponding delivery tasks for each vehicle accordingly.

[0090] Delivery tasks are sent to the corresponding unmanned delivery vehicles via wireless communication links. Upon receiving the task instruction, the unmanned delivery vehicle's onboard control system (i.e., the unmanned delivery vehicle's safety and collaborative delivery control system) combines the vehicle's real-time perception information, road environment information, dynamic traffic participant status, and the status of other collaborating vehicles to perform local safety checks, collaborative decision-making, and trajectory optimization. Therefore, the delivery platform is primarily responsible for acquiring, allocating, and issuing delivery tasks, while the onboard control system is mainly responsible for environmental perception, risk assessment, multi-vehicle collaborative decision-making, safe trajectory optimization, and vehicle chassis control execution. Through the coordinated operation of platform-side task scheduling and onboard safety control, unified management of delivery tasks can be achieved, while ensuring the vehicle's real-time safety response capability in complex and dynamic public road environments.

[0091] Delivery tasks are uniformly input and managed by the task acquisition and scheduling control center located on the delivery platform side. This control center is independent of each unmanned logistics vehicle and is used to receive delivery task requests sent by the order system, manual dispatch terminals, or business interfaces, and to obtain information such as task number, delivery origin, delivery destination, task time limit, task priority, cargo status, and loading and unloading requirements. At the same time, the task acquisition and scheduling control center on the platform side receives vehicle location, remaining battery power, cargo status, vehicle health status, and current task execution status from each unmanned logistics vehicle, and generates corresponding delivery task instructions for the corresponding vehicle, which are then sent to the corresponding unmanned logistics vehicle via wireless communication link.

[0092] The overall process is as follows: After the delivery platform's task acquisition and scheduling control center inputs the delivery task, it first determines the vehicle delivery target and task priority, and then issues the task instructions to the corresponding unmanned logistics vehicle; subsequently, the vehicle-side control system analyzes the delivery task, perceives the road environment in real time, and predicts potential risks; after identifying vehicle interaction conflicts, the system generates a target passage strategy; under cooperative constraints, it performs safety optimization on the vehicle's local trajectory; finally, after consistency verification by the safety supervision and redundancy control module, it issues control instructions to the vehicle chassis to achieve safe and stable automatic delivery.

[0093] In an exemplary embodiment, the above-described method, system, and apparatus are verified using a multi-vehicle collaborative delivery control system at an unsignalized intersection in an urban area as an example. In urban community terminal delivery scenarios, multiple unmanned logistics vehicles typically need to operate simultaneously in a shared public road environment, frequently passing through unsignalized intersections, narrow passing areas, and low-speed mixed-traffic roads within the community. In such scenarios, traffic conflicts can easily arise between vehicles due to path intersections. If each vehicle relies solely on independent decision-making for local avoidance, problems such as mutual waiting, decreased traffic efficiency, and even localized congestion can easily occur. This embodiment uses the example of three unmanned logistics vehicles simultaneously approaching an unsignalized intersection. The aforementioned safe collaborative delivery control system for the unmanned logistics vehicles is installed on each of the three vehicles, and the task acquisition and scheduling control center is located on the delivery platform side (connected and communicating with the control systems on the three unmanned logistics vehicles respectively).

[0094] Each unmanned logistics vehicle continuously collects information about the surrounding road environment and its own status through onboard LiDAR, integrated navigation unit, and wheel speed odometer. The LiDAR acquires real-time point cloud data of the location of road boundaries, static obstacles, and dynamic traffic participants around the vehicle. The integrated navigation unit simultaneously outputs the vehicle's attitude angle, angular velocity, and acceleration information. The vehicle positioning module integrates multi-source sensor data to generate the vehicle's real-time pose in the global coordinate system.

[0095] The task acquisition and scheduling control center receives order task information, assigns corresponding delivery target points to the three unmanned logistics vehicles, and calculates task priority based on the remaining delivery time of the order, the urgency of the task, and the current remaining battery power of the vehicle, thus generating three delivery tasks.

[0096] The delivery task scheduling module receives and parses delivery tasks, constructs a dynamic task cost function, and evaluates the vehicle's current delivery task in real time.

