Multi-robot safety prediction control method and device based on event triggering

Abstract

The invention relates to the technical field of mobile robot control, and discloses a multi-robot safety prediction control method and device based on event triggering, and the method comprises the steps: obtaining the real-time states of all robots and all obstacles in a target multi-robot system; based on the real-time state of each robot and each obstacle, forward propagation prediction is carried out by using a graph neural network model to obtain a control obstacle function value; calculating prediction uncertainty based on the control obstacle function value, and comparing the prediction uncertainty with a preset event triggering threshold value; if the predicted uncertainty meets a preset event triggering threshold value, updating a robot control sequence to obtain a control variable at the current moment; and controlling the robot to perform obstacle avoidance control based on the control variable at the current moment. According to the method, the resource utilization efficiency is remarkably improved while the safety and the stability of the multi-robot system are improved.

Description

Technical Field

[0001] This invention relates to the field of mobile robot control technology, specifically to a multi-robot safety predictive control method and device based on event triggering. Background Technology

[0002] Multi-robot systems are finding increasingly widespread applications in real-world scenarios such as warehousing and logistics, intelligent inspection, and unmanned delivery. In these scenarios, multi-robot systems can achieve efficient task execution and resource utilization through task decomposition and collaborative decision-making. However, with the increasing number of robots and the complexity of the environment, how to improve the system's operational efficiency and resource utilization while ensuring task completion and safety has become a current research focus. Summary of the Invention

[0003] This invention provides a multi-robot safety prediction control method and device based on event triggering, in order to solve the problems of centralized control being difficult to achieve, computational complexity rising sharply, and communication and computing resources being limited when the scale of robots in multi-robot systems increases and the complexity of the environment increases.

[0004] In a first aspect, the present invention provides an event-triggered multi-robot safety prediction and control method, the method comprising: The system acquires the real-time status of each robot and obstacle in the target multi-robot system, which is a four-wheel independently driven and four-wheel independently steering mobile robot system. Based on the real-time states of each robot and each obstacle, a graph neural network model is used to perform forward propagation prediction to obtain the control obstacle function value. The prediction uncertainty is calculated based on the control barrier function value, and then compared with the preset event trigger threshold. If the prediction uncertainty meets the preset event trigger threshold, the robot control sequence is updated to obtain the control variables at the current moment. The robot performs obstacle avoidance control based on the control variables at the current moment.

[0005] This invention provides a multi-robot safety predictive control method based on event triggering. For the real-time states of each robot and obstacle in a four-wheel independently driven, four-wheel independently steering mobile robot system, a graph neural network model is used for forward propagation prediction to obtain control obstacle function values. Based on these values, the prediction uncertainty is calculated and compared with a preset event triggering threshold. If the uncertainty meets the threshold, the robot control sequence is updated. This event-triggered mechanism allows for distributed optimization only when needed, effectively reducing network bandwidth consumption and node computational load, improving system resource utilization efficiency, and reducing computational complexity. Furthermore, by utilizing a model-based predictive control method using graph neural network control obstacle functions and event triggering, the system's communication and computational resource consumption can be significantly reduced while ensuring the safety of the multi-robot system and the completion of collaborative tasks.

[0006] Secondly, the present invention provides an event-triggered multi-robot safety prediction and control device, the device comprising: The acquisition module is used to acquire the real-time status of each robot and obstacle in the target multi-robot system; wherein, the target multi-robot system is a four-wheel independent drive and four-wheel independent steering mobile robot system; The prediction module is used to perform forward propagation prediction based on the real-time states of each robot and each obstacle, and obtain the control obstacle function value; The comparison module is used to calculate the prediction uncertainty based on the control barrier function value and compare the prediction uncertainty with a preset event trigger threshold. The update module is used to update the robot control sequence and obtain the control variables at the current moment if the prediction uncertainty meets the preset event trigger threshold. The control module is used to control the robot to perform obstacle avoidance based on the control variables at the current moment.

[0007] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the event-triggered multi-robot safety prediction control method of the first aspect or any corresponding embodiment described above.

[0008] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the event-triggered multi-robot safety prediction control method of the first aspect or any corresponding embodiment described above.

[0009] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the event-triggered multi-robot safety prediction control method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0010] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the first process of an event-triggered multi-robot safety predictive control method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a robot model in the global coordinate system and the local coordinate system according to an embodiment of the present invention; Figure 4 This is a second flowchart illustrating an event-triggered multi-robot safety predictive control method according to an embodiment of the present invention. Figure 5 This is a schematic diagram of the structure of a graph neural network model according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the third process of an event-triggered multi-robot safety predictive control method according to an embodiment of the present invention; Figure 7 This is a flowchart illustrating a method for reducing network communication resources and providing safety control for multi-robot model prediction according to an embodiment of the present invention. Figure 8 This is a schematic diagram of the training loss curve according to an embodiment of the present invention; Figure 9 These are schematic diagrams of the simulation trajectories of four robot systems according to embodiments of the present invention; Figure 10 This is a schematic diagram illustrating how a robot successfully avoids obstacles and neighboring robots during the process of completing the task of reaching the target point according to an embodiment of the present invention; Figure 11 This is a schematic diagram of the trigger interval recording of four robot systems according to an embodiment of the present invention; Figure 12 This is a schematic diagram of the simulation trajectory of four unmodified robot methods according to embodiments of the present invention; Figure 13This is a structural block diagram of an event-triggered multi-robot safety prediction and control device according to an embodiment of the present invention; Figure 14 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0014] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0015] As an optional application scenario of this invention, such as Figure 1 As shown, the multi-robot system employs a four-wheel independent drive and four-wheel independent steering mobile robot system. This system may include at least one terminal device and at least one server. Figure 1 The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0016] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0017] Model predictive control (MRC) methods are increasingly being applied in multi-robot systems due to their excellent online optimization and constraint handling capabilities. To further enhance system safety, obstacle control functions are integrated into MRC, enabling online detection and avoidance of collisions. In existing technologies, obstacle control functions designed using traditional analytical methods are effective for low-dimensional, simple structural problems. However, when facing complex dynamics, high-dimensional states, or large-scale agents, the design of traditional obstacle control functions is often difficult and lacks robustness.

