Multi-robot collaborative control method and system based on edge computing

Abstract

The present invention provides a multi-robot collaborative control method and system based on edge computing, which relates to the field of multi-robot control technology, including dividing a search and rescue area into multiple sub-areas, generating sub-area search and rescue tasks, generating an optimal matching scheme between a ground robot and a sub-area search and rescue task according to the sub-area search and rescue tasks and the capability parameters and current status of each ground robot, adopting different search and rescue strategies for different types of ground robots, and generating a heterogeneous robot collaborative search and rescue strategy by defining a formal language and lexical structure for collaborative control of each ground robot; each ground robot performs the search and rescue task in the corresponding sub-area according to the heterogeneous robot collaborative search and rescue strategy, and when it is detected that the resource usage or network status meets the preset conditions, a calculation offloading decision is initiated, and part of the search and rescue task is offloaded to other computing nodes according to the decision result to continue the search and rescue operation until the search and rescue task is completed.

Description

Technical Field

[0001] The present invention relates to multi-robot control technology, and in particular to a multi-robot collaborative control method and system based on edge computing. Background Art

[0002] With the development of robotics and wireless communication technologies, multi-robot collaborative search and rescue has received increasing attention. At the disaster site, using multiple robots for collaborative search and rescue can significantly improve search and rescue efficiency and reduce casualties. However, multi-robot collaborative search and rescue faces many technical challenges, such as autonomous navigation in complex environments, real-time obstacle avoidance, coordinated control between heterogeneous robots, and processing of massive sensor data.

[0003] Traditional multi-robot collaborative search and rescue control solutions mainly rely on centralized control, where a central controller uniformly dispatches and controls all robots. This solution has problems such as high computing and communication pressure, slow response speed, poor robustness and scalability, and is difficult to meet the real-time and reliability requirements in complex search and rescue environments.

[0004] In order to overcome the above problems, some scholars have proposed using edge computing technology to sink computing, storage and communication capabilities to the local robots to achieve distributed multi-robot collaborative control. By deploying computing and storage resources on the robot side, edge computing can achieve local data processing and real-time response, reducing the burden on the central controller. At the same time, edge computing supports direct communication and collaboration between heterogeneous robots, improving the flexibility and scalability of the system.

[0005] However, the existing multi-robot collaborative control scheme based on edge computing still has some shortcomings. On the one hand, due to the limitation of the robot's own computing and storage resources, the processing of complex tasks still faces computing bottlenecks, and it is necessary to introduce more efficient computing offloading and task decomposition mechanisms. On the other hand, the robot may encounter unexpected situations such as failure or energy exhaustion during the execution of the task, and it is necessary to design efficient task migration and fault-tolerant mechanisms to ensure the continuity and reliability of the search and rescue mission.

[0006] Therefore, the present invention proposes an improved multi-robot collaborative control scheme based on edge computing, which can further improve the computational efficiency, task continuity and system reliability of multi-robot collaborative search and rescue. Summary of the invention

[0007] The embodiments of the present invention provide a multi-robot collaborative control method and system based on edge computing, which can solve the problems in the prior art.

[0008] According to a first aspect of the embodiments of the present invention,

[0009] A multi-robot collaborative control method based on edge computing is provided, comprising:

[0010] The environmental characteristics and search and rescue needs of the search and rescue area are obtained through drones, the search and rescue area is divided into multiple sub-areas, and sub-area search and rescue tasks are generated. At the same time, the capability parameters and current status of each ground robot are obtained. According to the sub-area search and rescue tasks and the capability parameters and current status of each ground robot, the optimal matching scheme between the ground robot and the sub-area search and rescue tasks is generated;

[0011] The sub-region search and rescue mission and the optimal matching solution are sent to the corresponding ground robot. Different search and rescue strategies are adopted for different types of ground robots. Multiple different types of ground robots are modeled as a discrete event system subsystem. By defining the formal language and lexicon of the collaborative control of various ground robots, a collaborative search and rescue strategy for heterogeneous robots is generated.

[0012] Each ground robot performs search and rescue tasks in the corresponding sub-area according to the heterogeneous robot collaborative search and rescue strategy, and monitors the computing resource usage and network status of the drone node in real time. When it is detected that the resource usage or network status meets the preset conditions, the computing offloading decision is initiated, and part of the search and rescue tasks are offloaded to other computing nodes according to the decision results to continue the search and rescue operation until the search and rescue mission is completed.

[0013] In an optional embodiment,

[0014] According to the sub-region search and rescue mission and the capability parameters and current status of each ground robot, the optimal matching scheme between the ground robot and the sub-region search and rescue mission is generated, including:

[0015] According to the capability parameters and current status of each ground robot, the utility value of each ground robot in performing search and rescue tasks in different sub-areas is calculated;

[0016] According to the utility value of each ground robot performing search and rescue tasks in different sub-areas, the scheme matching objective function is constructed by considering the task completion time and energy consumption indicators;

[0017] A multi-objective optimization model for the multi-robot task allocation problem is constructed based on the solution matching objective function, where the decision variable is the allocation matrix of the ground robot and the sub-region search and rescue tasks, and the constraints include that at least one robot is allocated to each sub-region and each robot can only be allocated to one sub-region.

[0018] By solving the multi-objective optimization model of the multi-robot task allocation problem, an optimal task allocation solution is obtained.

[0019] In an optional embodiment,

[0020] By solving the multi-objective optimization model of the multi-robot task allocation problem, the optimal task allocation solution is obtained, including:

[0021] The allocation matrix between the ground robot and the sub-region search and rescue task is represented as a real-valued matrix, and a certain number of particles are randomly generated, where each particle represents a possible task allocation scheme;

[0022] For each particle, calculate its fitness value on the solution matching objective function, and compare the dominance relationship between the current position of each particle and its individual optimal solution, and update the individual optimal solution according to the first comparison result; compare the dominance relationship between the individual optimal solutions of all particles and the global optimal solution, and update the global optimal solution according to the second comparison result;

[0023] For each particle, according to its current speed, the updated individual optimal solution and the updated global optimal solution, update its speed and position to obtain an updated particle;

[0024] The original particles and the updated particles are merged to obtain an extended population, and the extended population is non-dominated sorted to obtain non-dominated solution sets of different levels;

[0025] Repeat the iteration until the termination condition is met, then output the current non-dominated solution set as the optimal solution set, and select the final optimal task allocation plan from the obtained optimal solution set according to the decision maker's preference.

