71 results about "Task mapping" patented technology

The invention provides a robot control method based on Model-Based and RL, and relates to the technical field of automatic control. According to the method, a high-level task atlas model and a predefined task action set are constructed, a target task is mapped into a state node, graph search is carried out in combination with a current world state, an optimal task action path is planned, and a first action instruction is output. And calling a baseline strategy in the simulation environment to generate a control instruction, executing feasibility verification, compensating the deviation between the simulation and the real environment through the error prediction model, correcting the real control instruction, and then driving the robot to execute. And dynamically updating the success probability weight of state transition in the atlas based on the execution result, and feeding back the executed state for subsequent decision. According to the method, the reliability, the adaptability and the execution safety of migrating a simulation strategy to a real robot are improved, and closed-loop optimization and continuous self-adaption of a complex task are realized.

The invention provides a single instruction stream multi-fiber programming model and a construction method thereof. The method comprises the following steps: constructing a task, thread and fiber three-level concurrent unit; different code logic units are configured for each task, and one task is mapped to a plurality of threads; mapping the threads and processor cores one by one, and mapping one thread to a plurality of threads; the threads within the thread are caused to share all registers on the processor core. According to the single-instruction-stream multi-fiber-thread programming model and the construction method thereof, the programming model is constructed based on the fiber threads, concurrent execution of asynchronous data streams is achieved, the resource utilization rate is effectively increased, and redundant calculation is eliminated.

A self-optimizing and self-programming computing system (SOSPCS) design framework that achieves both programmability and flexibility and exploits computing heterogeneity [e.g., CPUs, GPUs, and hardware accelerators (HWAs)] is provided. First, at compile time, a task pool consisting of hybrid tasks with different processing element (PE) affinities according to target applications is formed. Tasks preferred to be executed on GPUs or accelerators are detected from target applications by neural networks. Tasks suitable to run on CPUs are formed by community detection to minimize data movement overhead. Next, a distributed reinforcement learning-based approach is used at runtime to allow agents to map the tasks onto the network-on-chip-based heterogeneous PEs by learning an optimal policy based on Q values in the environment.

The invention relates to the technical field of resource scheduling, in particular to an AI-assisted multi-factory collaborative production scheduling system which comprises a path filing module, a task sorting module, a strategy switching module, a node evaluation module and a task mapping module. According to the invention, the device sharing node is identified through the transverse connection of the device and the path node, the resource cross interference degree is effectively mastered, the task priority is accurately adjusted, the coordination of resource allocation and task execution is optimized, and the flexible adjustment capability in the task execution process is improved; node resource state switching and fluctuation changes are dynamically monitored, abnormal node load and processing time features are accurately extracted, a path replacement strategy is finely judged in combination with equipment shutdown time sequence analysis and resource residual feature classification, and the flexibility and adaptability in multi-factory collaborative production scheduling are enhanced. Equipment faults and state fluctuation are pre-judged, and a path strategy is adjusted, so that the risk of task delay caused by production rhythm fluctuation and resource interference is effectively reduced.

The invention relates to the technical field of software, in particular to an automatic construction method based on dependency topology analysis and dynamic orchestration, which comprises the following steps of: reading and analyzing a predefined configuration file to construct a service dependency relationship graph; wherein nodes in the service dependency relationship graph represent services, and directed edges represent dependency relationships among the services. According to the method, the predefined configuration file is read and analyzed to construct the service dependency relationship graph, so that automation of dependency management is realized, and manual identification errors are avoided; through dynamic mapping of services and construction tasks, the problem of conversion from a logic architecture to physical execution is solved; based on collaborative optimization of topological sorting and a priority strategy, the construction efficiency and the resource utilization rate are improved, and time consumed by a critical path is shortened. Therefore, full-link automation from dependency identification and task mapping to intelligent arrangement is completed, and the efficiency bottleneck and error risk caused by manual intervention are effectively solved.

The invention provides a secondary equipment debugging intelligent auxiliary method and system based on a secondary logic model, and relates to the technical field of debugging assistance, and the method comprises the steps: collecting multi-source structured data related to secondary equipment; constructing a secondary logic model diagram for debugging assistance by taking the equipment nodes as objects and the communication relationship and the logic control relationship as connecting edges based on the multi-source structured data; obtaining a debugging task; mapping the debugging task to a corresponding node in the secondary logic model diagram through a semantic and configuration fusion matching mechanism; based on the secondary logic model diagram and the debugging task mapping result, generating a structured debugging operation template in a mode of combining template retrieval, state injection and logic verification; based on the structured debugging operation template, comparing and checking the running state data of the current debugging object equipment; and on the basis of the comparison check result and the structured debugging operation template, executing online monitoring of the link state, and assisting in realizing intelligent analysis of the debugging state.

