Distributed collaborative debugging method and system for industrial robots

Abstract

The present invention provides an industrial robot distributed collaborative debugging method and system, which relates to the field of distributed control technology, and comprises establishing a directed acyclic graph model of an industrial robot collaborative debugging task, analyzing the directed acyclic graph model by using a critical path method, and determining a critical path of the collaborative debugging task; performing reliability modeling on each industrial robot node in the collaborative debugging task, and calculating the failure probability of each collaborative debugging subtask under a current scheduling scheme in combination with the time redundancy characteristics of each collaborative debugging subtask; defining a state space of a task scheduling scheme according to the obtained directed acyclic graph model of the industrial robot collaborative debugging task, the critical path of the collaborative debugging task, and the failure probability of each collaborative debugging subtask under the current scheduling scheme, searching the state space by a dynamic programming algorithm, and obtaining a scheduling scheme with the minimum global failure probability as the optimal fault-tolerant scheduling strategy at the current moment.

Description

Technical Field

[0001] The present invention relates to distributed control technology, and in particular to a distributed collaborative debugging method and system for industrial robots. Background Art

[0002] With the development of industrial automation, industrial robots play an increasingly important role in modern manufacturing. Industrial robots can perform various complex tasks, such as assembly, welding, spraying, handling, etc., greatly improving production efficiency and product quality. However, the application of industrial robots also faces many challenges, especially in the scenario of multi-robot collaborative work, how to achieve efficient, reliable and flexible scheduling and debugging is a key issue.

[0003] Traditional industrial robot scheduling and debugging methods mainly adopt a centralized approach, with the central controller uniformly performing task allocation and execution control. This centralized method has some limitations: first, centralized control relies on the central node, which has the risk of single point failure; second, centralized scheduling is difficult to cope with large-scale, dynamically changing task requirements; in addition, the centralized solution has high communication overhead, slow response speed, and difficulty in achieving real-time control. In response to the above problems, some studies have proposed an agent-based distributed scheduling method. This method models the scheduling problem as a multi-agent system, where each robot node acts as an agent and generates a scheduling strategy through local perception and decision-making. Agents exchange information through communication protocols and negotiate using methods such as game theory and auction theory to reach a globally consistent scheduling decision. Although the agent-based distributed scheduling method has certain advantages, it still has some shortcomings: the communication and negotiation overhead between agents is high, and it is difficult to ensure the real-time nature of scheduling decisions; in addition, distributed decisions may lead to scheduling conflicts, affecting the reliability of the system.

[0004] Therefore, there is an urgent need for an efficient, reliable and flexible distributed collaborative debugging method and system for industrial robots to solve the current technical challenges and improve the performance and adaptability of multi-robot collaborative operations. This patent proposes an innovative solution to the above technical problems. Summary of the invention

[0005] The embodiments of the present invention provide a distributed collaborative debugging method and system for industrial robots, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provided is an industrial robot distributed collaborative debugging method, comprising:

[0008] Establish a directed acyclic graph model of the collaborative debugging task of the industrial robot, wherein the nodes represent the collaborative debugging subtasks, and the edges represent the dependencies and communication costs between the collaborative debugging subtasks. Use the critical path method to analyze the directed acyclic graph model to obtain the time redundancy characteristics of each collaborative debugging subtask, and determine the critical path of the collaborative debugging task based on the time redundancy characteristics of each collaborative debugging subtask.

[0009] The reliability model is built for each industrial robot node in the collaborative debugging task, and the failure probability of each collaborative debugging subtask under the current scheduling scheme is calculated by combining the time redundancy characteristics of each collaborative debugging subtask.

[0010] According to the obtained directed acyclic graph model of the industrial robot collaborative debugging task, the critical path of the collaborative debugging task and the failure probability of each collaborative debugging subtask under the current scheduling scheme, the state space of the task scheduling scheme is defined, and the state space is searched through the dynamic programming algorithm to obtain the scheduling scheme with the minimum global failure probability as the optimal fault-tolerant scheduling strategy at the current moment.

[0011] In an optional embodiment,

[0012] The directed acyclic graph model is analyzed by using the critical path method to obtain the time redundancy characteristics of each collaborative debugging subtask. The critical path of the collaborative debugging task is determined based on the time redundancy characteristics of each collaborative debugging subtask, including:

[0013] According to the scale and complexity of the collaborative debugging task of the industrial robot, a graph partitioning algorithm based on community discovery is adopted to divide the directed acyclic graph model into multiple subgraphs, wherein each subgraph represents a local view of the collaborative debugging task;

[0014] A local critical path analysis agent is deployed on each subgraph. The local critical path analysis agent predicts the execution time and communication delay between local collaborative debugging tasks based on the topological structure of the subgraph to which it belongs, using a pre-built time delay prediction model, and adopts a multi-granularity time redundancy analysis method to analyze the time redundancy characteristics of local collaborative debugging tasks. By calculating the time redundancy characteristics of each local collaborative debugging task, the local critical path information is identified.

[0015] Based on the obtained local critical path information, a local critical path view including local critical task identification, time redundancy attributes and dependencies with other subgraphs is generated, and the local critical path views are aggregated through a distributed graph traversal algorithm. A global dependency graph is constructed based on the aggregated local critical path views, and the final global critical path is identified by topological sorting and critical path analysis of the global dependency graph.

[0016] In an optional embodiment,

[0017] The local critical path views are aggregated by a distributed graph traversal algorithm, a global dependency graph is constructed based on the aggregated local critical path views, and the final global critical path is identified by topological sorting and critical path analysis of the global dependency graph, including:

[0018] According to the dependency relationship between each subgraph in the local critical path view, a local dependency graph is constructed, and based on the local dependency graph, a local convergence tree is generated using a minimum spanning tree algorithm;

[0019] According to the local convergence tree, the local critical path views are converged step by step through a point-to-point communication method, and the global critical path views obtained by convergence are used as subgraphs of the local dependency graph. The local dependency graphs are merged at overlapping boundary nodes to generate fragments of the global dependency graph.

[0020] Distribute the shards of the global dependency graph to a group of global topological sorting agents. Each global topological sorting agent runs a distributed topological sorting algorithm based on the assigned global dependency graph shards and obtains a globally consistent topological sorting result by exchanging the topological sorting information of the boundary nodes.

