A distributed collaborative safety control method for multi-agent systems

Abstract

The present invention belongs to the field of intelligent agent safety control technology and relates to a distributed collaborative safety control method for multi-agent systems. The method comprises: S1, establishing a kinematic model of omnidirectional mobile agents and generating nominal speed instructions based on preset trajectories; S2, determining the position information of each agent and broadcasting it to other agents; S3, determining a candidate zero-crossing barrier function; S4, using the candidate zero-crossing barrier function to generate safety constraints; S5, determining a loss function based on the nominal speed and solving the control variable that minimizes the loss function while satisfying the safety constraints as a safety control instruction; S6, controlling the movement of the agents according to the safety control instructions. The method has the beneficial effect of effectively avoiding collisions between agents in a large-scale intelligent agent cluster, even in the absence of global information and central nodes, while also avoiding falling into local optimal solutions, ensuring that the agents safely navigate to designated targets.

Description

Technical Field

[0001] The present invention relates to the field of intelligent agent safety control technology, and in particular to a distributed collaborative safety control method applied to a multi-agent system. Background Art

[0002] Safety refers to maintaining a sufficient distance between any two agents in a multi-agent swarm composed of multiple robots. The goal of safe control is to modify the nominal control instructions as little as possible without violating safety constraints, so that each agent can safely navigate to the designated target point.

[0003] With the advancement of artificial intelligence and robotics, multi-agent systems (MAS) have demonstrated unique advantages in various fields, including formation flying, swarm locomotion, and multi-robot collaborative operations. These systems consist of multiple agents with partial perception, communication, computation, and execution capabilities, interconnected through a network, demonstrating a high degree of autonomy, fault tolerance, collaboration, and flexibility. However, when the paths of multiple agents intersect, avoiding collisions between them presents a technical challenge.

[0004] After decades of development, numerous safety control methods have been proposed. These primarily include global motion planning based on path planning and trajectory optimization, and local motion planning based on reinforcement learning. Path planning methods can plan a safe path based on environmental information, avoiding deadlock. However, they rely on accurate environmental perception, require significant computational resources, and are prone to becoming stuck in local optimal solutions. Reinforcement learning methods can adapt to complex obstacle avoidance scenarios and potentially achieve optimal solutions. However, their drawbacks include significant computational resources and limited generalization capabilities. Obstacle function control is a relatively powerful and recently developed method. As a controller expressed in an optimized form, it ensures system safety with a concise controller form, simple computation, and strong real-time performance. With the widespread adoption of large-scale robot swarms in recent years, research has focused on the safe control of multi-agent systems. For example, autonomous vehicles must avoid other vehicles and pedestrians in complex traffic environments. In collaborative operations with drones and in drone delivery, collisions must be avoided when intersecting paths. In logistics warehouses, multiple handling robots must efficiently move goods within a single space.

[0005] Key issues to consider in multi-agent safety control include: each agent only has information about its own mission but lacks complete knowledge of the missions and states of other vehicles; and agents are coupled, meaning that whether an agent will collide depends not only on its own decisions but also on those of its neighbors. Multi-agent safety control primarily employs centralized and decentralized control algorithms to address these two key issues.

[0006] Centralized control algorithms assume the existence of a central node that can communicate with all agents. Each agent uploads its status and task information to the central node. The central node then centrally calculates control instructions for all agents and distributes them to all agents. The advantage is that the central node can access global information and make globally optimal decisions. However, in large-scale cluster scenarios spanning a large space, communication range limits the availability of a central node that can cover all agents. Furthermore, each agent must upload its status and task information to the central node. The larger the cluster, the higher the network bandwidth requirements and the slower the computation speed. Decentralized control decouples the system by pre-assigning coupling constraints. Each agent is deployed with an independent safety controller that only calculates its own safety control instructions and does not require task information from other agents. Decentralized algorithms offer the advantages of fast computation and ease of scalability. However, because the safety controller makes decisions based on local information, it is prone to falling into local optimal solutions. Summary of the Invention

[0007] Technical issues to be solved:

[0008] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a distributed collaborative safety control method applied to a multi-agent system, which solves the technical problem of how to avoid collisions between agents while avoiding falling into local optimal solutions.