[0097] The environmental perception and dynamic risk assessment module processes the collected environmental information. The system identifies other vehicles, pedestrians, and non-motorized vehicles within the intersection through point cloud clustering, and establishes state vector models for detected dynamic targets, extracting their position, speed, heading angle, and acceleration information. Subsequently, a short-time trajectory prediction algorithm estimates the future motion trend of the targets, and combines the minimum predicted distance between the vehicle and the target, the estimated collision time, and the probability of behavioral intent to construct a local dynamic risk field at the intersection, quantifying the risk of each potential conflict area.

[0098] After completing the environmental risk assessment, the multi-vehicle collaborative decision-making module begins operation. Based on the unmanned logistics vehicle's current location, target direction, and predicted trajectory, a set of vehicle traffic interaction states is constructed, identifying potential path conflicts between vehicles. By processing the local observation states of the vehicles, the local safety constraint value of the unmanned logistics vehicle is calculated. The collaborative passage cost of the other two unmanned logistics vehicles is obtained through onboard communication exchange, and a global collaborative constraint boundary is generated using a maximum consistency mechanism. Taking into account vehicle waiting time loss, additional energy consumption, conflict risk, and task fairness constraints, a multi-vehicle collaborative behavior strategy is automatically generated, determining the optimal collaborative passage strategy (i.e., the target passage strategy) under the premise of meeting delivery time requirements. Ultimately, it is determined that this unmanned logistics vehicle should prioritize passing through the intersection (at this point, the passage strategies generated by the other two unmanned logistics vehicles are slowing down to yield and briefly stopping to wait, respectively), thereby avoiding vehicles simultaneously entering the conflict area.

[0099] After the target traffic strategy is determined, the safe trajectory optimization and control phase begins. Multiple candidate control input sequences are generated based on the collaborative decision-making results, including different combinations of steering angles and longitudinal acceleration / deceleration, and corresponding candidate trajectories are developed based on the vehicle dynamics model. Subsequently, a control-layer safety screening function is introduced to filter each trajectory in real time, eliminating trajectories that may lead to collisions or entry into dangerous areas. Simultaneously, considering trajectory tracking error, ride comfort, control energy consumption, and risk costs, the remaining trajectories are comprehensively optimized, and the trajectory that meets safety constraints and has the lowest cost is selected as the final execution trajectory.

[0100] During vehicle operation, the safety monitoring and redundancy control module continuously monitors the minimum safe distance between vehicles and the estimated collision time, and verifies in real time whether there are any potential risks in the final trajectory. When a social vehicle suddenly enters an intersection or an unmanned logistics vehicle's control deviation increases, the system can immediately trigger a redundancy intervention mechanism, including limiting the maximum steering angle, actively decelerating and braking, or performing a minimum-risk stopping operation to ensure that the vehicle is always in a safe and controllable state.

[0101] Through the above process, all three unmanned logistics vehicles successfully completed the coordinated passage at the intersection. No collisions or deadlocks occurred during the entire process. The overall passage efficiency of the vehicles was significantly better than the traditional independent decision-making mode, effectively verifying the multi-vehicle collaborative decision-making capability and practical application value of this application in complex interactive scenarios on public roads.

[0102] In another exemplary embodiment, the above-described method, system, and apparatus are verified using active safety avoidance control of sudden dynamic obstacles during delivery on public roads as an example. During terminal delivery on urban public roads, unmanned logistics vehicles often encounter sudden dynamic obstacles such as pedestrians suddenly crossing the road, non-motorized vehicles illegally changing lanes, and other vehicles temporarily merging into their lanes. These scenarios are highly random and unpredictable, placing higher demands on the vehicle's real-time safety control capabilities. This embodiment uses the example of an unmanned logistics vehicle encountering a pedestrian suddenly crossing the road during delivery on an urban secondary road to illustrate the avoidance process in the method and apparatus proposed in this application.

[0103] The unmanned delivery vehicle travels normally along the target path according to the predetermined delivery task. The lidar installed on the top of the vehicle continuously scans the road environment ahead, while the combined navigation unit and wheel speed encoder provide real-time vehicle motion status information. The system continuously updates the vehicle's current position on the global map through the positioning module and maps the real-time environmental information onto a dynamic risk field.