[0026] In an optional embodiment,

[0027] Multiple different types of ground robots are modeled as a discrete event system subsystem. By defining the formal language and lexicon of the collaborative control of various ground robots, the heterogeneous robot collaborative search and rescue strategy is generated, including:

[0028] According to the task requirements and the capabilities of various ground robots, a series of language samples for communication interaction and behavioral coordination are generated and stored in the collaborative control language corpus;

[0029] Preprocess the collaborative control language corpus to form a standardized data set, and use word embedding technology to convert language samples into language sample embedding vectors;

[0030] A semantic neural network model based on an encoder-decoder architecture is constructed, wherein the semantic neural network model includes an encoder and a decoder. A task requirement encoder and a ground robot capability encoder based on a graph neural network are introduced into the encoder to encode the task requirement graph and the ground robot capability graph to obtain an embedded representation of each task node and ground robot node. At the same time, a pointer network-based copy mechanism is introduced into the decoder to calculate the attention distribution according to the current decoder and encoder outputs to obtain the probability of copying each position from the input sequence of the semantic neural network model.

[0031] The language sample embedding vector is input into the constructed semantic neural network model to generate a language symbol string for the collaborative control of heterogeneous robots. A semantic verification mechanism based on knowledge graph is introduced to check the semantic consistency and finally a formal description of the collaborative search and rescue strategy of heterogeneous robots is output.

[0032] In an optional embodiment,

[0033] When it is detected that the resource usage or network status meets the preset conditions, the computing offloading decision is initiated, and part of the search and rescue task is offloaded to other computing nodes according to the decision result to continue the search and rescue operation until the search and rescue task is completed, including:

[0034] The computation offloading decision problem is modeled as an offloading optimization model, where the task offloading objective function is to minimize the task execution time and energy consumption, the constraints are the task completion time and energy consumption limits, and the decision variables are the offloading decision of each task and the selection of the offloading target node;

[0035] The optimization algorithm is used to calculate the offloading optimization model and obtain the optimal calculation offloading plan. According to the optimal calculation offloading plan, part of the search and rescue mission is offloaded from the UAV to the selected ground robot node.

[0036] The selected ground robot node receives the data of the offloading task, performs the offloading task calculation, and returns the offloading task calculation result to the UAV;

[0037] The drone integrates the received calculation results of the offloaded mission with the results of other missions executed locally to form a complete search and rescue plan, and continues to command the ground robot to execute it until the search and rescue mission is completed.

[0038] In an optional embodiment,

[0039] The formula of the task offloading objective function is as follows:

[0040] ;

[0041] Where J represents the offloading optimization target, M represents the number of search and rescue missions, i represents the search and rescue mission index, N represents the number of ground robot nodes, j represents the ground robot node index, and x represents the number of ground robot nodes. ij represents a binary variable, t ij represents the time required for task i to be executed on ground robot node j, t i represents the time required for task i to execute on the local node, λ represents the trade-off coefficient between time and energy consumption, and e ij represents the energy consumed by executing task i on ground robot node j, e i It represents the energy consumed by executing task i on the local node.

[0042] In an optional embodiment,

[0043] The method further comprises:

[0044] Constructing a local task state model for the drone node and the backup node, wherein the input of the local task state model includes the attribute characteristics of the search and rescue task, the assigned search and rescue task list, and the completion progress of each search and rescue task, and the output is the completion probability and the estimated remaining time of the search and rescue task;

[0045] Build a federated learning framework between drone nodes and backup nodes. During the task execution, drone nodes and backup nodes use the task status data collected locally to obtain the model update through gradient calculation, and transmit the model update to the federated learning framework to obtain a global update solution.

[0046] The local task state model is updated using the global update scheme to obtain a new local task state model, and the iteration is repeated until the local task state model converges;

[0047] The drone node periodically sends heartbeat packets to the backup node, which monitors the status of the drone through the heartbeat packets. When the backup node fails to receive the heartbeat packets for multiple consecutive times, it determines that the drone is faulty and automatically triggers task migration.

[0048] During task migration, the backup node uses its own local task state model to continue executing the task. After the task migration is completed, the model parameters of the backup node's own local task state model are sent to the federated learning framework, and the received model parameters are distributed to other backup nodes through the federated learning framework.

[0049] According to a second aspect of the embodiments of the present invention,

[0050] Provided is a multi-robot collaborative control system based on edge computing, characterized in that it includes:

[0051] The first unit is used to obtain the environmental characteristics and search and rescue needs of the search and rescue area through the UAV, divide the search and rescue area into multiple sub-areas, generate sub-area search and rescue tasks, and simultaneously obtain the capability parameters and current status of each ground robot. According to the sub-area search and rescue tasks and the capability parameters and current status of each ground robot, an optimal matching scheme between the ground robot and the sub-area search and rescue tasks is generated;

[0052] The second unit sends the sub-area search and rescue tasks and the optimal matching solutions to the corresponding ground robots. Different search and rescue strategies are adopted for different types of ground robots. Multiple different types of ground robots are modeled as a discrete event system subsystem. By defining the formal language and lexicon of the collaborative control of various ground robots, a collaborative search and rescue strategy for heterogeneous robots is generated.

[0053] In the third unit, each ground robot performs search and rescue tasks in the corresponding sub-area according to the heterogeneous robot collaborative search and rescue strategy, and monitors the computing resource usage and network status of the drone node in real time. When it is detected that the resource usage or network status meets the preset conditions, the computing offloading decision is initiated, and part of the search and rescue task is offloaded to other computing nodes according to the decision result to continue the search and rescue operation until the search and rescue mission is completed.