[0009] Technical solution:

[0010] In order to achieve the above objectives, the main technical solutions adopted by the present invention include:

[0011] In a first aspect, the present invention provides a distributed collaborative safety control method for a multi-agent system, comprising:

[0012] S1, establish the kinematic model of the omnidirectional mobile agent and generate the nominal speed command according to the preset trajectory;

[0013] S2, determine the location information of each agent and broadcast it to other agents within the preset range of each agent;

[0014] S3, determining a candidate zero-crossing barrier function;

[0015] S4, based on the kinematic model and the relative position of each agent and all neighboring agents within its preset range, generates safety constraints using candidate zero-crossing barrier functions;

[0016] S5, determining the loss function based on the nominal speed, and solving the control quantity that minimizes the loss function while satisfying the safety constraints as the safety control instruction;

[0017] S6, controls the movement of the intelligent body according to the safety control instructions;

[0018] S7, update the agent position and determine whether all agents have reached the end point. If not, execute S2; if so, end.

[0019] Optionally, the kinematic model is ;

[0020] in, is the position of the agent, represented by a two-dimensional column vector in the two-dimensional world coordinate system represents the position of the i-th agent, It is The first derivative of the agent's position; It is the linear velocity of the intelligent body and the control input of the system. In the two-dimensional coordinate system, it is represented by a two-dimensional column vector Indicates the The speed of an agent.

[0021] Optionally, the zero-crossing barrier function for: ;

[0022] The constant , is the expected safe distance between two agents, is an additional safety margin to ensure robustness. , is the total number of agents in the cluster, symbol represents the two-norm of the vector, represents the position of the jth agent.

[0023] Optionally, safety constraints are generated based on the candidate zero-crossing barrier function, including:

[0024] The safety constraint g is obtained by derivation of the zero-crossing barrier function with respect to time t as follows: ;

[0025] in, is the decay rate of the barrier function, which is a constant greater than 0; is the sampling time, u is the control input, .

[0026] Optionally, the loss function is: ;

[0027] in, is the nominal control quantity, i Each agent provides a reference value of speed.

[0028] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements a distributed collaborative safety control method applied to a multi-agent system as described in any one of the first aspects above.

[0029] In a third aspect, the present invention provides a storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements a distributed collaborative safety control method applied to a multi-agent system as described in any one of the first aspects above.

[0030] Beneficial effects:

[0031] The beneficial effects of the present invention are as follows: Compared with the prior art, the distributed collaborative safety control method for multi-agent systems of the present invention has the following significant advantages:

[0032] (1) The proposed method uses a distributed optimization algorithm to enable each agent to consider the decision information of neighboring agents and achieve collaborative obstacle avoidance. This not only avoids collisions between agents but also effectively solves the problem of local optimal solutions caused by decentralized safety controllers that rely only on local information, thus ensuring the optimality of global path planning.

[0033] (2) This invention distributes the safety controller to each agent. Each agent only needs to communicate with adjacent agents through the local area network, solving the problem of limited communication range. It also improves the problem of insufficient network bandwidth caused by collecting information from all agents and the slow computing speed caused by solving the safety constraints of all robots in the cluster when a centralized controller is used to control the safety of large-scale robot clusters.