[0018] To address the aforementioned issues, a control obstacle function learning method based on graph neural networks is proposed. Graph control obstacle functions can automatically obtain usable control obstacle function expressions through end-to-end training in high-dimensional, nonlinear, and complex obstacle environments, significantly improving system security and scalability. However, this method has a new application-level limitation: it focuses on immediate, local security constraints, guaranteeing only immediate safety, and insufficiently considers global performance objectives such as global path optimality and energy efficiency, and is prone to policy inconsistencies or conservative behavior. Therefore, this paper considers combining the immediate safety guarantee of graph control obstacle functions with the global optimization capability of model predictive control to achieve a synergistic improvement in system security and overall performance in complex scenarios.

[0019] To address the real-time obstacle avoidance and communication resource optimization issues in multi-robot systems, this invention provides an event-triggered multi-robot safety predictive control method. The method models robots and obstacles as graph nodes, with each node containing features such as normalized position and velocity. Inter-node interactions are represented by weighted directed edges. A pre-trained graph neural network model is used to extract node features through a message passing mechanism, constructing a collision-free safety set between the robot and obstacles, as well as other robots. This generates control obstacle function constraints, which are then embedded into the model predictive control framework to explicitly ensure real-time system safety in the optimization problem. An event-triggered mechanism based on graph neural network prediction uncertainty is designed, updating the control input only when the uncertainty exceeds a threshold, effectively reducing communication and computational resource consumption. This model prediction and safety control method can achieve real-time obstacle avoidance and safe navigation for multiple four-wheeled independently driven, four-wheeled independently steering mobile robot systems. While improving the safety and stability of multi-robot systems, it significantly improves resource utilization efficiency.

[0020] According to an embodiment of the present invention, an event-triggered multi-robot safety predictive control method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0021] This embodiment provides an event-triggered multi-robot safety predictive control method, which can be used in the aforementioned terminal device. Figure 2 This is a flowchart of an event-triggered multi-robot safety predictive control method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain the real-time status of each robot and each obstacle in the target multi-robot system; wherein, the target multi-robot system is a four-wheel independently driven and four-wheel independently steerable mobile robot system.

[0022] Specifically, for a four-wheel independently steering and four-wheel independently driven mobile robot platform, which can effectively achieve real-time obstacle avoidance and safe navigation for multiple robots, the following assumptions are made regarding the four-wheel independently driven and four-wheel independently steering wheeled mobile robot system: the four-wheel independently driven and four-wheel independently steering wheeled mobile robot moves on a flat, hard surface, and there is no vertical movement, such as... Figure 3 As shown, this is the robot's pose at a certain moment in the global coordinate system. Using the global coordinate system, For the robot's own coordinate system, For the wheel coordinate system, and This represents the robot's position in the global coordinate system. Represents the robot's heading angle in the global coordinate system. The yaw angle represents the robot's yaw angle. The longitudinal distances from the front and rear wheels to the center of the robot are: The distances from the left and right wheels to the center of the robot are the lateral distances. Representing the longitudinal velocity, lateral velocity, and angular velocity in the robot's coordinate system. Represents the linear velocity of the four wheels. The corners representing the four wheels. This represents the angular velocity of the four wheels.

[0023] Step S202: Based on the real-time states of each robot and each obstacle, forward propagation prediction is performed using a graph neural network model to obtain the control obstacle function value.

[0024] Specifically, the graph neural network model is obtained through pre-training. The specific steps include: modeling each robot and obstacle in the multi-robot system as nodes in a graph structure, with edges representing the interaction relationships between nodes; inputting the constructed graph structure into the designed graph neural network model, which is then mapped to control obstacle function values ​​through global pooling, and performing end-to-end training to minimize the mixed loss of predicted values ​​and analytical control obstacle function values; based on the pre-trained graph neural network model, constructing a graph structure input with robot and obstacle states, predicting output control obstacle function values, using these control obstacle function values ​​and their gradients as safety constraints, and then embedding these constraints into the model's predictive control framework to explicitly ensure the real-time safety of the system in the optimization problem.

[0025] Furthermore, input node features Edge features By splicing node features and edge features, and fusing local interaction information through hierarchical message passing in graph neural networks, end-to-end learning of security indicators is achieved.

[0026] Furthermore, local topology information is received as input features, including node features and edge features, where the node features include... It is a node coordinates and coordinate, yes Directional velocity and Directional velocity, Represents a robot. Representing obstacles, the robot node and obstacle node are respectively... and ; edge features The target node coordinates to source node Distance at coordinates The target node coordinates to source node Distance at coordinates It is the distance between the two nodes. It is the normalized distance between the two nodes. , Represents the sine and cosine of the azimuth angle between the robots. , Let the sine and cosine of the azimuth angle between the robot and the obstacle be denoted by , and the robot-robot edge be denoted by . Robot-Obstacle Edge .