[0021] The global topological sequence is obtained based on the globally consistent topological sorting results. Combined with the time attributes in the global dependency graph, the global critical path is calculated through the dynamic programming algorithm to output the final global critical path.

[0022] In an optional embodiment,

[0023] The formula for calculating the failure probability of each collaborative debugging subtask under the current scheduling scheme is as follows:

[0024] ;

[0025] ;

[0026] Among them, F i represents the failure probability of the ith collaborative debugging subtask under the current scheduling scheme, i represents the ith collaborative debugging subtask, t represents the current time, and d i represents the latest completion time of the ith collaborative debugging subtask, v ij represents the dependent node j of the i-th collaborative debugging subtask, V i represents a node set, q ij (t) represents the impact of other node failures on dependent node v at time t ij The reliability of the failure probability, p ij (t) represents the dependent node v ij The failure probability of itself at time t, v ikrepresents the kth node that subtask i depends on during its execution. Represents node v ik The cumulative impact of failures on the reliability of the i-th collaborative debugging subtask, S ij Represents and depends on node v ij Other related industrial robot nodes.

[0027] In an optional embodiment,

[0028] The state space is searched by a dynamic programming algorithm to obtain a scheduling solution with the lowest global failure probability. The optimal fault-tolerant scheduling strategy at the current moment includes:

[0029] Construct a multi-agent reinforcement learning model and design an independent agent for each industrial robot node, including state space, action space, reward function and policy network. The state space represents the local information of the current industrial robot node, and the action space represents the optional operations of the industrial robot node in scheduling decision-making.

[0030] Each agent updates its own policy network parameters based on local state information and feedback rewards, generates scheduling decisions in real time, and broadcasts the generated scheduling decisions to other agents;

[0031] Each agent compares the received scheduling strategies of other agents with the local scheduling strategies, and determines whether there is a conflict based on the pre-defined decision conflict rules. If there is a conflict, it conducts multiple rounds of iterative negotiations with other agents through a distributed negotiation algorithm to generate an initial fault-tolerant scheduling strategy.

[0032] A distributed consistency algorithm is used to perform consistency check on the generated initial fault-tolerant scheduling strategy. The fault-tolerant scheduling strategy that passes the consistency check is synchronized to other intelligent agents and updated locally to obtain the final optimal fault-tolerant scheduling strategy.

[0033] In an optional embodiment,

[0034] The calculation formula for each agent to update its own strategy network parameters based on local state information and feedback rewards is as follows:

[0035] ;

[0036] in, represents the updated policy network parameters, θ i represents the policy network parameters before updating, α represents the learning rate, represents the gradient of the policy network parameters before updating, E k represents the expected value at time step k, r k (θ) represents the probability ratio, clip(·) represents the clipping operation,ε A is a hyperparameter representing the magnitude of the policy update. k represents the advantage function.

[0037] In an optional embodiment,

[0038] The method further comprises:

[0039] Obtain the status of each industrial robot node and the corresponding task execution progress information, detect node failure events through anomaly detection algorithms, and trigger the fault-tolerant scheduling mechanism when a node failure event is detected;

[0040] Based on the fault-tolerant scheduling mechanism, the impact of node failure events on each task is evaluated and the incremental failure probability of the task is calculated;

[0041] Based on the failure probability increment and the preset failure probability threshold, the affected failure tasks are determined and reconstructed. The reconstruction optimization target is constructed by combining the task importance, node reliability and load balancing. The optimal reconstruction scheme for the target task is determined by minimizing the reconstruction optimization target.

[0042] According to the target task reconstruction scheme, the failed task is migrated from the current robot node to the target node in the target task reconstruction scheme. At the same time, the incremental checkpoint technology is used to periodically save the task status on the current robot node, and the saved task status is transmitted to a preset storage node. Based on the task status in the storage node, the execution status of the failed task is restored on the target node.

[0043] According to a second aspect of the embodiments of the present invention,

[0044] Provided is an industrial robot distributed collaborative debugging system, comprising:

[0045] The first unit is used to establish a directed acyclic graph model of the collaborative debugging task of the industrial robot, wherein the nodes represent the collaborative debugging subtasks, the edges represent the dependencies and communication costs between the collaborative debugging subtasks, the directed acyclic graph model is analyzed by using the critical path method, the time redundancy characteristics of the collaborative debugging subtasks are obtained, and the critical path of the collaborative debugging task is determined based on the time redundancy characteristics of the collaborative debugging subtasks;

[0046] The second unit is used to perform reliability modeling on each industrial robot node in the collaborative debugging task, and calculate the failure probability of each collaborative debugging subtask under the current scheduling scheme in combination with the time redundancy characteristics of each collaborative debugging subtask;

[0047] The third unit is used to define the state space of the task scheduling scheme according to the obtained directed acyclic graph model of the industrial robot collaborative debugging task, the critical path of the collaborative debugging task, and the failure probability of each collaborative debugging subtask under the current scheduling scheme, and search the state space through a dynamic programming algorithm to obtain the scheduling scheme with the minimum global failure probability as the optimal fault-tolerant scheduling strategy at the current moment.

[0048] According to a third aspect of the embodiments of the present invention,

[0049] An electronic device is provided, comprising:

[0050] processor;

[0051] a memory for storing processor-executable instructions;

[0052] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0053] A fourth aspect of the embodiments of the present invention is:

[0054] 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.

[0055] In this embodiment, by establishing a directed acyclic graph model, the dependencies and communication costs between the various collaborative debugging subtasks can be intuitively represented. Using the critical path method, the time redundancy characteristics of each subtask can be analyzed, thereby determining the critical path of the entire collaborative debugging task, laying the foundation for subsequent scheduling optimization. By performing reliability modeling on each industrial robot node, the failure probability of each collaborative debugging subtask under the current scheduling scheme is calculated. This provides an important reliability indicator for subsequent fault-tolerant scheduling optimization. By optimizing the scheduling strategy, the failure probability of the entire collaborative debugging task is reduced, thereby improving the reliability and robustness of the task. At the same time, scheduling optimization based on time redundancy characteristics can also ensure the timely completion of the task, avoiding the failure of some key subtasks to cause the entire task to be completed on time. It can significantly improve the reliability, robustness and timeliness of the collaborative debugging task of industrial robots, and provide strong theoretical and technical support for achieving efficient and high-quality collaborative debugging of industrial robots. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of a distributed collaborative debugging method for industrial robots according to an embodiment of the present invention;

[0057] Figure 2 It is a schematic diagram of the structure of the distributed collaborative debugging system of industrial robots according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] 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.