[0034] (3) The agent broadcasts itself in the future The system uses position information from the next second to share motion intentions with neighboring agents in advance, reducing position errors caused by agent movement during the calculation process. An additional safety margin is also introduced to compensate for position prediction and calculation errors caused by network latency. This mechanism not only resolves the clock inconsistency problem caused by network latency in multi-agent systems, but also ensures that collaborative obstacle avoidance between agents can be carried out safely and effectively even in discontinuous calculations, significantly improving the robustness of multi-agent systems in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A block diagram of a control system provided by an embodiment of the present invention;

[0036] Figure 2 A schematic diagram of the initial state of a simulation provided by an embodiment of the present invention;

[0037] Figure 3A schematic diagram of simulation results of a distributed safety controller provided by an embodiment of the present invention;

[0038] Figure 4 A schematic diagram of simulation results of a distributed security controller provided by an embodiment of the present invention;

[0039] Figure 5 A schematic diagram of simulation results of a centralized security controller provided by an embodiment of the present invention;

[0040] Figure 6 A physical experiment platform for the feasibility of the embodiments of the present invention;

[0041] Figure 7 A diagram showing the results of a physical experiment provided by an embodiment of the present invention;

[0042] Figure 8 A flow chart of a distributed collaborative security control method applied to a multi-agent system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0044] This invention addresses the safety requirements of omnidirectional mobile robots performing tasks in large-scale robot cluster scenarios, and designs a distributed collaborative safety control method for multi-agent systems. On the one hand, any agent in the cluster only knows its own control objectives; on the other hand, agents can only communicate with adjacent agents, and there is no central node that can obtain global information. The goal of this method is that agents avoid collisions with other agents and reach their target positions based solely on their own information and that of adjacent agents. A key feature of this method is that during real-time interaction between agents, the state of any two agents in the cluster can always be ensured to be within the safe set. Furthermore, under the premise that the agents do not have global information, they can communicate with adjacent agents and modify the nominal control instructions as little as possible to achieve their respective mission objectives.

[0045] This method formulates the safety control problem as a quadratic programming problem with linear inequality constraints by constructing a robust control barrier function and generating safety constraints, and introduces a distributed optimization algorithm for solving it. In this invention, several important improvements are made to address the lack of global information and coupling constraints between agents in a multi-agent system. First, by designing a candidate zero-crossing barrier function and constructing safety constraints, it is ensured that the agent is safe for each adjacent agent; second, the safety control problem is converted into a quadratic programming problem, and the control input is optimized to meet all safety constraints at the same time; finally, a distributed optimization algorithm is introduced to achieve distributed iterative solution of the quadratic programming problem on each agent, avoiding the problems of limited communication range of the central node and insufficient network bandwidth. These three improvements enable the effective avoidance of collisions between agents in a large-scale agent cluster, even in the absence of global information and central nodes, and ensure that the agents can safely navigate to the designated target point.

[0046] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0047] The controlled object considered in this embodiment is A multi-agent system consists of omnidirectional agents. Here, an unmanned vehicle equipped with Mecanum wheels that can perform omnidirectional motion is used to illustrate the system. The kinematic model of this agent can be expressed as a first-order integrator model: ;

[0048] in, is the position of the agent, represented by a two-dimensional column vector in the two-dimensional world coordinate system represents the position of the i-th agent, It is The first derivative of the agent's position; It is the linear velocity of the intelligent body and the control input of the system. In the two-dimensional coordinate system, it is represented by a two-dimensional column vector Indicates the The speed of an agent.

[0049] In addition, a path tracing algorithm is used to derive a bounded, unverified, and safe nominal speed. The distributed safety controller designed in this embodiment performs safety filtering on the nominal speed and outputs safety control instructions to the motion controller, which ultimately drives the intelligent agent to move so that it reaches the target point without colliding with other intelligent agents.

[0050] exist Figure 1 The overall algorithm framework of the method provided in this embodiment is shown in FIG. Figure 2 、 Figure 3 、 Figure 4 and Figure 5 The comparison of Matlab simulation results of controllers designed based on the algorithm of this embodiment and the distributed algorithm is demonstrated. Figure 6 The algorithm was demonstrated in a physical experiment system. This system consists of the Optitrack motion capture system and several omnidirectional mobile robots powered by Raspberry Pis. All devices are connected to the same local area network via wireless network cards or cables, and information is transmitted between devices using the ROS distributed communication framework. Figure 7 Physical experimental results of the algorithm are presented.