[0027] Furthermore, the graph neural network learns the local feature information of the mobile robot in real time; the pre-trained graph neural network model is loaded, and in each control cycle, a graph structure containing node features and edge features is constructed as input based on the real-time state of the robot and obstacles. The control obstacle function value of the current robot system is predicted in real time through forward propagation. .

[0028] Step S203: Calculate the prediction uncertainty based on the control barrier function value, and compare the prediction uncertainty with the preset event trigger threshold.

[0029] Specifically, the prediction time domain of MPC (Model Predictive Control) is Np, which can solve for the control input at Np steps. However, the mechanism of MPC is to use only the first step and then repeat the solution to ensure that each step is precise control, which leads to wasted resources. Therefore, event triggering is designed to use neural networks to predict the uncertainty of the control obstacle function value. Small uncertainty means that the environment changes little, that is, the previous control input can be used.

[0030] Furthermore, during online operation, triggering conditions are designed based on the uncertainty of the graph neural network model's prediction of the current state control obstacle function value. When the prediction uncertainty of any robot exceeds a preset threshold, it indicates that the current environment has changed significantly enough. At this time, the model predictive control is updated and solved; otherwise, the existing control sequence is used.

[0031] Step S204: If the prediction uncertainty meets the preset event trigger threshold, the robot control sequence is updated to obtain the control variables at the current moment.

[0032] Step S205: Control the robot to perform obstacle avoidance control based on the control variables at the current moment.

[0033] This embodiment provides an event-triggered multi-robot safety predictive control method. For the real-time states of each robot and obstacle in a four-wheel independently driven, four-wheel independently steering mobile robot system, a graph neural network model is used for forward propagation prediction to obtain control obstacle function values. Based on these values, prediction uncertainty is calculated and compared with a preset event trigger threshold. If the prediction uncertainty meets the threshold, the robot control sequence is updated. Based on this event-triggered mechanism, distributed optimization is performed only when needed, effectively reducing network bandwidth consumption and node computational load, improving system resource utilization efficiency, and reducing computational complexity. Furthermore, by utilizing a model-based predictive control method using graph neural network control obstacle functions and event triggering, the system's communication and computational resource consumption can be significantly reduced while ensuring the safety of the multi-robot system and the completion of collaborative tasks.

[0034] This embodiment provides an event-triggered multi-robot safety predictive control method, which can be used in the aforementioned terminal device. Figure 4 This is a flowchart of an event-triggered multi-robot safety predictive control method according to an embodiment of the present invention, such as... Figure 4 As shown, the process includes the following steps: Step S401: Obtain the real-time status of each robot and obstacle in the target multi-robot system; wherein, the target multi-robot system is a four-wheel independently driven, four-wheel independently steering mobile robot system. For details, please refer to... Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0035] Step S402: Based on the real-time states of each robot and each obstacle, forward propagation prediction is performed using a graph neural network model to obtain the control obstacle function value.

[0036] Specifically, such as Figure 5 As shown, the graph neural network model includes an input feature encoding module, an input mapping module, a message passing module, an attention module, a message processing module, and a graph readout module; step S402 above includes: Step S4021: The real-time state of each robot and each obstacle is encoded into original node features by the input feature encoding module, and the original edge features are determined based on the original node features.

[0037] Specifically, the four-wheel independently driven, four-wheel independently steering wheeled mobile robot and obstacles in the environment are modeled as nodes in a graph structure. Node features include robot nodes and obstacle nodes, respectively. , Edge features include robot-robot edges and robot-obstacle edges, respectively. , Node features Sum of edge features , Indicates the number of nodes. Indicates the number of edges.

[0038] Step S4022: The original node features are mapped to the high-dimensional latent space through the input mapping module to obtain the initial node features.

[0039] Specifically, the input mapping module maps the original node features to a high-dimensional latent space, providing feature representations for subsequent message passing. The specific steps are as follows: Input Each node has 8-dimensional features, i.e. ; As input mapping weights; As a bias vector; compute node feature representations : (1) Activation process: right Perform a nonlinear transformation to output the initial node features. .

[0040] Step S4023: Based on the adjacent node features in the original edge features and initial node features, calculate the edge message features through the message passing module.

[0041] Specifically, the message passing module performs edge message calculation, calculating edge messages based on the features of adjacent nodes and edges to capture local interaction relationships; its specific steps include: Input: for Initial node characteristics of the edge, For the target node features, Edge features; concatenated to obtain: (2) MLP (Multilayer Perceptron) compresses the original 137-dimensional features to 64 dimensions (dimensionality reduction and feature extraction), extracting local features, which can be represented as: (3) Weight: As input mapping weights (that is, mapping 137-dimensional features to a 64-dimensional latent space); bias: As a bias vector; activation: right Perform nonlinear transformations; Dropout (a regularization technique for neural networks): Random deactivation during training; weights: As input mapping weights; bias: As a bias vector, MLP2 maintains its 64-dimensionality and undergoes further nonlinear transformation (feature refinement) in the 64-dimensional space to perform feature recombination and enhancement, which can be expressed as: (4) activation: right Perform nonlinear transformation; output side message characteristics .

[0042] Step S4024: The edge message features are weighted and calculated using the attention module to obtain the attention-enhanced edge message features.

[0043] Specifically, the attention application layer can be represented as: (5) Here, features represents the output of the feature transformation branch (i.e., the output of the message passing module), and scores represents the output of the attention branch. This represents the Sigmoid activation function (a type of sigmoid function). This indicates element-wise multiplication.