The invention discloses a high-end memory chip asynchronous test system based on a distributed architecture, which relates to the technical field of chip testing, and comprises a state modeling module, a master control end initialization and test end access, memory chip test task pool construction, test end state analysis in combination with historical test load and historical fault information, and memory chip asynchronous test. Generating an optimized scheduling strategy; the scheduling module is used for performing task division and asynchronous scheduling on the memory chip test task pool by adopting an intelligent scheduling algorithm based on an optimized scheduling strategy to generate an asynchronous data flow queue; according to the issuing module, the test end receives the asynchronous data flow queue in an asynchronous mode to carry out protocol unpacking and task mapping, and a memory chip function test task is obtained and issued; according to the method, the test throughput rate and the resource utilization rate are improved by asynchronously sequencing and scheduling tasks, the task scheduling delay is reduced, the load of each test end is balanced, and meanwhile, the horizontal expansibility and the fault-tolerant capability are enhanced.

This invention belongs to the technical field of embedded artificial intelligence and real-time system scheduling, and more specifically, relates to an adaptive load resource scheduling method and apparatus for embedded GPUs. The method involves offline acquisition and modeling of the platform's thermal throttling behavior, constructing DNN variants with different computational loads and accuracies, and profiling them under multiple frequency and TPC configurations to obtain a latency-power mapping table. In the online phase, based on task deadline margins and current temperature / frequency states, the method looks up the variant with the highest accuracy and meeting the deadline under the current thermal conditions and selects the corresponding TPC configuration from the profiling table. If necessary, phased online pruning and low-overhead switching are implemented, and the method collaborates with system-level DVFS / task mapping to optimize energy efficiency. This method can still guarantee the real-time performance of inference tasks under passive frequency reduction conditions and significantly reduce performance fluctuations caused by thermal throttling.

The invention provides a partition management method. According to the method, in the task mapping process, firstly, a partition to which each data item in a partition belongs is determined, the partition has a plurality of data, and the partition and an induction task have a corresponding relation; counting the data items belonging to the same partition to obtain a data item counting result; and then a data entry statistical result is sent to the management node, so that the management node makes a decision on whether to adjust the induction task or not. According to the method disclosed by the invention, the management node can make a decision on whether to adjust the number of induction tasks or not before the mapping task performs partition copying on the data items in the fragments to obtain intermediate data. Meanwhile, due to the fact that the decision is made in advance, a certain mapping task can be immediately read by the induction task after generating the intermediate data, and the induction task does not need to start to read the data items after all other mapping tasks generate the intermediate data. Therefore, the data reading efficiency is improved.

The application discloses a heterogeneous platform approximate computing task optimization mapping method based on DVFS and DPM. Firstly, real-time tasks with correlation are modeled as an approximate computing task model, so that a task directed acyclic graph (DAG), a task correlation matrix and a six-tuple representing task characteristics can be obtained; then, a mechanism of DVFS and DPM combination is introduced based on a heterogeneous multi-core platform; a problem description of task mapping based on QoS and energy joint optimization is constructed; a variable substitution method and a Big-M reconstruction method are used to process nonlinear terms in the problem, the task mapping problem is linearized, and an optimal solution is obtained through a Gurobi solver; a task layering method and a greedy algorithm are used to design a low-complexity heuristic algorithm, and the scalability of the mapping method is improved. The method of the application adopts the mechanism of DVFS+DPM joint optimization under the premise of meeting the system real-time, energy efficiency and reliability constraints, and improves the QoS of the system.

The invention discloses an AI interaction terminal control method and system based on a cloud platform, and belongs to the technical field of artificial intelligence. State parameter sets of a plurality of AI interaction terminals are collected and uploaded to the cloud platform; constructing a modeling engine including behavior pattern recognition and state coupling analysis based on the state parameters, and generating a multi-terminal coupling sensing model; terminal priorities are calculated according to the model, an asymmetric task mapping matrix is constructed, and a control strategy is generated; issuing a heterogeneous control instruction to the terminal and collecting execution feedback; self-learning updating of the model and the strategy is carried out in combination with the original state parameters and the feedback data; if the terminal fault is detected, task migration is carried out based on the coupling weight and the redundancy, and the control matrix is updated; and finally, cross-terminal linkage and control strategy self-evolution are realized through S-F-C closed-loop iteration. According to the invention, multi-terminal cooperative sensing, adaptive control and task robust migration are realized, and the intelligence and stability of an AI control system are improved.

This invention discloses an industrial agent reasoning framework and method based on physical first-principle constraints, belonging to the field of industrial intelligence technology. The framework includes a data access layer, a task parsing layer, a candidate solution generation layer, a physical constraint verification layer, a conflict rollback and replanning layer, an execution decision layer, and a knowledge accumulation layer. The physical constraint verification layer uses at least one physical first-principle model as a pre-constraint hard constraint gating before the candidate decision sequence enters the execution decision layer. It performs parameter mapping, boundary condition solving, consistency calculation, residual determination, and executability determination on the physical computation variables obtained from the structured reasoning task mapping, and outputs structured conflict information when the verification fails. The conflict rollback and replanning layer performs targeted rollback and replanning based on the violation type, violation node level, and source of missing boundary conditions in the structured conflict information. The execution decision layer only outputs execution-type output results for candidate decision sequences that have obtained the pass flag. The knowledge accumulation layer writes the structured conflict information and the corrected pass results into a rule base, template base, or historical case base to serve as filtering conditions, initial parameter values, tool sorting criteria, or threshold setting criteria during subsequent candidate generation. This invention can improve the physical feasibility, reliability, interpretability, and auditability of industrial agent decision-making results.