[0104] The delivery task scheduling module continuously assesses the vehicle's task status based on the current delivery target and remaining delivery time, and maintains the predetermined delivery route. As the vehicle approaches the exit of a residential area, the environmental perception and dynamic risk assessment module suddenly detects a pedestrian rapidly entering the driving area in front of the vehicle from an obstructed area on the right side of the road. The system immediately identifies this dynamic target and extracts its position, speed, and direction of movement information.

[0105] The environmental perception and dynamic risk assessment module predicts the short-term trajectory of pedestrians based on continuous observation results. It also calculates the estimated collision time and minimum predicted distance between vehicles and pedestrians. If the calculated risk field function shows a rapid increase, a collision risk is determined for the current trajectory, and local areas in the dynamic risk field are marked as high-risk zones.

[0106] Subsequently, the multi-vehicle collaborative decision-making module first determines whether there are other collaborative unmanned logistics vehicles or interference from other vehicles in the vicinity. Once it is confirmed that the current scenario is mainly a localized risk event for a single vehicle, the system automatically switches the control mode to active obstacle avoidance and raises the avoidance priority to the highest level.

[0107] After receiving the risk assessment results, the safety trajectory optimization control module immediately generates multiple candidate obstacle avoidance trajectories based on the original delivery trajectory, including a stationary deceleration and stopping trajectory, a left-deflection detour trajectory, and a right-deflection mild avoidance trajectory. All candidate trajectories are then screened one by one using the control layer safety screening function, eliminating trajectories that may enter dangerous areas or violate road boundary constraints. Simultaneously, considering vehicle dynamics feasibility and comfort constraints, the remaining trajectories undergo comprehensive cost optimization.

[0108] After optimization, a safe trajectory combining deceleration and gentle left-side deflection was selected as the final control scheme, and the corresponding control inputs were sent to the vehicle's actuators. The vehicle began to decelerate smoothly while making slight steering adjustments, successfully avoiding the pedestrian crossing. Throughout the obstacle avoidance process, the safety monitoring and redundant control module continuously monitored the minimum safe distance between the vehicle and the pedestrian and verified the estimated collision time in real time. Once it was detected that the pedestrian had completely left the vehicle's path and the local risk field had returned to normal, the active obstacle avoidance state was automatically deactivated, allowing the vehicle to gradually return to its original delivery path and continue completing subsequent delivery tasks.

[0109] Throughout the process, the unmanned logistics vehicle did not trigger an emergency stop, nor did it exhibit any sharp turns or trajectory instability, maintaining the continuity of the delivery task while ensuring pedestrian safety. This embodiment verifies that the present application possesses strong proactive risk prediction capabilities and highly reliable safety control capabilities in dynamic and sudden risk scenarios on urban public roads, effectively improving the actual road operation safety level of unmanned logistics vehicles.

[0110] Based on the above embodiments, this application addresses the common problems faced by unmanned logistics vehicles in urban public road terminal delivery scenarios, such as difficulty in coordinating traffic conflicts, unpredictable dynamic risks, and insufficient reliability of safety control. It provides a safe collaborative delivery control method, system, and device for unmanned logistics vehicles. By constructing a five-layer control system encompassing task scheduling, environmental risk assessment, multi-vehicle collaborative decision-making, safety trajectory optimization, and safety supervision redundancy, safe collaborative delivery of unmanned logistics vehicles in complex and dynamic traffic environments on urban public roads is achieved. This effectively solves the problems of insufficient coordination, insufficient safety, and insufficient dynamic response capabilities in existing technologies. It has significant advantages in safety, operational efficiency, and engineering feasibility, possesses strong engineering implementability, and can be widely applied to scenarios such as urban community delivery, park delivery, and public road terminal logistics transportation, demonstrating significant technological innovation value and promising industrial application prospects.