[0054] According to a third aspect of the embodiments of the present invention,

[0055] An electronic device is provided, comprising:

[0056] processor;

[0057] a memory for storing processor-executable instructions;

[0058] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0059] A fourth aspect of the embodiments of the present invention is:

[0060] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0061] In this embodiment, by using drones to obtain the environmental characteristics and requirements of the search and rescue area, the search and rescue area is reasonably divided into multiple sub-areas, and the optimal matching scheme between the ground robot and the sub-area is generated based on the multi-objective optimization model, which can make full use of robot resources and improve the efficiency and quality of search and rescue. Different types of ground robots are modeled as discrete event system subsystems, and heterogeneous robot collaborative search and rescue strategies are generated by defining formal languages ​​and lexical grammars, so as to realize collaborative control between different robots, give play to their respective advantages, and improve the search and rescue capabilities. The computing resource usage and network status of the drone nodes are monitored in real time. When resources are tight, the computing offloading decision is initiated, and part of the search and rescue tasks are offloaded to other computing nodes for continued execution, which avoids resource waste and search and rescue interruption and improves the utilization efficiency of computing resources. By introducing advanced technologies such as intelligent algorithms, multi-robot collaboration, and computing offloading, the efficiency and quality of search and rescue can be improved, the utilization of computing resources can be optimized, and the intelligence level and autonomy of the system can be enhanced, which has strong practical value and development prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 Schematic diagram of the process of a multi-robot collaborative control method based on edge computing according to an embodiment of the present invention;

[0063] Figure 2 This is a schematic diagram of the structure of a multi-robot collaborative control system based on edge computing in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0066] Figure 1 Schematic diagram of the process of a multi-robot collaborative control method based on edge computing according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0067] S101. Obtain the environmental characteristics and search and rescue needs of the search and rescue area through the drone, divide the search and rescue area into multiple sub-areas, generate sub-area search and rescue tasks, and simultaneously obtain the capability parameters and current status of each ground robot. Based on the sub-area search and rescue tasks and the capability parameters and current status of each ground robot, generate the optimal matching solution between the ground robot and the sub-area search and rescue tasks.

[0068] Exemplarily, first, an adaptive region division algorithm based on terrain features and search and rescue needs is designed. The drone obtains terrain images and elevation data of the search and rescue area through airborne sensors, and uses image segmentation and feature extraction techniques to identify different terrain types (such as plains, hills, mountains, etc.) and key objects (such as buildings, roads, waters, etc.). Then, combined with the priority of the search and rescue mission and the target distribution information, a graph-based clustering algorithm is used to adaptively divide the search and rescue area. The algorithm constructs a regional similarity matrix, considers multiple factors such as terrain complexity, obstacle density, and search and rescue priority, and aggregates regions with high similarity into a sub-region. Finally, according to the perception range and flight altitude of the drone, the size and shape of the sub-region are optimized to generate a regional division scheme suitable for multi-robot collaborative search and rescue.

[0069] In an optional embodiment,

[0070] According to the sub-region search and rescue mission and the capability parameters and current status of each ground robot, the optimal matching scheme between the ground robot and the sub-region search and rescue mission is generated, including:

[0071] According to the capability parameters and current status of each ground robot, the utility value of each ground robot in performing search and rescue tasks in different sub-areas is calculated;

[0072] According to the utility value of each ground robot performing search and rescue tasks in different sub-areas, the scheme matching objective function is constructed by considering the task completion time and energy consumption indicators;

[0073] A multi-objective optimization model for the multi-robot task allocation problem is constructed based on the solution matching objective function, where the decision variable is the allocation matrix of the ground robot and the sub-region search and rescue tasks, and the constraints include that at least one robot is allocated to each sub-region and each robot can only be allocated to one sub-region.

[0074] By solving the multi-objective optimization model of the multi-robot task allocation problem, an optimal task allocation solution is obtained.

[0075] Exemplarily, first, the utility value of each robot performing search and rescue tasks in different sub-areas is calculated based on the capability parameters (such as mobility, endurance, load capacity, etc.) and current status (such as position, power, fault condition, etc.) of each robot. For example, there are multiple ground robots and multiple sub-areas, where the utility value can be calculated by weighted summation of various indicator values ​​(such as coverage area, target detection probability, power consumption, etc.) when the robot performs the task.

[0076] According to the utility value of each ground robot performing search and rescue tasks in different sub-areas, the scheme matching objective function is constructed by taking into account the task completion time and energy consumption indicators. Based on the scheme matching objective function, a multi-objective optimization model for the multi-robot task allocation problem is constructed, in which the decision variables are the allocation matrix of ground robots and sub-area search and rescue tasks, and the constraints include that at least one robot is assigned to each sub-area and each robot can only be assigned to one sub-area.

[0077] The optimization model constructed by the optimization algorithm is solved to obtain the optimal decision variable matrix. According to the value of the decision variable matrix, the optimal matching scheme of the ground robot and the sub-region search and rescue mission is given. According to the matching scheme, the ground robot is deployed to perform the search and rescue mission. The actual data during the execution of the mission is collected and fed back into the utility value calculation and model parameter adjustment for subsequent optimization.

[0078] In this embodiment, by constructing a multi-objective optimization model and using the allocation matrix of ground robots and sub-region search and rescue tasks as decision variables, the optimal matching solution of robots and tasks can be obtained. This solution can fully consider the capability parameters and current status of each robot, maximize the potential of the robot, and improve the efficiency and quality of search and rescue. By assigning only one sub-region constraint to each robot, it avoids multiple robots from working repeatedly in the same area, and improves the efficiency of cooperation between robots. At the same time, the constraint of assigning at least one robot to each sub-region ensures that all areas can be searched and rescued in a timely manner.

[0079] In an optional embodiment,

[0080] By solving the multi-objective optimization model of the multi-robot task allocation problem, the optimal task allocation solution is obtained, including:

[0081] The allocation matrix between the ground robot and the sub-region search and rescue task is represented as a real-valued matrix, and a certain number of particles are randomly generated, where each particle represents a possible task allocation scheme;

[0082] For each particle, calculate its fitness value on the solution matching objective function, and compare the dominance relationship between the current position of each particle and its individual optimal solution, and update the individual optimal solution according to the first comparison result; compare the dominance relationship between the individual optimal solutions of all particles and the global optimal solution, and update the global optimal solution according to the second comparison result;

[0083] For each particle, according to its current speed, the updated individual optimal solution and the updated global optimal solution, update its speed and position to obtain an updated particle;

[0084] The original particles and the updated particles are merged to obtain an extended population, and the extended population is non-dominated sorted to obtain non-dominated solution sets of different levels;

[0085] Repeat the iteration until the termination condition is met, then output the current non-dominated solution set as the optimal solution set, and select the final optimal task allocation plan from the obtained optimal solution set according to the decision maker's preference.