[0051] First, refer to Figure 8 This embodiment provides a distributed collaborative security control method for a multi-agent system, including:

[0052] S1, establish the kinematic model of the omnidirectional mobile agent and generate the nominal speed command according to the preset trajectory.

[0053] S2, determine the location information of each agent and broadcast it to other agents within the preset range of each agent.

[0054] Assume that the agents in the cluster have the same control period , Greater than the time required for the distributed intelligent agent to solve the quadratic programming problem. At this moment, the agent samples its own position information through the motion capture system , and based on its current speed, predict Then, record your own location information ,in Is within the current control cycle , agent The actual amount of control performed. The agent latches this information during the control cycle And inform the adjacent agents through broadcasting. At the same time, obtain the sampling position information broadcast by the adjacent agents .

[0055] S3, determining a candidate zero-crossing barrier function.

[0056] ;

[0057] in, , .

[0058] is the expected safe distance between two agents, usually the sum of the expansion circle radii of agents i and j.

[0059] It is an additional safety margin that can compensate for the influence of predicted position error and calculation error on system safety. The definition of prediction error is as follows The prediction error can be regarded as the measurement error caused by the sensor, and the calculation error caused by the finite-step iteration can be regarded as the actuator error. Existing papers in related directions have proved that in sampling systems with measurement errors and actuator errors, by adding a safety margin, it can be ensured that the application of safety control quantities can still ensure that the state quantity is within the safety set. Therefore, this patent adds a safety margin term , in order to compensate for the prediction deviation and calculation error of the intelligent agent's own position information, generate control input, and still ensure that the system state strictly satisfies forward invariance.

[0060] S4, based on the kinematic model and the relative position of each agent and all its neighboring agents within a preset range, generates safety constraints using candidate zero-crossing barrier functions.

[0061] Design candidate zero-crossing barrier functions based on the relative distance between two different agents , and construct safety constraints to ensure that the agent is safe with respect to every neighboring agent.

[0062] Based on the theory of control obstacle function, About time Take the derivative and get a safety constraint.

[0063] This constraint limits the control inputs of the agents, ensuring that collisions do not occur between them, thereby defining a set of control strategies that satisfy safety conditions. To ensure that no two agents in a multi-agent system collide, it is usually necessary to apply the same safety constraint to all pairs of agents.

[0064] .

[0065] As a result, the number of constraints increases dramatically, slowing down the optimization problem and degrading the algorithm's real-time performance. Furthermore, establishing safety constraints between all agents requires global agent location information, which is likely infeasible given practical communication range and bandwidth limitations. To address these issues, the present invention provides a solution in S5.

[0066] S5, determine the loss function based on the nominal speed, and solve the control quantity that minimizes the loss function while satisfying the safety constraints as the safety control instruction.

[0067] The control objective of the multi-agent swarm safety control problem is to minimize the modification of nominal control instructions while ensuring that no two agents in the swarm collide. Therefore, a quadratic cost function can be used to compensate for the deviation between the control variable and the nominal control variable. By treating safety constraints as hard constraints in the optimization problem, the safety control problem can be formulated as a quadratic programming problem with multiple linear inequality constraints. This is a convex optimization problem.

[0068] Considering that the safety control problem is a local problem, which means that when the agents are far away from each other, their dynamic behavior may not be of concern. Therefore, if the safety constraints between agents that are far enough away are ignored, the above safety controller is still valid. At the same time, considering that the solution time of the multi-agent safety controller is not negligible, the controller can be modified as follows:

[0069] ;

[0070] ;

[0071] .

[0072] Indicates that at the sampling time , the set of other agents in the agent's neighborhood, defined as follows: ;

[0073] in, is the upper bound of the agent’s speed, is a constant greater than 0 defined in step 2.