[0111] This application proposes a distributed multi-vehicle cooperative decision-making mechanism based on zero-constraint violation, which ensures no path conflicts and no deadlock waiting during multi-vehicle cooperation, enabling cooperative traffic control of multiple unmanned logistics vehicles in public road environments. In typical scenarios such as unsignalized intersections, narrow road meeting points, and station merging, potential path conflicts between vehicles can be identified in advance, and cooperative strategies such as priority passage, deceleration and yielding, or stopping and waiting can be automatically generated based on task timeliness, vehicle status, and road environment, reducing ineffective waiting and repeated avoidance between vehicles.

[0112] This application possesses strong dynamic risk prediction and proactive risk avoidance capabilities. In response to emergencies such as pedestrians suddenly crossing the road, non-motorized vehicles intruding, and vehicles temporarily changing lanes, it can identify collision risks in advance through real-time perception and trajectory prediction, and complete deceleration, detour, or stopping control before the risk escalates.

[0113] At the vehicle control level, this application employs safety constraint-driven trajectory optimization control, which adjusts the vehicle trajectory in real time while ensuring that safety boundaries are not breached. The generated control trajectory is smoother, reducing mechanical shocks from rapid acceleration, deceleration, and sharp turns, improving vehicle stability and comfort, and reducing energy consumption.

[0114] This application adopts a modular system architecture, which does not rely on high-precision maps or complex roadside infrastructure. It can be deployed with only LiDAR, integrated navigation units, and basic positioning equipment, and can be directly adapted to existing low-speed unmanned logistics vehicle platforms. The system can meet the real-time operation requirements of scenarios such as urban community delivery, park logistics transfer, and campus delivery, and has strong engineering feasibility and promotional application value.

[0115] In summary, this application outperforms existing technologies in terms of multi-vehicle collaboration efficiency, active safety capabilities, and practical deployment feasibility. It can effectively improve the autonomous operation level of unmanned logistics vehicles on urban public roads and provide reliable technical support for the large-scale implementation of smart logistics at the last mile.

[0116] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0117] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0118] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0119] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0120] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0121] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.

[0122] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0123] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A safe collaborative delivery control method for unmanned logistics vehicles, characterized in that, include: Obtain delivery tasks; Determine the task status set based on delivery tasks and vehicle status information; The task scheduling result is determined based on the task state set using a dynamic task value function; Real-time acquisition of environmental information; construction of a dynamic risk field based on environmental information; A local traffic interaction state set is constructed based on the task state set and the dynamic risk field; a reference path is determined based on the task scheduling results; a predicted state sequence is obtained based on the local traffic interaction state set and the reference path; and candidate cooperative strategies are generated based on the reference path, the dynamic risk field, the predicted state sequence, and the traffic strategies of other unmanned logistics vehicles. A local security constraint value function is used to determine the security constraint value based on candidate cooperative strategies; candidate cooperative strategies corresponding to security constraint values ​​that are less than or equal to zero are retained. The target passage strategy is determined based on the globally unified collaborative constraint boundary and the retained candidate collaborative strategies; Multiple sets of candidate control input sequences are generated based on local traffic interaction states, target traffic strategies, vehicle dynamics constraints, and reference paths. The predicted candidate trajectory is obtained based on the candidate control input sequence using the vehicle dynamics state transition function. A control layer safety filtering function is used to filter the predicted candidate trajectories and determine the safe trajectory; The trajectory optimization objective function is adopted, and the final control input is obtained based on the safe trajectory. The unmanned logistics vehicle executes the final control input until the delivery task is completed; During the delivery process of unmanned logistics vehicles, it is determined whether to trigger a safety intervention mechanism based on the vehicle's own status information and dynamic risk field. When the safety intervention mechanism is triggered, an emergency control command is generated; The unmanned logistics vehicle executes emergency control commands.

2. The safe collaborative delivery control method for unmanned logistics vehicles according to claim 1, characterized in that, The environmental information includes observational information on road structure, obstacle boundaries, and dynamic traffic participants; Constructing a dynamic risk field based on environmental information includes: The observation information is mapped to a global reference coordinate system, and a target state vector of dynamic traffic participants is constructed. The risk field function is used to determine the current comprehensive risk value of the unmanned logistics vehicle based on the target state vector; A dynamic risk field is constructed based on the current comprehensive risk value of unmanned logistics vehicles.