[0086] Exemplarily, the robot-task assignment matrix is ​​represented as a real-valued matrix, where the columns are the number of robots and the elements in the matrix are the number of tasks. The elements in the matrix represent the probability of assigning robot i to task j.

[0087] Multiple particles are randomly generated, each particle represents a possible task allocation scheme. For each particle, its fitness value is calculated on the time target and energy consumption target. The current position of each particle is compared with its individual optimal solution in dominance relationship, and the individual optimal solution is updated according to the first comparison result; the individual optimal solution of all particles is compared with the global optimal solution in dominance relationship, and the global optimal solution is updated according to the second comparison result.

[0088] Generate mutation vectors according to the multi-objective differential evolution algorithm. For each particle, three different particles in the population are randomly selected, and the mutation vector, i.e. the current speed, is generated by controlling the step size of differential mutation.

[0089] The particle speed and position are updated according to the multi-objective particle swarm optimization algorithm. For each particle, the speed and position are updated according to its current speed, individual optimal solution and global optimal solution through inertia weight, acceleration constant and random number. The newly generated particle position is subjected to boundary processing, the elements in the position matrix are restricted to a given range, and normalized to ensure that the sum of the elements in each row is 1. The original particles are merged with the newly generated particles to obtain an extended population, and the extended population is non-dominated sorted to obtain non-dominated solution sets of different levels.

[0090] Select N particles from the non-dominated solution set as the new generation population. Based on indicators such as crowding distance and aggregation distance, particles with higher levels and more uniform distribution can be selected first. Determine whether the termination conditions are met, such as reaching the maximum number of iterations, the quality of the solution reaching the preset threshold, etc. If met, output the non-dominated solution set as the optimal solution; otherwise, return to the particle fitness calculation step and continue iterative optimization; finally, according to the decision maker's preference, select the final optimal task allocation plan from the obtained optimal solution set.

[0091] In this embodiment, the parallel search mechanism of the particle swarm algorithm can effectively explore different areas of the solution space, improve the global optimization capability, and avoid falling into the local optimum. By directly performing non-dominated sorting in the target space, it can effectively handle multi-objective optimization problems, and simultaneously optimize multiple indicators such as task completion time and energy consumption to obtain a more comprehensive and reasonable solution set. The output is a non-dominated solution set, from which decision makers can flexibly select the optimal task allocation solution that meets the requirements according to their actual preferences.

[0092] S102. Send the sub-region search and rescue mission and the optimal matching solution to the corresponding ground robot, adopt different search and rescue strategies for different types of ground robots, and model multiple different types of ground robots as a discrete event system subsystem, and generate a heterogeneous robot collaborative search and rescue strategy by defining the formal language and lexicon of the collaborative control of each ground robot;

[0093] Among them, targeted search strategies are designed for different types of heterogeneous robots. For example, for robots equipped with visual sensors, a target search algorithm based on probability graphs is adopted. Specifically, based on visual images and deep learning technology, a probability occupancy grid map of the search and rescue area is constructed to characterize the probability of the existence of targets in each grid. Then, using Bayesian inference and information entropy theory, the optimal search path is derived so that the robot can reach the grid with the highest probability of the target in the shortest time. During the search process, the robot dynamically updates the probability map based on the target information detected in real time, and adjusts the search path accordingly. For another example, for robots equipped with infrared sensors, a heat source-guided fast search strategy is adopted. The robot uses infrared thermal imaging technology to detect the distribution of heat sources in the environment in real time and identify suspected target areas. Then, combined with the location, area, temperature and other characteristics of the target area, the optimal search route is planned, and the suspected target area is searched and carefully checked.

[0094] Among them, the discrete event system subsystem includes state variables, which represent the robot's motion state (such as moving, searching, avoiding obstacles, etc.), event variables, which represent the robot's task events (such as target discovery, area search completion, etc.) and communication events (such as receiving task instructions, sending status information, etc.).

[0095] On this basis, the formal language and lexicon of the collaborative control of heterogeneous robots are defined to standardize the information interaction and behavior coordination between multiple robots. For example, when the visual robot finds a suspicious target, it sends a "target discovery" event to the infrared robot, requesting it to go to the area for heat source detection; when the infrared robot confirms the existence of the target, it sends a "target confirmation" event to the central controller, requesting other robots to go to support. Finally, the multi-robot collaborative control strategy is formally analyzed and experimentally verified using simulation tools to ensure the safety and effectiveness of the collaborative search and rescue of heterogeneous robots.

[0096] In an optional embodiment,

[0097] Multiple different types of ground robots are modeled as a discrete event system subsystem. By defining the formal language and lexicon of the collaborative control of various ground robots, the heterogeneous robot collaborative search and rescue strategy is generated, including:

[0098] According to the task requirements and the capabilities of various ground robots, a series of language samples for communication interaction and behavioral coordination are generated and stored in the collaborative control language corpus;

[0099] Preprocess the collaborative control language corpus to form a standardized data set, and use word embedding technology to convert language samples into language sample embedding vectors;

[0100] A semantic neural network model based on an encoder-decoder architecture is constructed, wherein the semantic neural network model includes an encoder and a decoder. A task requirement encoder and a ground robot capability encoder based on a graph neural network are introduced into the encoder to encode the task requirement graph and the ground robot capability graph to obtain an embedded representation of each task node and ground robot node. At the same time, a pointer network-based copy mechanism is introduced into the decoder to calculate the attention distribution according to the current decoder and encoder outputs to obtain the probability of copying each position from the input sequence of the semantic neural network model.

[0101] The language sample embedding vector is input into the constructed semantic neural network model to generate a language symbol string for the collaborative control of heterogeneous robots. A semantic verification mechanism based on knowledge graph is introduced to check the semantic consistency and finally a formal description of the collaborative search and rescue strategy of heterogeneous robots is output.