[0044] Furthermore, the input data is ;calculate Dimension is ,in, The calculation formula is: (6) in, (Attention branch output) during calculation, It will be broadcast to Then, element-wise multiplication is performed, with the message vector of each edge being multiplied by a scalar weight. The attention module then outputs the final value. The (weighted edge message) model now selects the importance of different edges through an attention mechanism.

[0045] Step S4025: The attention-enhanced edge message features are aggregated by the message processing module to obtain the updated node features.

[0046] In some optional implementations, step S4025 above includes: Step a1: Aggregate the attention-enhanced edge message features according to the target node index to obtain the neighborhood information features of each node.

[0047] Specifically, message aggregation involves aggregating edge messages by target node index to form a summary of neighborhood information for each node. The specific steps include: the input being edge message features. Target node index The operation is to process each edge, and... Accumulated to This is used to sum and aggregate data over the target node, which is the recipient of the message. It is an abbreviation for aggregated messages, representing the sum of all incoming edge messages received by each node. It is a scattering summation algorithm that distributes edge messages to corresponding nodes according to the target node index and accumulates them, that is: (7) Output aggregated messages .

[0048] Step a2: The neighborhood information features of each node are fused with the initial node features to obtain the fused node features.

[0049] Specifically, node updates involve fusing the node's own characteristics with aggregated neighborhood messages to update the node representation. The specific steps include: Input: The current node is in a hidden state. Aggregated neighborhood messages; splicing: MLP3 performs the following calculations: (8) activation: right Perform a nonlinear transformation; weights: As input mapping weights; bias: As a bias vector; Dropout: For random deactivation during training, MLP4 performs the following calculations: (9) activation: right Perform a nonlinear transformation; weights: As input mapping weights; bias: As a bias vector; residual: for The original information has been preserved.

[0050] Step a3: Perform layer normalization on the fused node features to obtain the updated node features.

[0051] Specifically, input The mean value is calculated along the feature dimension using the following formula: (10) Formula for calculating variance for: (11) The normalized expression is: (12) Dimension is Output the normalized node features, i.e., the updated node features. .

[0052] Step S4026: The updated node features are aggregated and mapped through the graph readout module to obtain the control barrier function value.

[0053] Specifically, node-level features (i.e., updated node features) are aggregated into graph-level representations and mapped to the final control barrier function values. The specific steps include: Input: ;calculate: Dimension is The formula for calculating MLP5 is: (13) activation: Dimension is Weight: As input mapping weights; bias: As the bias vector; the formula for calculating MLP6 is: (14) Weight: As input mapping weights; bias: As a bias vector; output the predicted control barrier function value .

[0054] Step S403: Calculate the prediction uncertainty based on the control barrier function value, and compare the prediction uncertainty with the preset event trigger threshold.

[0055] Specifically, the mean value of the control obstacle is obtained, the prediction uncertainty is calculated based on the control obstacle function value and the mean value of the control obstacle, and the prediction uncertainty is compared with the preset event trigger threshold.

[0056] Furthermore, the event-triggered mechanism integrates Dropout layers from graph neural networks to achieve uncertainty estimation: unlike related applications during the training phase, during the model deployment and inference phase, by preserving the random deactivation characteristic of Dropout layers, the same input graph structure is evaluated. implement The system will generate a random forward propagation; because the occlusion state of neurons is random in each forward propagation, the system will generate... The predicted outputs (i.e., control barrier function values) of the groups differ, and this is measured by... The degree of fluctuation in a predicted value can quantify the model's epistemic uncertainty regarding the current input: the greater the fluctuation, the lower the model's confidence in predicting the current state.

[0057] Specifically, record The next sample output is , for The mean value of the Dropout sampling output, i.e., the mean value of the control obstacle, is calculated using the following formula: (15) Furthermore, during the operation phase, the real-time control obstacle function value is acquired, and the standard deviation of the predicted control obstacle function is estimated using the following formula. As an uncertainty: (16) in, A graphical representation of the current state. It is the predicted value of the output control barrier function. The standard deviation is used to quantify the uncertainty of the model with respect to the current input.

[0058] Furthermore, the triggering conditions are as follows: (17) in, It is a preset uncertainty threshold. For the first The moment of the next trigger, index Initialize to 0.

[0059] Furthermore, the specific algorithm flow is as follows: (1) Initialization: In At that time, the graph neural network model calculates the initial control barrier function value. and uncertainty And optimize the solution of the model predictive control cost function (i.e., robot safety predictive control model); (2) loop (for each time step) ): Execution control: The first control input of the optimal control sequence calculated using the previous model predictive control; State awareness: Acquiring the new state of the system and constructing a new graph. Uncertainty: Calculate the prediction uncertainty under the current graphical state. Trigger condition: If The event triggering indicates a complex environment, meaning the previous control sequence cannot meet the obstacle avoidance requirements of the multiple robots. Therefore, it's necessary to call the graph neural network model, re-solve the cost function, and update the control sequence; otherwise... Instead of recalculating the model predictive control, the control sequence from the previous cycle is used.

[0060] Step S404: If the prediction uncertainty meets the preset event trigger threshold, the robot control sequence is updated to obtain the control variables at the current moment. For details, please refer to [link to relevant documentation]. Figure 2 Step S204 of the illustrated embodiment will not be described again here.

[0061] Step S405: Control the robot to perform obstacle avoidance based on the control variables at the current moment. See details below. Figure 2 Step S205 of the illustrated embodiment will not be described again here.