[0074] It is not difficult to see that the original quadratic programming problem has coupled linear constraints. The decision of the agent depends on its neighboring agents This creates an algebraic loop that prevents the controller from being directly deployed in a distributed manner.

[0075] The coefficient matrix of the quadratic term in the original quadratic programming problem is positive definite, and the constraints are all linear inequalities. Therefore, the original quadratic programming problem is a convex optimization problem. Based on the strong duality principle, we can solve the dual problem of the original quadratic programming problem:

[0076] ;

[0077] ;

[0078] in It is a constraint The corresponding Lagrange multiplier is, , is the Lagrangian function of the original quadratic programming problem, defined as follows:

[0079] .

[0080] The Lagrangian function contains two sets of variables, is the control input, Is a constraint variable. Solve the inner control input , and then the outer constraint variables are updated using the gradient ascent method.

[0081] The specific method for solving the dual problem is given as follows:

[0082] 1. Initialize the initial value of the constraint variable , solve the inner control input ;

[0083] .

[0084] 2. Use gradient ascent to update outer constraint variables ;

[0085] .

[0086] in depending on Moment, Agent and The number of other agents in their respective neighborhoods and , defined as follows: .

[0087] 3. Order ,in is the maximum number of iterations limited by network communication delay.

[0088] Agent Update control input locally and the constraint variables associated with themselves .

[0089] Agent Iteratively update the constraint variables related to itself locally When , only the local iteration result of the previous iteration round is needed and the result of iterating on adjacent agents . While calculating the local control input Only local bound variables are required . Therefore, the method for solving the dual problem is completely distributed.

[0090] It should be noted that and The same constraint corresponds to the agent. and agents But since they have the same iteration formula, .

[0091] S6, controls the movement of the intelligent body according to the safety control instructions.

[0092] S7, update the agent position and determine whether all agents have reached the end point. If not, execute S2; if so, end.

[0093] Optionally, the kinematic model is: ;

[0094] in, is the position of the agent, represented by a two-dimensional column vector in the two-dimensional world coordinate system represents the position of the i-th agent, It is The first derivative of the agent's position; It is the linear velocity of the intelligent body and the control input of the system. In the two-dimensional coordinate system, it is represented by a two-dimensional column vector Indicates the The speed of an agent.

[0095] Optionally, the zero-crossing barrier function h is:

[0096] ;

[0097] The constant , is the expected safe distance between two agents, is an additional safety margin to ensure robustness. , is the total number of agents in the cluster, symbol represents the two-norm of the vector, represents the position of the jth agent.

[0098] Optionally, safety constraints are generated based on the candidate zero-crossing barrier function, including:

[0099] The safety constraint g is obtained by taking the time derivative of the zero-crossing barrier function as follows: ;

[0100] in, is the decay rate of the barrier function, which is a constant greater than 0; is the sampling time, For control input.

[0101] Optionally, the loss function is: ;

[0102] in, is the nominal control quantity, i Each agent provides a reference value of speed, Representative i The terminal position of the agent, is the actual output control quantity.

[0103] Optionally, a distributed optimization algorithm is used to solve the loss function to obtain the control input, including:

[0104] Inner control input based on Lagrangian function to solve loss function , use the gradient ascent method to update the outer constraint variable λ.

[0105] The distributed security control method designed by the present invention has the following characteristics:

[0106] Unlike traditional safety control methods, the distributed safety controller does not rely on its own sensors to detect the location of neighboring agents. Instead, it broadcasts its own location information to neighboring agents and simultaneously receives location information broadcast by neighboring agents. This approach ensures that different agents can collectively address the same problem even if their clocks are out of sync.

[0107] By deploying the distributed controller on each agent, each agent uses its own local information to solve its own safety control instructions, ensuring the safety of the multi-agent system. Furthermore, the safety control decisions made by this control method based on each agent's local information are globally optimal.