[0102] Exemplarily, the collected collaborative control language corpus is first cleaned and annotated, and the language samples are paired with the corresponding task requirements and robot capabilities to form a standardized data set; using word embedding technologies such as Word2Vec, the words in the language samples are mapped to language sample embedding vectors as input features of the neural network; the task requirements and robot capabilities are feature extracted and encoded to form a structured input representation.

[0103] A sequence-to-sequence neural network model based on the Transformer architecture is designed, and the self-attention mechanism is used to achieve long-distance dependency modeling of input sequences and output sequences. In the encoder part of the Transformer, a task requirement and robot capability encoder based on a graph neural network is introduced to embed the structured information of tasks and robots into the language generation process. Among them, the graph attention network can encode the task requirement graph and the robot capability graph to obtain the embedded representation of each task node and robot node. By fusing these embedded representations with the output of the Transformer encoder, the structured information of tasks and robots can be introduced into the language generation process.

[0104] In the decoder part of Transformer, a copy mechanism based on the Pointer Network is introduced. The Pointer Network uses an attention distribution to decide which position of the element to copy from the input sequence. At a time step of the decoder, the Pointer Network calculates the attention distribution based on the current decoder and encoder outputs, where the attention distribution represents the probability of copying each position from the input sequence at a time step. Then, the decoder decides whether to generate a word from the vocabulary or copy the word from the input sequence based on the obtained attention distribution.

[0105] Using standardized data sets, the designed neural network model is trained through end-to-end supervised learning to optimize model parameters. During the training process, a reward mechanism based on reinforcement learning is introduced to optimize the model in a fine-grained manner based on indicators such as the quality and diversity of the generated language. Regularization techniques such as Early Stopping and Dropout are used to prevent overfitting of the model and improve the generalization ability of the model.

[0106] The trained neural network model is deployed to the online reasoning service to realize real-time language generation function, and the generated language symbol string is post-processed, including formatting, error checking, grammar correction, etc., to ensure that the generated language meets the requirements of collaborative control.

[0107] At the same time, a semantic verification mechanism based on knowledge graph is introduced to check the semantic consistency of the generated language and improve the accuracy and interpretability of language generation. Specifically, it is necessary to first build a knowledge graph related to the robot collaborative search and rescue mission. The knowledge graph is a structured knowledge base that contains knowledge such as entities (such as robot type, task type, etc.), relationships between entities, and attributes of entities. It can be extracted from an existing domain knowledge base or corpus, or semi-automatically / automatically constructed through knowledge graph automatic construction technology. The generated language description is semantically parsed to identify the entity mentions contained therein and link them to the corresponding entity nodes in the knowledge graph. Semantic parsing can be achieved using natural language processing technologies such as named entity recognition and relationship extraction. A series of semantic constraint rules are constructed based on domain knowledge and task requirements. These rules define the semantic constraints that should be satisfied between elements such as entities and relationships in the language description. For example, whether a certain type of robot can perform a certain task type, whether there is a sequential dependency relationship between different tasks, etc. The results of semantic parsing and entity linking are verified with the constraint rules. Check whether the entities and relationships in the language description meet the predefined semantic constraints, so as to determine whether the description is semantically consistent and reasonable. For the parts where semantic inconsistency is detected, correction suggestions can be given or a regeneration mechanism can be triggered to output a semantically consistent collaborative control strategy description.

[0108] In this embodiment, by constructing a semantic neural network model and encoding the task requirements and robot capabilities as graph embedding vector inputs, it is possible to comprehensively consider a variety of heterogeneous information, generate a language symbol string description of collaborative control, and realize efficient collaborative control of heterogeneous robots. The language description generation problem is modeled as an encoder-decoder framework, and a formal language description of the robot collaborative strategy is generated end-to-end without the need for artificial rules, reducing the complexity of strategy design. A semantic verification mechanism based on knowledge graph is introduced, which can check the semantic rationality and consistency of the generated strategy to ensure that the generated formal description meets actual needs. The encoder side uses graph neural networks to learn the structured representation of task requirements and robot capabilities, and the decoder side introduces a pointer network replication mechanism to capture the alignment relationship between input and output, which improves the expressive power of the model, can realize the automatic generation of heterogeneous robot collaborative control strategies, and ensure the rationality of strategy descriptions. It has both interpretability, generalization ability and adaptability, and is an efficient intelligent collaborative control method.

[0109] S103. Each ground robot performs search and rescue tasks in the corresponding sub-area according to the heterogeneous robot collaborative search and rescue strategy, and monitors the computing resource usage and network status of the drone node in real time. When it is detected that the resource usage or network status meets the preset conditions, the computing offloading decision is initiated, and according to the decision result, part of the search and rescue task is offloaded to other computing nodes to continue the search and rescue operation until the search and rescue task is completed.

[0110] Among them, computing resource usage mainly includes indicators such as CPU occupancy, memory occupancy, and storage space occupancy. Network conditions mainly include indicators such as network bandwidth, transmission delay, and packet loss rate.

[0111] The ground robot periodically sends resource detection requests to the drone node to obtain its computing resource usage. After receiving the request, the drone node evaluates its own computing resource usage and returns the evaluation results to the ground robot. Based on the collected evaluation results, the ground robot determines whether the computing resources of the drone node meet the preset conditions. The preset conditions can be set according to the needs of the search and rescue mission and the performance requirements of the system, such as CPU usage not exceeding 80%, memory usage not exceeding 70%, etc.

[0112] The ground robot regularly detects the network status between the ground robot and the drone node to evaluate network bandwidth, transmission delay, packet loss rate and other indicators. The detection method can use ICMP, TCP / UDP and other network diagnostic protocols. Based on the detection results, the ground robot determines whether the network status meets the preset conditions. The preset conditions can be set according to the real-time and reliability requirements of the search and rescue mission for data transmission, such as network bandwidth not less than 1Mbps, transmission delay not more than 50ms, packet loss rate not more than 1%, etc.