[0062] This embodiment provides an event-triggered multi-robot safety predictive control method for a four-wheel independent steering and four-wheel independent drive mobile robot platform. It can effectively achieve real-time obstacle avoidance and safe navigation for multiple robots. During online operation, triggering conditions are designed based on the uncertainty of the control obstacle function value predicted by the graph neural network model for the current state. This constraint is then embedded into the model predictive control framework, explicitly ensuring the real-time safety of the system in the optimization problem. By introducing the event-triggered mechanism into the model predictive control architecture, the solution is optimized only when necessary, effectively reducing the consumption of computational resources and balancing the computational efficiency and safety of the multi-robot system. This allows the multi-robot system to intelligently respond to changes even with limited resources.

[0063] This embodiment provides an event-triggered multi-robot safety predictive control method, which can be used in the aforementioned terminal device. Figure 6 This is a flowchart of an event-triggered multi-robot safety predictive control method according to an embodiment of the present invention, such as... Figure 6 As shown, the process includes the following steps: Step S601: Obtain the real-time status of each robot and obstacle in the target multi-robot system; wherein, the target multi-robot system is a four-wheel independently driven, four-wheel independently steering mobile robot system. For details, please refer to... Figure 4 Step S401 of the illustrated embodiment will not be described again here.

[0064] Step S602: Based on the real-time states of each robot and each obstacle, a graph neural network model is used to perform forward propagation prediction to obtain the control obstacle function values. For details, please refer to [link to details]. Figure 4 Step S402 of the illustrated embodiment will not be described again here.

[0065] Step S603: Calculate the prediction uncertainty based on the control barrier function value, and compare the prediction uncertainty with a preset event trigger threshold. For details, please refer to [link to relevant documentation]. Figure 4 Step S403 of the illustrated embodiment will not be described again here.

[0066] Step S604: If the prediction uncertainty meets the preset event trigger threshold, the robot control sequence is updated to obtain the control variables at the current moment.

[0067] Specifically, step S604 includes: Step S6041: Construct the kinematic model of the four-wheel independently driven and four-wheel independently steered mobile robot.

[0068] Specifically, the kinematic equations of the four-wheel independently driven and four-wheel independently steering mobile robot are as follows, that is, the expression of the kinematic model of the four-wheel independently driven and four-wheel independently steering mobile robot is as follows: (18) in, Position the robot in the global coordinate system velocity in direction, Position the robot in the global coordinate system velocity in direction, The robot's heading angle in the global coordinate system rate of change, , , , , and Indicates intermediate variables. Let the coordinates of the four wheels be in the robot's coordinate system. Represents the linear velocity of the four wheels. The corners representing the four wheels. This represents the robot's heading angle in the global coordinate system.

[0069] Step S6042: Establish a robot safety predictive control model based on the kinematic model of the four-wheel independently driven and four-wheel independently steering mobile robot.

[0070] In some optional implementations, step S6042 above includes: Step b1: Establish a robot motion prediction model based on the kinematic model of the four-wheel independently driven and four-wheel independently steering mobile robot.

[0071] Specifically, for ease of derivation, the kinematic model of the above four-wheel independently driven and four-wheel independently steering mobile robot is denoted as: (19) in, For state variables, For input variables, Represents state variables The derivative with respect to time.

[0072] Furthermore, by linearizing the above nonlinear kinematic equations and obtaining their Taylor expansion, retaining the first-order terms and ignoring the higher-order terms, we can obtain: (20) in, Represents the linearized state variables. This represents the linearized input variable.

[0073] Furthermore, while linearizing, the first-order quotient method is used to discretize formula (20), transforming the nonlinear model of four-wheel independent drive and four-wheel independent steering into a linear time-varying model: (twenty one) In the above formula, , , For function Regarding state variables Jacobian matrix, for control variables Jacobian matrix, To use time intervals, The identity matrix is ​​the one corresponding to the dimension. and Indicates an intermediate variable.

[0074] The original system state was: (twenty two) In the above formula, It represents the 3-dimensional real space.

[0075] Extended state variables for: (twenty three) in, The control parameters are the four-wheel steering angle and the four-wheel drive speed.

[0076] Controlling the amount of change for: (twenty four) Introducing the above extended state variables into the control change variables yields: (25) in, Indicates the original system state. , Indicates the control dimension. , For the system augmentation matrix, The input is the augmented matrix.

[0077] (26) (27) In the above formula, The dimension is The identity matrix.

[0078] The system's output equation is: (28) in, Indicates in The output vector predicted at each time step, i.e., the prediction of the system output. Represents the output matrix of the control system. It includes the dynamic relationships of the system, describing the relationship between inputs and states. Indicates in The prediction results of the state variables at time 1.

[0079] Based on the above steps, the prediction model of the model predictive controller can be obtained: (29) in, This represents the predicted time-domain output vector. ; Represents the state mapping matrix, ; This represents a sequence of control changes. ; This indicates the control input increment in the prediction model.

[0080] Based on the expression of the prediction model of the above-mentioned predictive controller, the following objective function is established: (30) in, , These are the weight matrices for the state variables and the control increments, respectively. Represents system state limitations. This indicates a limitation on the control increment, ensuring the continuity of the controller and the responsiveness of the actuator.

[0081] Step b2: Based on the robot motion prediction model, with the goal of minimizing cost, the control increment to be solved is used as the optimization variable to construct the objective function.

[0082] Specifically, in multi-robot systems there are There are 1 robot, and each robot predicts a step size of 1. Their respective target points are By combining the prediction outputs of all robots, the global prediction vector is defined as: (31) in, For the first The output sequence of a robot in the prediction time domain.