[0059] 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.

[0060] Figure 1 FIG. 1 is a flow chart of a distributed collaborative debugging method for an industrial robot according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0061] S101. Establish a directed acyclic graph model of the collaborative debugging task of the industrial robot, wherein the nodes represent the collaborative debugging subtasks, and the edges represent the dependencies and communication costs between the collaborative debugging subtasks. The directed acyclic graph model is analyzed using the critical path method to obtain the time redundancy characteristics of the collaborative debugging subtasks, and the critical path of the collaborative debugging task is determined based on the time redundancy characteristics of the collaborative debugging subtasks.

[0062] The time redundancy characteristics of each collaborative debugging subtask include the earliest start time, the latest start time and the time margin of each subtask.

[0063] In an optional embodiment,

[0064] The directed acyclic graph model is analyzed by using the critical path method to obtain the time redundancy characteristics of each collaborative debugging subtask. The critical path of the collaborative debugging task is determined based on the time redundancy characteristics of each collaborative debugging subtask, including:

[0065] According to the scale and complexity of the collaborative debugging task of the industrial robot, a graph partitioning algorithm based on community discovery is adopted to divide the directed acyclic graph model into multiple subgraphs, wherein each subgraph represents a local view of the collaborative debugging task;

[0066] A local critical path analysis agent is deployed on each subgraph. The local critical path analysis agent predicts the execution time and communication delay between local collaborative debugging tasks based on the topological structure of the subgraph to which it belongs, using a pre-built time delay prediction model, and adopts a multi-granularity time redundancy analysis method to analyze the time redundancy characteristics of local collaborative debugging tasks. By calculating the time redundancy characteristics of each local collaborative debugging task, the local critical path information is identified.

[0067] Based on the obtained local critical path information, a local critical path view including local critical task identification, time redundancy attributes and dependencies with other subgraphs is generated, and the local critical path views are aggregated through a distributed graph traversal algorithm. A global dependency graph is constructed based on the aggregated local critical path views, and the final global critical path is identified by topological sorting and critical path analysis of the global dependency graph.

[0068] For example, first, according to the scale and complexity of the collaborative debugging task of industrial robots, the directed acyclic graph model of the entire task is distributedly partitioned. A graph partitioning algorithm based on community detection, such as the Louvain algorithm and the spectral clustering algorithm, can be used to divide the directed acyclic graph into multiple subgraphs. Each subgraph represents a local view of a debugging task, preserving the dependencies and communication costs between local tasks.

[0069] A local critical path analysis agent is deployed on each subgraph. The critical path analysis agent is responsible for analyzing the time redundancy characteristics of local tasks within the subgraph to which it belongs and identifying local critical paths. Specifically, the critical path analysis agent first predicts the execution time and communication delay between local tasks based on the topological structure of the subgraph using the time delay prediction model. Then, a multi-granularity time redundancy analysis method is used to analyze the time redundancy characteristics of local tasks at two granularities: the task level and the operation level. By calculating the earliest start time, the latest start time, and the free float time of each local task, where the free float time is the difference between the latest start time and the earliest start time of each task, the critical tasks and critical paths that have a significant impact on the local debugging progress are identified. The critical path is the path composed of all tasks with a free float time of 0, and the critical tasks are tasks with a free float time of 0, that is, tasks on the critical path.

[0070] Among them, the time delay prediction model can be built based on the long short-term memory neural network (LSTM) through historical data training and online learning to capture the dynamic change of time delay. During the task execution process, the edge weights in the directed acyclic graph model are dynamically updated to reflect the current time delay prediction value. By adjusting the model parameters in real time, the adaptability and robustness of the critical path analysis can be improved.

[0071] Each critical path analysis agent generates a local critical path view based on the local critical path analysis results. The local critical path view contains the identification of local critical tasks, time redundancy attributes, and dependencies with other subgraphs. By defining the communication protocol between subgraphs, the critical path analysis agents exchange local critical path view information through a message passing mechanism. Distributed graph traversal algorithms, such as distributed breadth-first search and distributed depth-first search, are used to achieve efficient propagation and convergence of local critical path views. Specifically, each critical path analysis agent uses the local critical path view generated by itself as the root node. The agent sends its own local view to other agents it depends on. The agent that receives the local view will merge it with its own local view. The merged view will continue to propagate to the next layer of dependent agents. This process continues until all agents obtain a complete global view.

[0072] Based on the aggregation of local critical path views, a distributed critical path fusion strategy is designed to globally integrate local critical path information. A fusion method based on the global dependency graph is used to construct a global critical path view of the entire collaborative debugging task. Specifically, a global dependency graph is constructed based on the dependencies between subgraphs contained in the critical path analysis agent. The nodes of the global dependency graph represent local critical tasks, and the edges represent the global dependencies between local critical tasks. By performing topological sorting and critical path analysis on the global dependency graph, the global critical paths that have a significant impact on the entire collaborative debugging task are identified.

[0073] In this embodiment, a large-scale directed acyclic graph model is divided into multiple subgraphs through a graph partitioning algorithm, which reduces the scale of problems that a single node needs to handle, thereby reducing the complexity of the solution and improving the analysis and calculation efficiency. Deploying a local critical path analysis agent on each subgraph can realize parallel analysis of local views, make full use of distributed computing resources, and improve analysis efficiency. Using a pre-built time delay prediction model, the execution time and communication delay between collaborative debugging tasks can be predicted more accurately, providing more reliable input data for time redundancy analysis, thereby improving the accuracy of the analysis results. Using a multi-granularity time redundancy analysis method, collaborative debugging tasks can be analyzed from different time granularities, critical path information can be fully identified, and the comprehensiveness and depth of the analysis can be improved. The local critical path views are efficiently converged through a distributed graph traversal algorithm, and a global dependency graph is constructed based on the convergence results, which realizes efficient analysis and optimization of large-scale complex collaborative debugging tasks, improves the accuracy and efficiency of critical path identification, and provides strong support for improving the reliability and real-time performance of collaborative debugging tasks.