3. The safe collaborative delivery control method for unmanned logistics vehicles according to claim 1, characterized in that, The local security constraint value function is used to determine the security constraint value based on candidate cooperative strategies, including: Using formula Based on the candidate collaborative strategy, the security constraint value is determined; In the formula, Indicates unmanned logistics vehicle The safety constraint value, Indicates the length of the prediction time domain. Indicates unmanned logistics vehicle At the current control moment Based on vehicle's own status information and candidate collaborative strategies The predicted first Vehicle state in each predicted step; Indicates the first The decision-level security constraint value for each prediction step. ; Indicates unmanned logistics vehicle Vehicle-to-vehicle safety distance constraints Indicates unmanned logistics vehicle Road boundary constraints, Indicates unmanned logistics vehicle Collision time constraints.

4. The safe collaborative delivery control method for unmanned logistics vehicles according to claim 1, characterized in that, The target travel strategy is determined based on globally unified collaborative constraint boundaries and retained candidate collaborative strategies, including: Determine the cooperative passage value of the unmanned logistics vehicle; obtain the local value of other vehicles; the other vehicles include nearby vehicles that have an interaction relationship with the unmanned logistics vehicle; The global unified collaborative constraint boundary is determined based on the collaborative passage cost and the local cost. Determine whether the retained candidate collaboration strategies satisfy the global unified collaboration constraint boundary; If the global unified collaborative constraint boundary is satisfied, the retained candidate collaborative strategy will be used as the target passage strategy. If the global unified collaborative constraint boundary is not satisfied, the candidate collaborative strategy is adjusted until the global unified collaborative constraint boundary is satisfied. If a candidate cooperative strategy that satisfies the globally unified cooperative constraint boundary cannot be generated, a parking strategy is generated as the target passage strategy.

5. The safe collaborative delivery control method for unmanned logistics vehicles according to claim 1, characterized in that, The sequence dimension of the candidate control input sequence is the prediction time domain length; Using the vehicle dynamics state transition function, the predicted candidate trajectory is obtained based on the candidate control input sequence, including: Using formula Based on the candidate control input sequence, a predicted candidate trajectory is determined; where, Indicates unmanned logistics vehicle At the current control moment The predicted first The candidate vehicle state for each prediction step. This represents the vehicle dynamics state transition function. Indicates the first Candidate control inputs for each prediction step.

6. The safe collaborative delivery control method for unmanned logistics vehicles according to claim 1, characterized in that, The control layer safety filtering function is used to filter the predicted candidate trajectories and determine the safe trajectories, including: Using formula The control layer safety value for the candidate vehicle state at each prediction step in the predicted candidate trajectory is determined; where, Indicates unmanned logistics vehicle At the current control moment The predicted first The control layer safety value for the candidate vehicle state in each prediction step. Indicates unmanned logistics vehicle At the current control moment The predicted first The distance between the candidate vehicle state and dynamic obstacles in each prediction step. This indicates the setting of a safe interval distance. Indicates unmanned logistics vehicle At the current control moment The predicted first The remaining safe distance from the candidate vehicle state to the road boundary in each prediction step. Indicates unmanned logistics vehicle At the current control moment The predicted first Minimum brakeable safety margin for candidate vehicle states in each prediction step. When the control layer safety value of the candidate vehicle state in each prediction step of the predicted candidate trajectory is greater than zero, the predicted candidate trajectory is taken as the safe trajectory; otherwise, the predicted candidate trajectory is discarded.

7. The safe collaborative delivery control method for unmanned logistics vehicles according to claim 1, characterized in that, The trajectory optimization objective function is used to obtain the final control input based on the safe trajectory, including: Using formula Determine the comprehensive cost value of the safety trajectory; where, Indicates comprehensive value. Indicates the length of the prediction time domain. Indicates the first The cost of trajectory tracking error per prediction step, Indicates the first The comfort cost of each predicted step, Indicates the first The energy cost of each prediction step Indicates the first The dynamic risk cost of each prediction step; The safe trajectory corresponding to the minimum overall cost will be used as the final execution trajectory. The first candidate control input in the candidate control input sequence corresponding to the final execution trajectory is extracted as the final control input.