[0083] Furthermore, the system's global target vector is composed of the sub-targets of all robots, defined as: (32) in, Indicates the first The target state of the robot.

[0084] Furthermore, the cost function simultaneously minimizes the target error and control increment of all robots: (33) in, Here is the state weight matrix for each robot. This is the control increment weight matrix for each robot.

[0085] Furthermore, substituting the above prediction model into the above cost function, that is, substituting the above formula (29) into the above formula (33), we can obtain: (34) in, , , The weight matrix represents the output error. This represents the weight matrix that controls the increment.

[0086] The objective function in the standard quadratic programming form is as follows: (35) in, , , Represents the Hessian matrix, This represents a vector of linear terms.

[0087] Step b3: Construct robot obstacle avoidance safety constraints based on the control obstacle function values.

[0088] Specifically, based on a pre-trained graph neural network model, a graph structure input is constructed using the robot's state and the obstacle's state. Neighborhood interaction information is aggregated through a message passing mechanism to predict the output control obstacle function value. This control obstacle function value and its gradient are then used as safety constraints.

[0089] Furthermore, assume that all obstacles in the environment can be considered as having a position in a global coordinate system. , radius is For any circular obstacle, with respect to any initial state variable... The state variables include the heading angle and coordinates of the robot or obstacle, if there exists a very large time interval. This ensures that the system state always remains within a safe set. Inside, it is called a safe set. It has forward invariance, when the set When a system possesses this property, it adheres to the security constraints of that set. The condition for a set to satisfy these security constraints can be expressed as: (36) in, Representing system safety constraints, it is essentially a mapping function that maps the reachable set of system states to the real number field. Represents a state variable.

[0090] Furthermore, for multi-robot systems, the collision-free safety set is represented as: (37) in, and Number the robot. , Represents robots and Safety constraint functions between them.

[0091] Furthermore, in the sensing radius The robot inside is the neighbor robot. ,Right now: (38) in, and They are robots and The location.

[0092] Furthermore, The graph neural network model is a graph representation of the current state. Its output control barrier function prediction value is , As a graph control barrier function value based on graph neural networks, if a control input set exists... and , making all There is a control input satisfy: (39) in, It is a strictly increasing continuous function that satisfies And it usually has linear or Lipschitz properties (Lipschitz continuity) to adjust the safe convergence rate; Indicates the radius of induction Neighbor robots At any moment State variables, Indicates the radius of induction Neighbor robots At any moment Input variables.

[0093] Among them, discrete difference Defined as the forward interpolation predicted by a graph neural network model: (40) System dynamics (i.e., the state of a multi-robot system) are described by the following formula: (41) Furthermore, the above formula (39) is used as a safety constraint to ensure robot obstacle avoidance, namely robot obstacle avoidance safety constraint.

[0094] Step b4: Obtain the control variable constraints, control increment constraints, and robot terminal set constraints. Use the robot obstacle avoidance safety constraints, control variable constraints, control increment constraints, and robot terminal set constraints as the constraints of the objective function to obtain the robot safety predictive control model.

[0095] Specifically, due to actuator constraints, the expressions for the constraints (i.e., control variable constraints) of the linear and angular velocities of the four wheels are as follows: (42) in, , , This is the Kronecker product (a special matrix operation in linear algebra). It is an identity matrix with dimension 8. For the control input of the system at the previous moment, This represents a lower triangular matrix of all 1s. This represents the control time domain, used to predict control actions within a certain number of future steps.

[0096] Furthermore, the expressions for the constraints on the velocity increments and angular velocity increments of the four wheels (i.e., control increment constraints) are as follows: (43) Furthermore, the expression for the terminal set constraint to ensure the robot's operational tasks is as follows: (44) in, This represents the set of terminal constraints.

[0097] Furthermore, based on the prediction model and the initial control increment sequence, an objective function is established, along with the objective constraints. The expression for the robot safety predictive control model is as follows: (45) (46) (47) (48) (49) Step S6043: Solve the robot safety predictive control model to obtain the control increment sequence at the current moment.

[0098] Specifically, with the objective function, constraints, and safety constraints on the control barrier function input to state safety, the optimal control increment sequence is obtained. The problem of solving the problem is transformed into an optimization control problem. In the optimization solution process, a state error triggering mechanism is introduced.

[0099] Furthermore, the cost at the current time step is determined based on the objective function. With the goal of minimizing the cost, the objective function is solved based on the objective constraints to obtain the control increment sequence at the current time step, i.e., at each time step... Based on the objective function described above, the cost at the current moment is determined. With the goal of minimizing the cost, the objective function is solved based on the objective constraints to obtain the control increment sequence at the current moment. : (50) Step S6044: Obtain the control variable from the previous time step, add the first control variable in the control increment sequence at the current time step to the control variable from the previous time step, and obtain the control variable at the current time step.

[0100] Specifically, by adding the first element of the optimal control increment sequence to the control input at the previous time step, we can obtain the system at each current time step. System input at time for: (51) Step S605: Control the robot to perform obstacle avoidance control based on the control variables at the current moment.

[0101] Specifically, by inputting control inputs into the robot system at each sampling moment, safe obstacle avoidance control of the robot can be achieved.

[0102] The event-triggered multi-robot safety predictive control method provided in this embodiment utilizes a graph neural network to learn the graph neural network control obstacle function to obtain safety constraints, thereby achieving collision avoidance for multiple robots. Furthermore, the safety constraints, control variable constraints, control increment constraints, and terminal set constraints are used as constraints of the objective function to ensure safe obstacle avoidance for the robots. Then, the control increment sequence is obtained by solving the problem. The first control variable in the control increment sequence at the current moment is added to the control variable at the previous moment to obtain the control variable at the current moment. Using the control variable at the current moment to control the robot for obstacle avoidance control can effectively achieve real-time obstacle avoidance and safe navigation for multiple robots.