[0074] In an optional embodiment,

[0075] The local critical path views are aggregated by a distributed graph traversal algorithm, a global dependency graph is constructed based on the aggregated local critical path views, and the final global critical path is identified by topological sorting and critical path analysis of the global dependency graph, including:

[0076] According to the dependency relationship between each subgraph in the local critical path view, a local dependency graph is constructed, and based on the local dependency graph, a local convergence tree is generated using a minimum spanning tree algorithm;

[0077] According to the local convergence tree, the local critical path views are converged step by step through a point-to-point communication method, and the global critical path views obtained by convergence are used as subgraphs of the local dependency graph. The local dependency graphs are merged at overlapping boundary nodes to generate fragments of the global dependency graph.

[0078] Distribute the shards of the global dependency graph to a group of global topological sorting agents. Each global topological sorting agent runs a distributed topological sorting algorithm based on the assigned global dependency graph shards and obtains a globally consistent topological sorting result by exchanging the topological sorting information of the boundary nodes.

[0079] The global topological sequence is obtained based on the globally consistent topological sorting results. Combined with the time attributes in the global dependency graph, the global critical path is calculated through the dynamic programming algorithm to output the final global critical path.

[0080] Exemplarily, in the process of aggregating local critical path views, a local aggregation agent is first deployed on each subgraph. Each local aggregation agent establishes a communication connection with the local critical path analysis agent of the subgraph to which it belongs through a subscription-publishing mechanism. The local critical path analysis agent publishes the generated local critical path view to the local aggregation agent.

[0081] After receiving the local critical path view, the local convergence agent extracts the dependencies between the subgraphs and constructs a local dependency graph. A local convergence tree is generated by applying the minimum spanning tree algorithm on the local dependency graph. The local convergence tree defines the convergence order and path of the local critical path view.

[0082] According to the local convergence tree, local convergence agents converge local critical path views step by step along the edges of the convergence tree through point-to-point communication. During the convergence process, local convergence agents reduce the received local critical path views according to predefined reduction actions and merge redundant critical path metadata and edge information.

[0083] The global critical path view obtained by aggregation uses an incremental graph merging algorithm during the construction of the global dependency graph. Each local aggregation agent uses the local critical path view obtained by aggregation as a subgraph of the local dependency graph, and generates each fragment of the global dependency graph by merging the local dependency graphs at the overlapping boundary nodes.

[0084] Distribute the shards of the global dependency graph to a set of global topological sorting agents. Each global topological sorting agent runs a distributed topological sorting algorithm based on the assigned global dependency graph shard to determine the global topological order of each operation. The global topological sorting agents achieve globally consistent topological sorting results by exchanging topological sorting information of boundary nodes.

[0085] The global critical path analysis agent obtains the global topological sequence from the global topological sorting agent, combines the time attributes in the global dependency graph, and calculates the global critical path through a dynamic programming algorithm. In this process, by introducing a critical path pruning strategy, local paths that are unlikely to become global critical paths are pruned to improve computational efficiency.

[0086] Finally, the global critical path analysis agent outputs the global critical path results and feeds them back to the local convergence agent. The local convergence agent verifies and adjusts the local critical path according to the global critical path, resolves potential path conflicts, and ensures the consistency and reliability of the global critical path in the local view.

[0087] In this embodiment, by constructing a local dependency graph and using the minimum spanning tree algorithm to generate a local convergence tree, an efficient local view convergence path can be determined, and point-to-point communication is used to perform step-by-step convergence, avoiding the communication bottleneck problem caused by centralized convergence and improving convergence efficiency. The global critical path view obtained by convergence is used as a subgraph of the local dependency graph. By merging the local dependency graph at the overlapping boundary, a fragment of the global dependency graph can be generated, supporting the distributed construction of large-scale global dependency graphs, reducing single-point computing and storage pressure. Based on the globally consistent topological sorting results, the global topological sequence is obtained. Combined with the time attributes in the global dependency graph, the dynamic programming algorithm can be used to accurately calculate the global critical path, and the final global critical path can be output, avoiding errors and conflicts that may occur during the convergence of local critical path views. A distributed computing architecture is adopted, which has good scalability and flexibility. The number of proxy nodes participating in the calculation can be dynamically adjusted according to actual needs, and efficient resource utilization can be achieved through dynamic load balancing and other means.

[0088] S102. Perform reliability modeling on each industrial robot node in the collaborative debugging task, and calculate the failure probability of each collaborative debugging subtask under the current scheduling scheme in combination with the time redundancy characteristics of each collaborative debugging subtask;

[0089] Exemplarily, first, the state space of the reliability model is defined according to the operating state and failure mode of the industrial robot node. Each state in the state space represents a possible operating state of the node, such as normal working state, different types of failure states, and maintenance state. The transition between states represents the change of the node state, and the transition rate reflects the frequency and probability of state change.

[0090] Secondly, the historical operation data and fault logs of the industrial robot nodes are used to estimate the transfer rate parameters of the reliability model. The transfer probability and average transfer time between states are calculated by counting the frequency and time interval of node transfers between different states. The transfer rate matrix of the reliability model is obtained by using methods such as maximum likelihood estimation or Bayesian estimation to characterize the dynamic characteristics of node state transfer.

[0091] Then, based on the reliability model, the reliability index of the industrial robot node is calculated. The instantaneous probability distribution solution method of the Markov process is used to calculate the probability of the node being in different states within a given time range. Combining the task execution time and state duration of the node, the reliability function of the node during the task execution process is calculated, that is, the probability of the node running without failure within a specified time.

[0092] At the same time, the time redundancy characteristics of the task are introduced into the reliability modeling process. For each collaborative debugging subtask, the industrial robot node sequence on its critical path is extracted to construct the reliability dependency graph of the task. The reliability dependency graph is a directed acyclic graph, where the nodes represent the industrial robot nodes that the subtask depends on during execution, the edges represent the reliability dependency relationship between the nodes, and the edge weights represent the degree of influence of node failure on the reliability of the subtask.

[0093] Combined with the time redundancy information such as the earliest start time, the latest start time, and the free floating time of the subtask, the minimum reliability requirement threshold of each subtask is calculated. The minimum reliability requirement threshold reflects the maximum failure probability that the subtask can tolerate. Exceeding this threshold will cause the subtask to fail to be completed on time. By comparing the minimum reliability requirement threshold of the subtask with the failure probability of its dependent nodes, the reliability bottleneck is identified, providing a decision-making basis for subsequent scheduling optimization.