[0108] The input of the control method is the information of adjacent intelligent agents. Even when facing large-scale intelligent clusters in large-scale scenarios, a single intelligent agent only needs the information of adjacent intelligent agents, which greatly reduces the network bandwidth demand and reduces the computing speed. It is also suitable for scenarios where there is no central node.

[0109] It should be noted that in the method proposed in the present invention, when the intelligent agent calculates the local security issues within its own neighborhood, it is completely based on local information and there is no need to obtain the global position information of all intelligent agents.

[0110] Specifically, each agent Broadcast its own location information and receive broadcast information from neighboring agents , can only obtain the location of the neighboring agents with which it communicates directly. Neighborhood Set The definition of depends on the relative distance between agents. The construction of the neighborhood set only depends on the location information of local agents, does not require the knowledge of all agents, and is completely implemented locally in the agent.

[0111] The present invention uses a distributed optimization algorithm to ensure that when an agent solves a local optimization problem locally, it iterates based only on the decision information of agents in its neighborhood, rather than the decision information of all agents in the cluster. This ensures the fully distributed nature of the safety control method proposed in the present invention.

[0112] This processing flow is consistent with the core feature of the present invention, namely, each intelligent agent completes security control only through local interaction, avoiding centralized global information dependence and solving the problems of limited communication range and insufficient network bandwidth.

[0113] The present invention selects the obstacle avoidance problem of 8 omnidirectional Mecanum wheeled vehicles whose kinematic models are approximately first-order integrators and avoid each other to reach their respective target points in the field as another embodiment, and combines Figure 2 、 Figure 3 、 Figure 4 and Figure 5 Details are as follows:

[0114] In the simulation environment, eight circular agents with a radius of 0.15m are considered. Simulation time , the initial moment The initial position of the agent and mission objectives The configuration is shown in Table 1.

[0115] Table 1

[0116]

[0117] Define the agent's nominal controller .use Update the agent's state, is the simulation step size.

[0118] The controller parameters of the distributed safety control simulation experiment are shown in Table 2 below.

[0119] Table 2

[0120]

[0121] Under the same initial conditions, the distributed safety controller proposed in this invention and the existing centralized safety controller and distributed safety controller are respectively applied in the multi-agent system. The motion trajectory of each agent is as follows: Figure 2 、 Figure 3 、 Figure 4 and Figure 5 As shown in the simulation results, , any two agents in the cluster and The relative distances are In the cluster, the minimum agent distances between the distributed security controller, the centralized security controller and the distributed security controller of this embodiment are 、 and Although the distributed safety controller's separation distance is slightly lower than that of the centralized solution due to sampling errors and iterative calculation errors, its actual safety distance still strictly meets the system requirements by adding a safety margin, thus ensuring the safety of the multi-agent system. This shows that the designed distributed controller can effectively coordinate local constraints while avoiding reliance on global information, achieving safety performance similar to that of the centralized solution.

[0122] Under the same simulation time and parameter settings, the agent clusters deploying the distributed safety controller, the centralized safety controller and the distributed safety controller designed by the present invention take ,and and This is because the security control decisions made by the distributed and centralized security controllers are globally optimal, while the agents deployed with the decentralized security controllers make local optimal decisions based on their own local information. Therefore, the output security control instructions are more conservative and take longer to reach the target point.

[0123] In order to further verify the feasibility of the algorithm designed in this invention, Figure 6 Physical experiments were carried out on the physical experimental platform shown in the figure. The experimental results are shown in the figure below. Figure 7 shown.

[0124] The distributed safety controller proposed in this invention is deployed on the Raspberry Pi development board of each agent. Each agent constructs a local safety problem based on its own and adjacent agent's position information and its own nominal speed, and solves it locally using the Lagrangian relaxation method. Iteration, agent use Iteration results and the iterative results of adjacent agents Calculate the Lagrange multiplier for the current iteration , and use Update Round iteration results , and then Notify neighboring agents by broadcasting. The result of the iteration As the actual speed instruction of the agent at the current moment .