[0113] The goal of computation offloading decision is to select the optimal computation offloading scheme and offload part of the search and rescue mission from the UAV node to other computing nodes to improve the execution efficiency and quality of the search and rescue mission.

[0114] In an optional embodiment,

[0115] When it is detected that the resource usage or network status meets the preset conditions, the computing offloading decision is initiated, and part of the search and rescue task is offloaded to other computing nodes according to the decision result to continue the search and rescue operation until the search and rescue task is completed, including:

[0116] The computation offloading decision problem is modeled as an offloading optimization model, where the task offloading objective function is to minimize the task execution time and energy consumption, the constraints are the task completion time and energy consumption limits, and the decision variables are the offloading decision of each task and the selection of the offloading target node;

[0117] The optimization algorithm is used to calculate the offloading optimization model and obtain the optimal calculation offloading plan. According to the optimal calculation offloading plan, part of the search and rescue mission is offloaded from the UAV to the selected ground robot node.

[0118] The selected ground robot node receives the data of the offloading task, performs the offloading task calculation, and returns the offloading task calculation result to the UAV;

[0119] The drone integrates the received calculation results of the offloaded mission with the results of other missions executed locally to form a complete search and rescue plan, and continues to command the ground robot to execute it until the search and rescue mission is completed.

[0120] For example, when the ground robot detects that the computing resource usage or network status of the drone node meets the preset conditions, the computing offloading decision process is triggered. The goal of the computing offloading decision is to select the optimal computing offloading solution to offload part of the search and rescue mission from the drone node to other computing nodes to improve the execution efficiency and quality of the search and rescue mission.

[0121] The computation offloading decision problem can be modeled as an optimization problem. The optimization goal is to minimize the task execution time and energy consumption under the constraints of task completion time and energy consumption limit. The decision variables include the offloading decision of each task (binary variable, indicating whether to offload) and the selection of the offloading target node (integer variable, indicating which computing node to offload to).

[0122] Optionally, a heuristic algorithm or an approximate algorithm can be used to solve the problem. Common heuristic algorithms include genetic algorithm, particle swarm optimization algorithm, ant colony algorithm, etc. Taking genetic algorithm as an example, the unloading decision variable can be encoded as a chromosome, and the population can be continuously optimized through genetic operations such as selection, crossover, and mutation, and finally converge to the optimal solution.

[0123] According to the optimal computation offloading scheme obtained by optimization, the UAV node offloads part of the search and rescue mission to the selected ground robot node. The offloading process includes the transmission of mission data and the migration of mission computation.

[0124] During the mission data transmission phase, the drone node sends mission-related data (such as map information of the search and rescue area, target feature information, etc.) to the target node through the wireless network. In order to improve the efficiency and reliability of data transmission, technologies such as block transmission and multi-channel parallel transmission can be used. At the same time, in order to ensure the confidentiality and integrity of the data, the transmitted data can be encrypted and digitally signed.

[0125] In the task calculation migration phase, the target node starts to perform task calculation immediately after receiving the task data. The calculation process may involve complex algorithms and models, such as target detection, path planning, decision optimization, etc. In order to speed up the calculation process, the GPU, FPGA and other acceleration devices of the target node can be used. After the calculation is completed, the target node returns the calculation results to the drone node.

[0126] After receiving the calculation results of the offloaded task, the drone node integrates the results with the results of other tasks executed locally to form a complete search and rescue plan. According to the search and rescue plan, the drone node continues to command the ground robot to perform the search and rescue operation until the search and rescue mission is completed.

[0127] Optionally, during the entire computation offloading process, a sound fault-tolerant mechanism and exception handling strategy are established to prevent task execution failure due to node failure, network interruption, etc.

[0128] In an optional embodiment,

[0129] ;

[0130] Where J represents the offloading optimization target, M represents the number of search and rescue missions, i represents the search and rescue mission index, N represents the number of ground robot nodes, j represents the ground robot node index, and x represents the number of ground robot nodes. ij represents a binary variable, t ij represents the time required for task i to be executed on ground robot node j, t i represents the time required for task i to execute on the local node, λ represents the trade-off coefficient between time and energy consumption, and e ij represents the energy consumed by executing task i on ground robot node j, e i It represents the energy consumed by executing task i on the local node.

[0131] In this embodiment, by constructing an optimization model and solving the optimal offloading solution, computing resources can be reasonably allocated, and tasks can be offloaded to relatively idle nodes for execution, thereby improving the computing resource utilization efficiency of the entire system and avoiding resource waste. The load situation of the drone node calculation detected in real time can be used to dynamically decide whether task offloading is needed, thereby achieving dynamic load balancing of the system and improving operational stability. The objective function of the optimization model takes into account the task execution time and energy consumption. The optimal solution obtained can shorten the task completion delay, reduce energy consumption, and improve time and energy utilization efficiency. The technical solution based on computing offloading optimization can improve the computing resource utilization efficiency, improve latency and energy efficiency, ensure task execution continuity, support heterogeneous computing collaboration, enhance system reliability, and has good flexibility and scalability. It is an effective mobile edge computing solution.

[0132] In an optional embodiment,

[0133] The method further comprises:

[0134] Constructing a local task state model for the drone node and the backup node, wherein the input of the local task state model includes the attribute characteristics of the search and rescue task, the assigned search and rescue task list, and the completion progress of each search and rescue task, and the output is the completion probability and the estimated remaining time of the search and rescue task;

[0135] Build a federated learning framework between drone nodes and backup nodes. During the task execution, drone nodes and backup nodes use the task status data collected locally to obtain the model update through gradient calculation, and transmit the model update to the federated learning framework to obtain a global update solution.

[0136] The local task state model is updated using the global update scheme to obtain a new local task state model, and the iteration is repeated until the local task state model converges;

[0137] The drone node periodically sends heartbeat packets to the backup node, which monitors the status of the drone through the heartbeat packets. When the backup node fails to receive the heartbeat packets for multiple consecutive times, it determines that the drone is faulty and automatically triggers task migration.

[0138] During task migration, the backup node uses its own local task state model to continue executing the task. After the task migration is completed, the model parameters of the backup node's own local task state model are sent to the federated learning framework, and the received model parameters are distributed to other backup nodes through the federated learning framework.