[0094] Based on the reliability dependency graph of the subtask and the node failure probability distribution, a graph-based reliability propagation algorithm is designed to calculate the failure probability of the subtask under the current scheduling scheme. The reliability propagation algorithm traverses the reliability dependency graph in a depth-first search or breadth-first search manner, and recursively calculates the failure probability of the subtask based on the dependency relationship and failure probability between nodes.

[0095] In an optional embodiment,

[0096] The formula for calculating the failure probability of each collaborative debugging subtask under the current scheduling scheme is as follows:

[0097] ;

[0098] ;

[0099] Among them, F i represents the failure probability of the ith collaborative debugging subtask under the current scheduling scheme, i represents the ith collaborative debugging subtask, t represents the current time, and d i represents the latest completion time of the ith collaborative debugging subtask, v ij represents the dependent node j of the i-th collaborative debugging subtask, V i represents a node set, q ij (t) represents the impact of other node failures on dependent node v at time t ij The failure probability affected by the reliability, p ij (t) represents the dependent node v ij The failure probability of itself at time t, v ik represents the kth node that subtask i depends on during its execution. Represents node v ik The cumulative impact of failures on the reliability of the i-th collaborative debugging subtask, S ij Represents and depends on node v ij Other related industrial robot nodes.

[0100] In this embodiment, by constructing a dynamic reliability model based on Markov process, the dynamic change characteristics of the operating state of the industrial robot node can be characterized, providing a theoretical basis for reliability analysis. Using Markov process theory, the reliability function of the industrial robot node running without fault within a specified time can be calculated, providing key input for subsequent reliability analysis and scheduling optimization. A graph-based reliability propagation algorithm is designed, which can efficiently calculate the failure probability of subtasks under the current scheduling scheme, providing important reliability evaluation indicators for scheduling optimization. By integrating time redundancy features, dynamic reliability modeling and graph-based propagation analysis, the technical solution can more accurately evaluate the reliability level of collaborative debugging tasks.

[0101] S103. Define the state space of the task scheduling scheme based on the obtained directed acyclic graph model of the industrial robot collaborative debugging task, the critical path of the collaborative debugging task, and the failure probability of each collaborative debugging subtask under the current scheduling scheme, search the state space through a dynamic programming algorithm, and obtain the scheduling scheme with the minimum global failure probability as the optimal fault-tolerant scheduling strategy at the current moment.

[0102] Among them, the failure probability refers to the probability that the collaborative debugging subtask cannot be completed on time under the current scheduling scheme.

[0103] Exemplarily, first, the state space of the task scheduling scheme is defined. Each state in the state space represents a possible task scheduling scheme, including the execution order of subtasks, the assigned industrial robot nodes, and the attributes such as the start time and completion time of the task. The transition between states represents the change of the scheduling scheme, such as changing the execution order of subtasks, reallocating industrial robot nodes, etc. Secondly, the state representation and transfer function of the state space are designed. Using a vectorized state representation method, each scheduling scheme state is represented as a multidimensional vector, and each dimension of the vector corresponds to a different scheduling attribute. The transfer function defines the legal transfer rules between states to ensure that the generated new state satisfies the task dependency and resource occupancy constraints. Then, the objective function of the state space is defined, that is, the minimization of the global failure probability. According to the failure probability of each subtask under the current scheduling scheme, the global failure probability of the entire collaborative debugging task is calculated. The global failure probability can be the weighted sum of the failure probabilities of all subtasks, and the weight reflects the importance of different subtasks to the success of the overall task.

[0104] The dynamic programming algorithm is applied to search the state space to find the scheduling solution with the lowest global failure probability. The dynamic programming algorithm decomposes the original problem into overlapping sub-problems and solves the sub-problems recursively to finally obtain the optimal solution to the original problem.

[0105] Specifically, an initial scheduling scheme is selected as the starting state, and its global failure probability is calculated. For the current state, all possible successor states are generated, and the global failure probability of each successor state is calculated. A memoization search technique is used to avoid repeated calculation of states that have been visited. For each successor state, it is compared with the currently known optimal scheduling scheme. If the global failure probability of the successor state is greater than the current optimal solution, the state is pruned and its subtree is no longer searched; if the global failure probability of the successor state is less than the current optimal solution, the optimal solution is updated, and the successor state is added to the candidate state set for the next round of search. Finally, the recursive, pruning, and updating stages are repeated until all possible states are searched or the preset search depth limit is reached. The scheduling scheme with the smallest global failure probability found during the search process is returned as the optimal fault-tolerant scheduling strategy at the current moment.

[0106] In an optional embodiment,

[0107] The state space is searched by a dynamic programming algorithm to obtain a scheduling solution with the lowest global failure probability. The optimal fault-tolerant scheduling strategy at the current moment includes:

[0108] Construct a multi-agent reinforcement learning model and design an independent agent for each industrial robot node, including state space, action space, reward function and policy network. The state space represents the local information of the current industrial robot node, and the action space represents the optional operations of the industrial robot node in scheduling decision-making.

[0109] Each agent updates its own policy network parameters based on local state information and feedback rewards, generates scheduling decisions in real time, and broadcasts the generated scheduling decisions to other agents;

[0110] Each agent compares the received scheduling strategies of other agents with the local scheduling strategies, and determines whether there is a conflict based on the pre-defined decision conflict rules. If there is a conflict, it conducts multiple rounds of iterative negotiations with other agents through a distributed negotiation algorithm to generate an initial fault-tolerant scheduling strategy.

[0111] A distributed consistency algorithm is used to perform consistency check on the generated initial fault-tolerant scheduling strategy. The fault-tolerant scheduling strategy that passes the consistency check is synchronized to other intelligent agents and updated locally to obtain the final optimal fault-tolerant scheduling strategy.

[0112] Exemplarily, first, a multi-agent reinforcement learning model is constructed. An independent agent is designed for each robot node, including a state space, an action space, a reward function, and a policy network. The state space represents the local information of the current robot node, such as task execution progress, resource occupancy, etc. The action space represents the optional operations of the robot node in scheduling decisions, such as task allocation, execution order adjustment, etc. The reward function evaluates the decision-making performance of the agent based on indicators such as task completion quality and time efficiency. The policy network is based on a deep neural network, which generates an action probability distribution based on the state input to guide the decision-making behavior of the agent. Secondly, the communication and interaction mechanism between agents is defined, allowing agents to share state information and decision results. By designing a suitable communication protocol and data exchange format, the agent can broadcast its own state information regularly and receive status updates from other agents. At the same time, agents can exchange decision information, such as task allocation schemes, scheduling order, etc., to achieve coordination and optimization of distributed decision-making. In order to promote cooperation between agents, a global reward function is designed to evaluate the collaborative performance of the entire multi-agent system. The global reward function comprehensively considers the contributions of all agents, such as total task completion time, average resource utilization, etc., and encourages agents to take into account global optimization goals when making local decisions.