[0103] The following specific embodiments illustrate the detailed steps and effectiveness of an event-triggered multi-robot safety predictive control method.

[0104] Example 1: A multi-robot model prediction and safety control method that reduces network communication resources, applicable to four-wheel independent drive and four-wheel independent steering mobile robot systems, such as... Figure 7 As shown, the specific steps of the multi-robot model prediction and safety control method include: Step 1: Model each robot and obstacle in the multi-robot system as nodes in a graph structure, with edges representing the interaction relationships between nodes; Step 2: Input the constructed graph structure into the designed graph neural network model, and finally map it to the control barrier function value through global pooling. Perform end-to-end training to minimize the mixed loss of the predicted value and the analytical control barrier function value. Step 3: Based on the pre-trained graph neural network model, construct a graph structure input with the robot and obstacle states, predict the output control obstacle function value, use the control obstacle function value and its gradient as safety constraints, and then embed the constraints into the model predictive control framework to explicitly ensure the real-time safety of the system in the optimization problem. Step 4: During online operation, design triggering conditions based on the uncertainty of the control barrier function value predicted by the graph neural network model for the current state; Step 5: For a four-wheel independent steering and four-wheel independent drive mobile robot platform, this method can effectively achieve real-time obstacle avoidance and safe navigation for multiple robots.

[0105] Example 2: The robot's sensing radius is 3 meters. Each robot only considers the environment within its 3.0-meter range. The increased prediction uncertainty of the graph neural network model indicates that the robot's perception range is limited when dealing with obstacles or other robots. Each robot is triggered independently and adapts to changes in the local environment. The simulation of the four-wheel independent drive and four-wheel independent steering mobile robot system is carried out. Specifically, in order to verify the effectiveness of the multi-robot model prediction and safety control method, an obstacle avoidance control verification example is designed in MATLAB (a commercial data software) for the four-wheel independent drive and four-wheel independent steering mobile robot system.

[0106] The simulation settings include a robot radius of 0.2 meters, an obstacle radius of 0.4 meters, a safety boundary of 0.3 meters, a sensing radius of 3 meters, a prediction step size of 5, a sampling time of 0.08 seconds, and an event triggering mechanism with an uncertainty threshold of 0.08. During non-triggering moments, the control input from the previous optimization solution is reused to reduce computational overhead. Each robot makes independent decisions within the system, which consists of 4 robots and 4 obstacles, with 8 nodes and 44 robot-to-robot and robot-to-obstacle edges.

[0107] Figure 8 To train a loss graph between predicted values ​​and geometric control obstacle function values, using the geometric control obstacle function value as the supervision signal, a graph structure pre-constructed based on the robot and obstacles is divided into a training set and a validation set. A graph neural network model is used to train the training set and the validation set respectively to obtain the loss between the predicted value of the control obstacle function and the geometric control obstacle function value, and then a training loss graph between the predicted value and the geometric control obstacle function value is drawn.

[0108] Figure 9 To set up simulation trajectories for four four-wheeled independently driven and four-wheeled independently steering mobile robots, each moving from its own starting point to its own target point. Figure 9 In the diagram, A represents the positional relationship between robot 1 and the obstacle. Figure 10 For an enlarged view of A, such as Figure 10 As shown, the four robots successfully avoided obstacles and neighboring robots while completing the task of reaching the target point. Figure 11 The diagram shows the trigger interval records for four robot systems based on an event-triggered mechanism, as follows: Figure 11 As shown, the solution time interval of the model predictive control method without event triggering is 0.08s, which can be seen as effectively reducing the number of triggers and saving network resources.

[0109] Figure 12 To use relevant control barrier functions, such as Figure 12 As shown, the simulation trajectory diagram of the model predictive controller without event triggering mechanism is based solely on the raw physical data such as the distance and speed of obstacles, without performing node feature aggregation of graph neural networks and focusing on key obstacles. It can be seen that robots 1 and 4 can complete obstacle avoidance in environments with few obstacles. However, in scenarios with narrow passages and multiple obstacles, robots 1 and 3 cannot avoid obstacles in time due to simultaneously dealing with multiple low-risk constraints, resulting in them entering unsafe areas. Since no event triggering mechanism is designed, complete model predictive control and optimization calculations must be performed at each sampling time, which makes the time for the robots to reach the target point longer than the simulation time of the method of this invention.

[0110] This embodiment also provides an event-triggered multi-robot safety prediction and control device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0111] This embodiment provides an event-triggered multi-robot safety prediction and control device, such as... Figure 13 As shown, it includes: The acquisition module 1301 is used to acquire the real-time status of each robot and each obstacle in the target multi-robot system; wherein, the target multi-robot system is a four-wheel independent drive and four-wheel independent steering mobile robot system; Prediction module 1302 is used to perform forward propagation prediction based on the real-time states of each robot and each obstacle, and obtain the control obstacle function value. Comparison module 1303 is used to calculate the prediction uncertainty based on the control barrier function value and compare the prediction uncertainty with a preset event trigger threshold; The update module 1304 is used to update the robot control sequence if the prediction uncertainty meets the preset event trigger threshold, so as to obtain the control variables at the current moment. The control module 1305 is used to control the robot to perform obstacle avoidance based on the control variables at the current moment.