[0139] Exemplarily, the drone node and the backup node each build their own local task state model to characterize the state and progress of task execution. The input of the model may include the attribute characteristics of the task, the list of assigned subtasks, the completion progress of each subtask, etc., and the output may be the completion probability of the task, the estimated remaining time, etc. The structure of the model may adopt a deep neural network, a graph neural network, etc., and the specific model architecture and hyperparameters may be optimized according to the characteristics of the task and the distribution of the data.

[0140] A federated learning framework is built between the drone nodes and the backup nodes to coordinate the model training and updating between different nodes. The framework mainly includes the following modules: (1) Communication module: responsible for transmitting model parameters, gradients, aggregation results and other data between nodes, using secure communication protocols such as SSL and TLS; (2) Aggregation module: responsible for aggregating the model updates received from other nodes, and can use aggregation algorithms such as FedAvg and FedProx; (3) Update module: responsible for updating the local model using the aggregated updates locally, and can use optimization algorithms such as gradient descent and Adam.

[0141] During the task execution process, the drone nodes and backup nodes use the locally collected task status data to train and update the local model. Specifically, each node first uses local data independently to calculate the gradient or weight update of the model. Each node sends the calculated update amount to the aggregation module of the federated learning framework, and the aggregation module aggregates the received update amount to obtain a global model update plan. After the aggregation module sends the global update plan to each node, each node uses the global update plan to update the local model to obtain a new local model. Repeat the above steps until the model converges or reaches the preset number of iterations.

[0142] When a drone node fails and needs to migrate tasks, the backup node can directly use its own local model to continue the task without obtaining complete task status data from the failed node. In order to ensure that the model of the backup node is consistent with the model of the failed node, after the task migration is completed, the backup node can send its own model parameters to the federated learning framework and distribute them to other nodes through the aggregation module to achieve global synchronization of the model.

[0143] In this embodiment, by constructing a local task state model, the task completion probability and estimated remaining time can be output according to the attributes, allocation and completion progress of the search and rescue task, providing decision support for task scheduling and management. Using the federated learning framework, the drone node and the backup node can update the parameters of the local task state model in real time during the task execution, so that the model can continue to learn and improve the accuracy of task state prediction. When a drone node fails, the backup node can automatically take over the task execution and continue to execute using its own local task state model, improving the fault tolerance and reliability of the system. The collaborative work of heterogeneous computing resources of drone nodes and backup nodes is realized, making full use of all computing power in the system and reducing network communication overhead; at the same time, each node only needs to update the local model, reducing computing overhead and achieving a balance between computing and communication overhead.

[0144] Figure 2Schematic diagram of the structure of a multi-robot collaborative control system based on edge computing according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0145] The first unit is used to obtain the environmental characteristics and search and rescue needs of the search and rescue area through the UAV, divide the search and rescue area into multiple sub-areas, generate sub-area search and rescue tasks, and simultaneously obtain the capability parameters and current status of each ground robot. According to the sub-area search and rescue tasks and the capability parameters and current status of each ground robot, an optimal matching scheme between the ground robot and the sub-area search and rescue tasks is generated;

[0146] The second unit sends the sub-area search and rescue tasks and the optimal matching solutions to the corresponding ground robots. Different search and rescue strategies are adopted for different types of ground robots. Multiple different types of ground robots are modeled as a discrete event system subsystem. By defining the formal language and lexicon of the collaborative control of various ground robots, a collaborative search and rescue strategy for heterogeneous robots is generated.

[0147] In the third unit, each ground robot performs search and rescue tasks in the corresponding sub-area according to the heterogeneous robot collaborative search and rescue strategy, and monitors the computing resource usage and network status of the drone node in real time. When it is detected that the resource usage or network status meets the preset conditions, the computing offloading decision is initiated, and part of the search and rescue task is offloaded to other computing nodes according to the decision result to continue the search and rescue operation until the search and rescue mission is completed.

[0148] According to a third aspect of the embodiments of the present invention,

[0149] An electronic device is provided, comprising:

[0150] processor;

[0151] a memory for storing processor-executable instructions;

[0152] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0153] A fourth aspect of the embodiments of the present invention is:

[0154] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0155] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-robot collaborative control method based on edge computing, characterized in that: include: The environmental characteristics and search and rescue needs of the search and rescue area are obtained through drones, the search and rescue area is divided into multiple sub-areas, and sub-area search and rescue tasks are generated. At the same time, the capability parameters and current status of each ground robot are obtained. According to the sub-area search and rescue tasks and the capability parameters and current status of each ground robot, the optimal matching scheme between the ground robot and the sub-area search and rescue tasks is generated; The sub-region search and rescue mission and the optimal matching solution are sent to the corresponding ground robot. Different search and rescue strategies are adopted for different types of ground robots. Multiple different types of ground robots are modeled as a discrete event system subsystem. By defining the formal language and lexicon of the collaborative control of various ground robots, a collaborative search and rescue strategy for heterogeneous robots is generated. Each ground robot performs search and rescue tasks in the corresponding sub-area according to the heterogeneous robot collaborative search and rescue strategy, and monitors the computing resource usage and network status of the drone node in real time. When it is detected that the resource usage or network status meets the preset conditions, the computing offloading decision is initiated, and part of the search and rescue tasks are offloaded to other computing nodes according to the decision results to continue the search and rescue operation until the search and rescue mission is completed; According to the task requirements and the capabilities of various ground robots, a series of language samples for communication interaction and behavioral coordination are generated and stored in the collaborative control language corpus; Preprocess the collaborative control language corpus to form a standardized data set, and use word embedding technology to convert language samples into language sample embedding vectors; A semantic neural network model based on an encoder-decoder architecture is constructed, wherein the semantic neural network model includes an encoder and a decoder. A task requirement encoder and a ground robot capability encoder based on a graph neural network are introduced into the encoder to encode the task requirement graph and the ground robot capability graph to obtain an embedded representation of each task node and ground robot node. At the same time, a pointer network-based copy mechanism is introduced into the decoder to calculate the attention distribution according to the current decoder and encoder outputs to obtain the probability of copying each position from the input sequence of the semantic neural network model. The language sample embedding vector is input into the constructed semantic neural network model to generate a language symbol string for the collaborative control of heterogeneous robots. A semantic verification mechanism based on knowledge graph is introduced to check the semantic consistency and finally a formal description of the collaborative search and rescue strategy of heterogeneous robots is output.