[0113] In terms of learning algorithms, a distributed reinforcement learning algorithm is used to achieve parallel learning and strategy optimization of agents. Each agent independently updates its own policy network parameters based on local state transfer and reward feedback. At the same time, collaborative learning mechanisms such as shared experience replay or joint policy optimization are introduced to promote knowledge sharing and policy coordination among agents. Agents can regularly upload collected state transfer samples to the central experience replay storage for other agents to learn and update strategies. In addition, policy network parameters can be synchronized regularly between agents to achieve joint policy optimization and accelerate the convergence process.

[0114] In order to ensure the consistency and feasibility of distributed scheduling decisions, the information transmission and decision synchronization mechanism between agents is designed. After generating the scheduling strategy, the agent needs to broadcast the decision results to other agents and wait for the confirmation and feedback of other agents. After receiving the decision information of other agents, the agent detects whether there is a decision conflict, such as task allocation conflict, resource occupation conflict, etc. A rule-based or constraint-based conflict detection algorithm is used to identify the inconsistency between decisions. When a decision conflict is detected, the negotiation mechanism between agents is triggered. Agents negotiate through communication protocols such as request-response mode or multiple rounds of iterations. The negotiation process adopts a distributed negotiation algorithm, such as the distributed constrained optimization (DCOP) algorithm or a bargaining-based negotiation algorithm. During the negotiation process, the agent generates candidate conflict resolution solutions based on factors such as task priority and node reliability. The candidate solutions are evaluated, and the optimal solution is selected considering indicators such as global constraint satisfaction and scheduling performance.

[0115] The agent exchanges the generated conflict resolution plan with other agents and receives feedback. Through multiple rounds of iteration and feedback, the agent continuously optimizes and amends the plan until a consensus is reached. A distributed consensus algorithm, such as Paxos or Raft, is used to reach a decision consensus among agents. Through multiple rounds of voting and confirmation, the agents gradually converge to a consistent scheduling decision as the final optimal fault-tolerant scheduling strategy. The distributed consensus algorithm ensures the consistency and feasibility of the decision. When the agents reach a consensus on the decision, the optimal fault-tolerant scheduling strategy is synchronized to all agents. Each agent updates its local scheduling decision to ensure the consistency of the decision.

[0116] In an optional embodiment,

[0117] The calculation formula for each agent to update its own strategy network parameters based on local state information and feedback rewards is as follows:

[0118] ;

[0119] in, represents the updated policy network parameters, θ irepresents the policy network parameters before updating, α represents the learning rate, represents the gradient of the policy network parameters before updating, E k represents the expected value at time step k, r k (θ) represents the probability ratio, clip(·) represents the clipping operation, ε A is a hyperparameter representing the magnitude of the policy update. k represents the advantage function.

[0120] In this embodiment, by modeling each industrial robot node as an independent agent, distributed intelligent scheduling decisions can be realized with the help of reinforcement learning methods, making full use of the local state information of each node, and improving the timeliness and pertinence of the decision. By adopting a distributed multi-agent solution, a large amount of state transmission between nodes and central nodes can be avoided, communication overhead can be reduced, and the real-time and high availability of the system can be improved. The introduction of a distributed negotiation algorithm can coordinate scheduling decisions between agents, eliminate conflicts, enhance the global coordination of decisions, and improve the effectiveness and consistency of the overall scheduling solution. The use of a distributed consistency algorithm to check and synchronize the initial scheduling strategy can ensure that all agents ultimately reach a consensus on the optimal fault-tolerant scheduling strategy to avoid the uncertainty of distributed decision-making. The intelligent and adaptive scheduling of the collaborative debugging tasks of industrial robots is realized. While improving the quality of decision-making, it also has the advantages of high real-time performance, high scalability, and high fault tolerance, which can effectively improve the robustness and efficiency of collaborative debugging tasks.

[0121] In an optional embodiment,

[0122] The method further comprises:

[0123] Obtain the status of each industrial robot node and the corresponding task execution progress information, detect node failure events through anomaly detection algorithms, and trigger the fault-tolerant scheduling mechanism when a node failure event is detected;

[0124] Based on the fault-tolerant scheduling mechanism, the impact of node failure events on each task is evaluated and the incremental failure probability of the task is calculated;

[0125] Based on the failure probability increment and the preset failure probability threshold, the affected failure tasks are determined and reconstructed. The reconstruction optimization target is constructed by combining the task importance, node reliability and load balancing. The optimal reconstruction scheme for the target task is determined by minimizing the reconstruction optimization target.

[0126] According to the target task reconstruction scheme, the failed task is migrated from the current robot node to the target node in the target task reconstruction scheme. At the same time, the incremental checkpoint technology is used to periodically save the task status on the current robot node, and the saved task status is transmitted to a preset storage node. Based on the task status in the storage node, the execution status of the failed task is restored on the target node.

[0127] For example, a monitoring agent is first deployed on each robot node to regularly collect information such as the node's hardware status, software status, and task execution progress. The monitoring agent sends the collected data to the central monitoring server to achieve real-time monitoring of the node status and task progress.

[0128] The central monitoring server receives data from the monitoring agent and detects node failure events in real time through anomaly detection algorithms such as threshold detection, statistical analysis, or machine learning algorithms. When failures such as node response timeout, hardware failure, or software crash are detected, the fault-tolerant scheduling mechanism is triggered.

[0129] Based on the fault-tolerant scheduling mechanism, the fault-tolerant scheduler receives node failure events and evaluates the impact of failed nodes on each task. By analyzing the dependencies and execution status of tasks, the incremental probability of task failure caused by node failure is calculated. Taking into account the criticality and time constraints of tasks, the affected key tasks are determined.