[0112] The event-triggered multi-robot safety predictive control device provided in this embodiment of the invention can execute the event-triggered multi-robot safety predictive control method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the various modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0113] Figure 14 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0114] The following is a detailed reference. Figure 14 The diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 1401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1402 or a program loaded from memory 1408 into random access memory (RAM) 1403. The RAM 1403 also stores various programs and data required for the operation of the electronic device. The processor 1401, ROM 1402, and RAM 1403 are interconnected via a bus 1404. An input / output (I / O) interface 1405 is also connected to the bus 1404.

[0115] Typically, the following devices can be connected to I / O interface 1405: input devices 1406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 1407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 1408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1409. Communication device 1409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 14 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0116] Figure 14 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0117] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, it implements the event-triggered multi-robot safety predictive control method shown in the above embodiments.

[0118] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0119] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A multi-robot safety predictive control method based on event triggering, characterized in that, The method includes: The real-time status of each robot and obstacle in the target multi-robot system is obtained; wherein, the target multi-robot system is a four-wheel independently driven and four-wheel independently steering mobile robot system; Based on the real-time states of each robot and each obstacle, a graph neural network model is used to perform forward propagation prediction to obtain the control obstacle function value. The prediction uncertainty is calculated based on the control barrier function value, and the prediction uncertainty is compared with a preset event trigger threshold. If the prediction uncertainty meets the preset event trigger threshold, the robot control sequence is updated to obtain the control variables at the current moment. The robot performs obstacle avoidance control based on the control variables at the current moment.

2. The method according to claim 1, characterized in that, The graph neural network model includes an input feature encoding module, an input mapping module, a message passing module, an attention module, a message processing module, and a graph readout module; the step of using the graph neural network model to perform forward propagation prediction based on the real-time states of each robot and each obstacle to obtain the control obstacle function value includes: The input feature encoding module encodes the real-time states of each robot and obstacle into raw node features, and determines raw edge features based on the raw node features. The input mapping module maps the original node features to a high-dimensional latent space to obtain the initial node features. Based on the adjacent node features in the original edge features and the initial node features, the edge message features are calculated through the message passing module. The attention module performs weighted calculations on the edge message features to obtain attention-enhanced edge message features. The message processing module performs message aggregation on the attention-enhanced edge message features to obtain updated node features. The updated node features are aggregated and mapped using the graph readout module to obtain the control barrier function value.

3. The method according to claim 2, characterized in that, The step of aggregating the attention-enhanced edge message features through the message processing module to obtain updated node features includes: The attention-enhanced edge message features are aggregated according to the target node index to obtain the neighborhood information features of each node; The neighborhood information features of each node are fused with the initial node features to obtain the fused node features; The fused node features are subjected to layer normalization to obtain the updated node features.

4. The method according to claim 1, characterized in that, The step of calculating the prediction uncertainty based on the control barrier function value and comparing the prediction uncertainty with a preset event trigger threshold includes: Obtain the mean value of the control obstacle, calculate the prediction uncertainty based on the control obstacle function value and the mean value of the control obstacle, and compare the prediction uncertainty with a preset event trigger threshold.

5. The method according to claim 1, characterized in that, If the prediction uncertainty meets the preset event trigger threshold, the robot control sequence is updated to obtain the control variables at the current moment, including: Construct a kinematic model for a four-wheel independently driven and four-wheel independently steering mobile robot; A robot safety predictive control model is established based on the kinematic model of the four-wheel independently driven and four-wheel independently steering mobile robot. Solving the robot safety predictive control model yields the control increment sequence at the current moment; Obtain the control variable from the previous time step, and add the first control variable in the control increment sequence at the current time step to the control variable from the previous time step to obtain the control variable at the current time step.

6. The method according to claim 5, characterized in that, The establishment of a robot safety predictive control model based on the kinematic model of the four-wheel independently driven and four-wheel independently steering mobile robot includes: A robot motion prediction model is established based on the kinematic model of the four-wheel independently driven and four-wheel independently steering mobile robot. Based on the robot motion prediction model, with the goal of minimizing cost, the control increment to be solved is used as the optimization variable to construct the objective function; Based on the control obstacle function value, construct robot obstacle avoidance safety constraints; Obtain the control variable constraints, control increment constraints, and robot terminal set constraints. Use the robot obstacle avoidance safety constraints, control variable constraints, control increment constraints, and robot terminal set constraints as constraints of the objective function to obtain the robot safety predictive control model.

7. A multi-robot safety predictive control device based on event triggering, characterized in that, The device includes: The acquisition module is used to acquire the real-time status of each robot and each obstacle in the target multi-robot system; wherein, the target multi-robot system is a four-wheel independently driven and four-wheel independently steering mobile robot system; The prediction module is used to perform forward propagation prediction based on the real-time states of each robot and each obstacle, and obtain the control obstacle function value. The comparison module is used to calculate the prediction uncertainty based on the control barrier function value and compare the prediction uncertainty with a preset event trigger threshold. The update module is used to update the robot control sequence if the prediction uncertainty meets the preset event trigger threshold, so as to obtain the control variables at the current moment. The control module is used to control the robot to perform obstacle avoidance based on the control variables at the current moment.

8. An electronic device, characterized in that, include: A memory and a processor are interconnected, the memory storing computer instructions, and the processor executing the computer instructions to perform the event-triggered multi-robot safety predictive control method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the event-triggered multi-robot safety predictive control method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions for causing a computer to execute the event-triggered multi-robot safety predictive control method according to any one of claims 1 to 6.

Images