[0125] Finally, the safety controller (deployed on the Raspberry Pi) outputs safety control commands via the serial port to the motion controller (deployed on the underlying development board), driving the intelligent agent's movement. This process changes the position of the intelligent agent, completing safety control at the current moment.

[0126] Physical experiments demonstrate that the distributed safety controller of this embodiment can navigate along a pre-set path from a starting point. Furthermore, when multiple agents encounter conflicting pre-set paths, the agents coordinate with each other to avoid obstacles and ultimately reach their respective destinations safely. The resulting safety decisions are globally optimal, avoiding deadlock in the safety controller when multiple agents encounter conflicting paths.

[0127] In a second aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements a distributed collaborative safety control method applied to a multi-agent system as described in any one of the first aspects above.

[0128] In a third aspect, an embodiment of the present invention provides a storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements a distributed collaborative security control method applied to a multi-agent system as described in any one of the first aspects above.

[0129] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0130] Obviously, those skilled in the art may make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if such modifications and variations fall within the scope of the claims and their equivalents, the present invention shall also include such modifications and variations.

[0131] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A distributed collaborative safety control method for a multi-agent system, characterized in that: include: S1, establish the kinematic model of the omnidirectional mobile agent and generate the nominal speed command according to the preset trajectory; S2, determine the location information of each agent and broadcast it to other agents within the preset range of each agent; S3, determining a candidate zero-crossing barrier function; S4, based on the kinematic model and the relative position of each agent and all neighboring agents within its preset range, generates safety constraints using candidate zero-crossing barrier functions; S5, based on the nominal speed, determines the loss function and solves the control quantity that minimizes the loss function while satisfying the safety constraints on each agent through distributed iteration, which is used as the safety control instruction; Each agent constructs a local safety problem based on its own and neighboring agent's position information and its own nominal velocity, and solves it iteratively using the Lagrangian relaxation method locally; Iteration, agent use Iteration results and the iterative results of adjacent agents Calculate the Lagrange multiplier for the current iteration , and use Update Round iteration results , and then Inform neighboring agents through broadcasting; S6, controls the movement of the intelligent body according to the safety control instructions; S7, update the agent position and determine whether all agents have reached the end point. If not, execute S2; if yes, end; The kinematic model is: ； in, is the position of the agent, represented by a two-dimensional column vector in the two-dimensional world coordinate system represents the position of the i-th agent, It is The first derivative of the agent's position; It is the linear velocity of the intelligent body and the control input of the system. In the two-dimensional coordinate system, it is represented by a two-dimensional column vector Indicates the The speed of each agent; The zero-crossing barrier function for: ； The constant , is the expected safe distance between two agents, is an additional safety margin to ensure robustness. , is the total number of agents in the cluster, symbol represents the two-norm of the vector, represents the position of the jth agent; Generate safety constraints based on candidate zero-crossing barrier functions, including: The safety constraint g is obtained by derivation of the zero-crossing barrier function with respect to time t as follows: ； in, is the decay rate of the barrier function, which is a constant greater than 0; is the sampling time, is the control input, ; The agent locally updates the control input and its own constraint variables as follows: Initialize the initial value of the constraint variable , solve the inner control input ; ； in, is the nominal control quantity, i Each agent provides a reference value of speed; Update outer constraint variables using gradient ascent ; ； in depending on Moment, Agent and The number of other agents in their respective neighborhoods and , defined as follows: ； make ,in is the maximum number of iterations limited by network communication delay.

2. A distributed collaborative safety control method for a multi-agent system according to claim 1, characterized in that: The loss function is: 。 3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements a distributed collaborative safety control method applied to a multi-agent system as described in any one of claims 1 to 2.

4. A storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, wherein: When the processor executes the computer program, it implements a distributed collaborative safety control method applied to a multi-agent system as described in any one of claims 1 to 2.

Images