[0130] Specifically, when a failure of an industrial robot node is detected, all subtasks that directly or indirectly depend on the failed node are identified in the task dependency graph. These subtasks are the subtasks affected by the failure of the node. For each affected subtask, the failure probability of the subtask in the current state is recalculated based on the failure mode of the failed node and the task execution progress. The newly calculated failure probability is usually higher than the initial failure probability. The updated failure probability is compared with the initial failure probability, and the difference between the two is the increment of the failure probability of the subtask due to the node failure. Since there is a dependency relationship between tasks, an increase in the failure probability of a subtask will cause an increase in the failure probability of the subsequent subtasks that depend on it. Therefore, it is necessary to traverse the task dependency graph, propagate the failure probability increment of each subtask along the dependency edge to the subsequent nodes, and cumulatively calculate the final failure probability increment of each subtask. The failure probability increments of all subtasks are summarized to obtain the failure probability increment of the entire collaborative debugging task, which is used as a key indicator for evaluating the impact of node failure.

[0131] According to the pre-set failure probability threshold, such as 0.5, when the failure probability of a task exceeds 50%, it is considered a failed task. All subtasks are traversed, and their failure probability increments are added to the initial failure probability. For subtasks whose sum of failure probabilities exceeds the threshold, they are marked as failed tasks and added to the failed task set.

[0132] Different weight coefficients are assigned according to the criticality of the task, execution priority, etc. The historical reliability data of the node (such as the mean trouble-free operation time) is used as a reference for selecting the migration target. The current load of the node is used as a parameter to avoid migrating tasks to overloaded nodes. The weighted sum of the above three indicators can be used as the reconstruction optimization objective function, and the goal is to minimize this function.

[0133] The problem is modeled as an integer programming problem, where the decision variable is the task-to-node assignment matrix, where the constraints include that each task can only be assigned to one node and the load of each node does not exceed the upper limit. An exploratory algorithm (such as simulated annealing) or an optimization solver (such as Gurobi) can be used to solve the integer programming problem. The optimal solution obtained is the optimal reconstruction solution, which gives the target migration node for each failed task.

[0134] According to the optimal reconstruction plan, each failed task is migrated to the target node. The incremental checkpoint technology is used to save the task status snapshot at the current node at a preset time interval (such as 10 seconds). The task status snapshot is transferred to the preset storage node for backup, and the latest task status snapshot in the storage node is loaded at the target node. Based on the loaded status, the execution of the task is resumed at the target node to achieve seamless switching of task execution.

[0135] In this embodiment, the status of the industrial robot node and the task execution progress information are monitored in real time through the anomaly detection algorithm, so that the node failure event can be discovered in time and the triggering preparation for fault-tolerant scheduling can be made. The fault-tolerant scheduler can evaluate the impact of the node failure event on each task, calculate the increment of the task failure probability, and provide a basis for formulating an appropriate fault-tolerant scheduling strategy. The incremental checkpoint technology is used to regularly save the task status and transmit it to the storage node, which provides support for the execution status recovery of the task on the target node and improves the continuity and consistency of fault-tolerant recovery. It can efficiently respond to the failure event of the industrial robot node, identify the real failed tasks, and optimize the reconstruction and deployment of these tasks to the appropriate nodes, while ensuring the consistent migration of the task execution status, thereby minimizing the impact of the failure event on the entire collaborative debugging task and improving the robustness and reliability of task execution.

[0136] Figure 2 FIG. 1 is a schematic diagram of the structure of the distributed collaborative debugging system for industrial robots according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0137] The first unit is used to establish a directed acyclic graph model of the collaborative debugging task of the industrial robot, wherein the nodes represent the collaborative debugging subtasks, the edges represent the dependencies and communication costs between the collaborative debugging subtasks, the directed acyclic graph model is analyzed by using the critical path method, the time redundancy characteristics of the collaborative debugging subtasks are obtained, and the critical path of the collaborative debugging task is determined based on the time redundancy characteristics of the collaborative debugging subtasks;

[0138] The second unit is used to perform reliability modeling on each industrial robot node in the collaborative debugging task, and calculate the failure probability of each collaborative debugging subtask under the current scheduling scheme in combination with the time redundancy characteristics of each collaborative debugging subtask;

[0139] The third unit is used to define the state space of the task scheduling scheme according to the obtained directed acyclic graph model of the industrial robot collaborative debugging task, the critical path of the collaborative debugging task, and the failure probability of each collaborative debugging subtask under the current scheduling scheme, and search the state space through a dynamic programming algorithm to obtain the scheduling scheme with the minimum global failure probability as the optimal fault-tolerant scheduling strategy at the current moment.

[0140] According to a third aspect of the embodiments of the present invention,

[0141] An electronic device is provided, comprising:

[0142] processor;

[0143] a memory for storing processor-executable instructions;

[0144] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0145] A fourth aspect of the embodiments of the present invention is:

[0146] 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.

[0147] 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.

[0148] 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 distributed collaborative debugging method for industrial robots, characterized in that: include: Establish a directed acyclic graph model of the collaborative debugging task of the industrial robot, wherein the nodes represent the collaborative debugging subtasks, and the edges represent the dependencies and communication costs between the collaborative debugging subtasks. Analyze the directed acyclic graph model using the critical path method to obtain the time redundancy characteristics of each collaborative debugging subtask, and determine the critical path of the collaborative debugging task based on the time redundancy characteristics of each collaborative debugging subtask, wherein the time redundancy characteristics include the earliest start time, the latest start time and the time margin of each subtask; The reliability model is built for each industrial robot node in the collaborative debugging task, and the failure probability of each collaborative debugging subtask under the current scheduling scheme is calculated by combining the time redundancy characteristics of each collaborative debugging subtask. According to the obtained directed acyclic graph model of the collaborative debugging task of the industrial robot, the critical path of the collaborative debugging task, and the failure probability of each collaborative debugging subtask under the current scheduling scheme, the state space of the task scheduling scheme is defined, and the state space is searched by a dynamic programming algorithm to obtain the scheduling scheme with the minimum global failure probability as the optimal fault-tolerant scheduling strategy at the current moment; The formula for calculating the failure probability of each collaborative debugging subtask under the current scheduling scheme is as follows: ； ； Among them, F i represents the failure probability of the ith collaborative debugging subtask under the current scheduling scheme, i represents the ith collaborative debugging subtask, t represents the current time, and d i represents the latest completion time of the ith collaborative debugging subtask, v ij represents the dependent node j of the i-th collaborative debugging subtask, V i represents a node set, q ij (t) represents the impact of other node failures on dependent node v at time t ij The reliability of the failure probability, p ij (t) represents the dependent node v ij The failure probability of itself at time t, v ik represents the kth node that subtask i depends on during its execution. Represents node v ik The cumulative impact of failures on the reliability of the i-th collaborative debugging subtask, S ij Represents and depends on node v ij Other related industrial robot nodes.