8. The safe collaborative delivery control method for unmanned logistics vehicles according to claim 1, characterized in that, Determining whether to trigger a safety intervention mechanism based on the vehicle's own state information and dynamic risk field includes: The vehicle safety margin is determined based on the vehicle's own state information and the dynamic risk field; The estimated collision time is determined based on the dynamic risk field. When the vehicle safety margin is less than the set safety boundary or the expected collision time is less than the collision warning time threshold, a safety intervention mechanism is triggered.

9. A safe collaborative delivery control system for unmanned logistics vehicles, characterized in that, The method for safe collaborative delivery control of unmanned logistics vehicles according to any one of claims 1-8 includes: a delivery task scheduling module, an environmental perception and dynamic risk assessment module, a multi-vehicle collaborative decision-making module, and a safe trajectory optimization control module connected in sequence, as well as a safety supervision and redundancy control module; the safety supervision and redundancy control module is connected to the environmental perception and dynamic risk assessment module. The task scheduling module is used to acquire delivery tasks; the task scheduling module is also used to determine a task status set based on the delivery tasks and the vehicle's own status information; the task scheduling module is also used to determine the task scheduling result based on the task status set using a dynamic task value function; The environmental perception and dynamic risk assessment module is used to acquire environmental information in real time; the environmental perception and dynamic risk assessment module is also used to construct a dynamic risk field based on the environmental information. The multi-vehicle collaborative decision-making module is used to construct a local traffic interaction state set based on a task state set and a dynamic risk field; the multi-vehicle collaborative decision-making module is also used to determine a reference path based on task scheduling results; the multi-vehicle collaborative decision-making module is also used to obtain a predicted state sequence based on the local traffic interaction state set and the reference path; the multi-vehicle collaborative decision-making module is also used to generate candidate collaborative strategies based on the reference path, the dynamic risk field, the predicted state sequence, and the passage strategies of other unmanned logistics vehicles; the multi-vehicle collaborative decision-making module is also used to determine a safety constraint value based on the candidate collaborative strategies using a local safety constraint value function; the multi-vehicle collaborative decision-making module is also used to retain candidate collaborative strategies corresponding to safety constraint values ​​less than or equal to zero; the multi-vehicle collaborative decision-making module is also used to determine a target passage strategy based on a globally unified collaborative constraint boundary and the retained candidate collaborative strategies; The safe trajectory optimization control module is used to generate multiple sets of candidate control input sequences based on local traffic interaction states, target traffic strategies, vehicle dynamics constraints, and reference paths. The module also uses a vehicle dynamics state transition function to obtain predicted candidate trajectories based on the candidate control input sequences. Furthermore, it uses a control layer safety screening function to filter the predicted candidate trajectories and determine the safe trajectory. Finally, it uses a trajectory optimization objective function to obtain the final control input based on the safe trajectory. The unmanned logistics vehicle executes the final control input until the delivery task is completed. The safety monitoring and redundancy control module is used to determine whether to trigger a safety intervention mechanism based on the vehicle's own status information and dynamic risk field during the unmanned logistics vehicle's delivery task. When the safety intervention mechanism is triggered, the safety monitoring and redundancy control module generates an emergency control command, and the unmanned logistics vehicle executes the emergency control command.

10. A safe collaborative delivery control device for unmanned logistics vehicles, characterized in that, This includes a task acquisition and scheduling control center and multiple safe collaborative delivery control systems for unmanned logistics vehicles as described in claim 9; The task acquisition and scheduling control center is deployed on the delivery platform side; the safe collaborative delivery control system of the unmanned logistics vehicle is installed on the unmanned logistics vehicle; the task acquisition and scheduling control center is connected to the safe collaborative delivery control systems of multiple unmanned logistics vehicles; The task acquisition and scheduling control center is used to acquire order task information and vehicle status information of multiple unmanned logistics vehicles, generate delivery tasks, and transmit the delivery tasks to the corresponding unmanned logistics vehicle's safe collaborative delivery control system in real time. The safe and collaborative delivery control system of the unmanned logistics vehicle is used to complete the delivery task.

Images