2. The method according to claim 1, characterized in that According to the sub-region search and rescue mission and the capability parameters and current status of each ground robot, the optimal matching scheme between the ground robot and the sub-region search and rescue mission is generated, including: According to the capability parameters and current status of each ground robot, the utility value of each ground robot in performing search and rescue tasks in different sub-areas is calculated; According to the utility value of each ground robot performing search and rescue tasks in different sub-areas, the scheme matching objective function is constructed by considering the task completion time and energy consumption indicators; A multi-objective optimization model for the multi-robot task allocation problem is constructed based on the solution matching objective function, where the decision variable is the allocation matrix of the ground robot and the sub-region search and rescue tasks, and the constraints include that at least one robot is allocated to each sub-region and each robot can only be allocated to one sub-region. By solving the multi-objective optimization model of the multi-robot task allocation problem, an optimal task allocation solution is obtained.

3. The method according to claim 2, characterized in that By solving the multi-objective optimization model of the multi-robot task allocation problem, the optimal task allocation solution is obtained, including: The allocation matrix between the ground robot and the sub-region search and rescue task is represented as a real-valued matrix, and a certain number of particles are randomly generated, where each particle represents a possible task allocation scheme; For each particle, calculate its fitness value on the solution matching objective function, and compare the dominance relationship between the current position of each particle and its individual optimal solution, and update the individual optimal solution according to the first comparison result; compare the dominance relationship between the individual optimal solutions of all particles and the global optimal solution, and update the global optimal solution according to the second comparison result; For each particle, according to its current speed, the updated individual optimal solution and the updated global optimal solution, update its speed and position to obtain an updated particle; The original particles and the updated particles are merged to obtain an extended population, and the extended population is non-dominated sorted to obtain non-dominated solution sets of different levels; Repeat the iteration until the termination condition is met, then output the current non-dominated solution set as the optimal solution set, and select the final optimal task allocation plan from the obtained optimal solution set according to the decision maker's preference.

4. The method according to claim 1, characterized in that: When it is detected that the resource usage or network status meets the preset conditions, the computing offloading decision is initiated, and part of the search and rescue task is offloaded to other computing nodes according to the decision result to continue the search and rescue operation until the search and rescue task is completed, including: The computation offloading decision problem is modeled as an offloading optimization model, where the task offloading objective function is to minimize the task execution time and energy consumption, the constraints are the task completion time and energy consumption limits, and the decision variables are the offloading decision of each task and the selection of the offloading target node; The optimization algorithm is used to calculate the offloading optimization model and obtain the optimal calculation offloading plan. According to the optimal calculation offloading plan, part of the search and rescue mission is offloaded from the UAV to the selected ground robot node. The selected ground robot node receives the data of the offloading task, performs the offloading task calculation, and returns the offloading task calculation result to the UAV; The drone integrates the received calculation results of the offloaded mission with the results of other missions executed locally to form a complete search and rescue plan, and continues to command the ground robot to execute it until the search and rescue mission is completed.

5. The method according to claim 4, characterized in that The formula of the task offloading objective function is as follows: ； Where J represents the offloading optimization target, M represents the number of search and rescue missions, i represents the search and rescue mission index, N represents the number of ground robot nodes, j represents the ground robot node index, and x represents the number of ground robot nodes. ij represents a binary variable, t ij represents the time required for task i to be executed on ground robot node j, t i represents the time required for task i to execute on the local node, λ represents the trade-off coefficient between time and energy consumption, and e ij represents the energy consumed by executing task i on ground robot node j, e i It represents the energy consumed by executing task i on the local node.

6. The method according to claim 1, characterized in that The method further comprises: Constructing a local task state model for the drone node and the backup node, wherein the input of the local task state model includes the attribute characteristics of the search and rescue task, the assigned search and rescue task list, and the completion progress of each search and rescue task, and the output is the completion probability and the estimated remaining time of the search and rescue task; Build a federated learning framework between drone nodes and backup nodes. During the task execution, drone nodes and backup nodes use the task status data collected locally to obtain the model update through gradient calculation, and transmit the model update to the federated learning framework to obtain a global update solution. The local task state model is updated using the global update scheme to obtain a new local task state model, and the iteration is repeated until the local task state model converges; The drone node periodically sends heartbeat packets to the backup node, which monitors the status of the drone through the heartbeat packets. When the backup node fails to receive the heartbeat packets for multiple consecutive times, it determines that the drone is faulty and automatically triggers task migration. During task migration, the backup node uses its own local task state model to continue executing the task. After the task migration is completed, the model parameters of the backup node's own local task state model are sent to the federated learning framework, and the received model parameters are distributed to other backup nodes through the federated learning framework.

7. A multi-robot collaborative control system based on edge computing, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to obtain the environmental characteristics and search and rescue needs of the search and rescue area through the UAV, divide the search and rescue area into multiple sub-areas, generate sub-area search and rescue tasks, and simultaneously obtain the capability parameters and current status of each ground robot. According to the sub-area search and rescue tasks and the capability parameters and current status of each ground robot, an optimal matching scheme between the ground robot and the sub-area search and rescue tasks is generated; The second unit sends the sub-area search and rescue tasks and the optimal matching solutions to the corresponding ground robots. Different search and rescue strategies are adopted for different types of ground robots. Multiple different types of ground robots are modeled as a discrete event system subsystem. By defining the formal language and lexicon of the collaborative control of various ground robots, a collaborative search and rescue strategy for heterogeneous robots is generated. In the third unit, each ground robot performs search and rescue tasks in the corresponding sub-area according to the heterogeneous robot collaborative search and rescue strategy, and monitors the computing resource usage and network status of the drone node in real time. When it is detected that the resource usage or network status meets the preset conditions, the computing offloading decision is initiated, and part of the search and rescue task is offloaded to other computing nodes according to the decision result to continue the search and rescue operation until the search and rescue mission is completed.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images