2. The method according to claim 1, characterized in that The directed acyclic graph model is analyzed by using the critical path method to obtain the time redundancy characteristics of each collaborative debugging subtask. The critical path of the collaborative debugging task is determined based on the time redundancy characteristics of each collaborative debugging subtask, including: According to the scale and complexity of the collaborative debugging task of the industrial robot, a graph partitioning algorithm based on community discovery is adopted to divide the directed acyclic graph model into multiple subgraphs, wherein each subgraph represents a local view of the collaborative debugging task; A local critical path analysis agent is deployed on each subgraph. The local critical path analysis agent predicts the execution time and communication delay between local collaborative debugging tasks based on the topological structure of the subgraph to which it belongs, using a pre-built time delay prediction model, and adopts a multi-granularity time redundancy analysis method to analyze the time redundancy characteristics of local collaborative debugging tasks. By calculating the time redundancy characteristics of each local collaborative debugging task, the local critical path information is identified. Based on the obtained local critical path information, a local critical path view is generated, which includes local critical task identification, time redundancy attributes and dependencies with other subgraphs. The local critical path views are aggregated through a distributed graph traversal algorithm, and a global dependency graph is constructed based on the aggregated local critical path views. The final global critical path is identified by topological sorting and critical path analysis of the global dependency graph.

3. The method according to claim 2, characterized in that The local critical path views are aggregated through a distributed graph traversal algorithm, a global dependency graph is constructed based on the aggregated local critical path views, and the final global critical path is identified by topological sorting and critical path analysis of the global dependency graph, including: According to the dependency relationship between each subgraph in the local critical path view, a local dependency graph is constructed, and based on the local dependency graph, a local convergence tree is generated using a minimum spanning tree algorithm; According to the local convergence tree, the local critical path views are converged step by step through a point-to-point communication method, and the global critical path views obtained by convergence are used as subgraphs of the local dependency graph. The local dependency graphs are merged at overlapping boundary nodes to generate fragments of the global dependency graph. Distribute the shards of the global dependency graph to a group of global topological sorting agents. Each global topological sorting agent runs a distributed topological sorting algorithm based on the assigned global dependency graph shards and obtains a globally consistent topological sorting result by exchanging the topological sorting information of the boundary nodes. The global topological sequence is obtained based on the globally consistent topological sorting results. Combined with the time attributes in the global dependency graph, the global critical path is calculated through the dynamic programming algorithm to output the final global critical path.

4. The method according to claim 1, characterized in that The state space is searched by a dynamic programming algorithm to obtain a scheduling solution with the lowest global failure probability. The optimal fault-tolerant scheduling strategy at the current moment includes: Construct a multi-agent reinforcement learning model and design an independent agent for each industrial robot node, including state space, action space, reward function and policy network. The state space represents the local information of the current industrial robot node, and the action space represents the optional operations of the industrial robot node in scheduling decision-making. Each agent updates its own policy network parameters based on local state information and feedback rewards, generates scheduling decisions in real time, and broadcasts the generated scheduling decisions to other agents; Each agent compares the received scheduling strategies of other agents with the local scheduling strategies, and determines whether there is a conflict based on the pre-defined decision conflict rules. If there is a conflict, it conducts multiple rounds of iterative negotiations with other agents through a distributed negotiation algorithm to generate an initial fault-tolerant scheduling strategy. A distributed consistency algorithm is used to perform consistency check on the generated initial fault-tolerant scheduling strategy. The fault-tolerant scheduling strategy that passes the consistency check is synchronized to other intelligent agents and updated locally to obtain the final optimal fault-tolerant scheduling strategy.

5. The method according to claim 4, characterized in that The calculation formula for each agent to update its own strategy network parameters based on local state information and feedback rewards is as follows: ； in, represents the updated policy network parameters, θ i represents the policy network parameters before updating, α represents the learning rate, represents the gradient of the policy network parameters before updating, E k represents the expected value at time step k, r k (θ) represents the probability ratio, clip(·) represents the clipping operation, ε A is a hyperparameter representing the magnitude of the policy update. k represents the advantage function.

6. The method according to claim 1, characterized in that The method further comprises: Obtain the status of each industrial robot node and the corresponding task execution progress information, detect node failure events through anomaly detection algorithms, and trigger the fault-tolerant scheduling mechanism when a node failure event is detected; Based on the fault-tolerant scheduling mechanism, the impact of node failure events on each task is evaluated and the incremental failure probability of the task is calculated; Based on the failure probability increment and the preset failure probability threshold, the affected failure tasks are determined and reconstructed. The reconstruction optimization target is constructed by combining the task importance, node reliability and load balancing. The optimal reconstruction scheme for the target task is determined by minimizing the reconstruction optimization target. According to the target task reconstruction scheme, the failed task is migrated from the current robot node to the target node in the target task reconstruction scheme. At the same time, the incremental checkpoint technology is used to periodically save the task status on the current robot node, and the saved task status is transmitted to a preset storage node. Based on the task status in the storage node, the execution status of the failed task is restored on the target node.

7. An industrial robot distributed collaborative debugging system, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to establish a directed acyclic graph model of the collaborative debugging task of the industrial robot, wherein the nodes represent the collaborative debugging subtasks, the edges represent the dependencies and communication costs between the collaborative debugging subtasks, the directed acyclic graph model is analyzed by using the critical path method, the time redundancy characteristics of the collaborative debugging subtasks are obtained, and the critical path of the collaborative debugging task is determined based on the time redundancy characteristics of the collaborative debugging subtasks; The second unit is used to perform reliability modeling on each industrial robot node in the collaborative debugging task, and calculate the failure probability of each collaborative debugging subtask under the current scheduling scheme in combination with the time redundancy characteristics of each collaborative debugging subtask; The third unit is used to define the state space of the task scheduling scheme according to the obtained directed acyclic graph model of the industrial robot collaborative debugging task, the critical path of the collaborative debugging task, and the failure probability of each collaborative debugging subtask under the current scheduling scheme, and search the state space through a dynamic programming algorithm to obtain the scheduling scheme with the minimum global failure probability as the optimal fault-tolerant scheduling strategy at the current moment.

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