The invention relates to the technical field of computing power scheduling, and discloses a computing power intermediate station resource layer full-link cooperative scheduling method oriented to multiple service scenes. According to the method, a resource request carrying a service scene label is received, the requirements for GPU computing power, a memory and storage IO are analyzed, and a standardized resource descriptor is generated. Meanwhile, states of all physical nodes are dynamically collected, a global resource view is constructed, resource descriptors are matched with the view, priorities are distributed in combination with service scene labels, and a resource reservation mechanism is triggered. And mapping the task to a target node according to a matching result, and executing GPU computing power division, memory allocation and network bandwidth reservation. Strategies are dynamically adjusted in the scheduling process, and efficient utilization of resources is ensured.

The application provides a method and system for single-step debugging in a data processing flow, and belongs to the field of data processing. The method comprises the following steps: receiving a single-step debugging start instruction, mapping the whole flow to a flow instance by analyzing the content of the data processing flow, and mapping the tasks being executed or having been executed in the flow to task instances; executing the current task instance, updating the task instance state and attribute assignment after the execution is completed; interrupting the flow instance after the execution of the task instance is completed; when the returned task instance information is received, the execution state and input and output data of the task instance are visually displayed according to the task instance state; when the next step running instruction is received, the flow instance identifier is input, the next task to be executed of the flow instance is queried, and the task instance is encapsulated and run, and the above steps are repeated to complete the whole flow debugging. The application can realize flow interruption and recovery, real-time visualization of intermediate data, greatly improve the debugging efficiency, and help to quickly locate errors.

The application discloses a quantum computing task mapping method and a quantum computer operating system, idle quantum bits in a quantum chip are dynamically partitioned in real time according to actual requirements of a quantum computing task to be executed, a quantum bit topology structure obtained can be matched with the quantum computing task to be executed, no waiting time is generated in the matching, and the matching degree is high, so that the resource utilization rate of the quantum chip is greatly improved, and the execution timeliness of the quantum computing task in a program waiting queue is effectively improved; by using the mapping method, each quantum computing task in the program waiting queue can quickly find an optimal partition area on idle quantum bits of a required quantum chip in a quantum chip cluster of the system to perform mapping.

The invention relates to the technical field of multi-robot cooperative control, in particular to a multi-robot cooperative control method and system based on a multi-modal large model. The method comprises the steps of obtaining multi-modal environment data and a natural language instruction containing at least one target task in response to a heterogeneous sensor, uploading the multi-modal environment data and the natural language instruction to a cloud end, and generating an analyzable task plan according to the multi-modal environment data and the natural language instruction through a cognitive loop; splitting the analyzable task plan into sub-tasks corresponding to each robot, generating a task mapping table based on the sub-tasks, and issuing the task mapping table to the robot, so as to execute the task mapping table through a control loop of the robot; the problem that an existing robot cannot perform effective cooperative operation in the process of executing the intention of a worker in time can be effectively solved.

Disclosed herein is a method for performing a robot skill based on skill uncertainty using a large language model. The method includes generating a subtask list using a large language model by receiving a target task and environment information, mapping a subtask in the subtask list into a skill embedding space through an abstract skill policy network, and performing the subtask by decoding the mapped subtask through a manipulation skill policy network.

This invention relates to an energy-saving scheduling method for inaccurate hybrid critical tasks based on a genetic algorithm, comprising the following steps: establishing an inaccurate hybrid critical task scheduling model; encoding the mapping of inaccurate hybrid critical tasks to multiple processors; formulating a fitness function to evaluate and select task mapping schemes to processors; performing crossover and mutation operations on existing task mapping schemes; and calculating the energy-efficient speed of each processor's task set. This invention, based on a genetic algorithm, seeks task mapping schemes that balance the load across multiple processors and calculates the optimal speed for calculating the energy consumption of each processor, ensuring that all tasks meet deadline requirements and reducing system energy consumption.

A data transmission method and apparatus, the method comprising: mapping M tasks of a first service to N logical channels, M and N being positive integers greater than or equal to 2; each task in the M tasks comprises one or more messages; and executing a logical channel priority (LCP), the logical channels for executing the LCP including L logical channels in the N logical channels, and L being a positive integer less than or equal to N. The method can improve the data transmission performance.

The application discloses a kind of unified impedance control method and system of redundant mechanical arm, control method includes: obtaining the state data of each joint of redundant mechanical arm and the desired pose of end effector, and calculating end actual pose;Task mapping matrix and task space impedance control quantity are obtained by solving;Zero space base vector is constructed, and corresponding zero space control quantity is calculated;Task mapping matrix and zero space base vector are fused to construct generalized mapping matrix, and task space impedance control quantity and zero space control quantity are fused to construct generalized control force vector;Generalized jacobian matrix and seven-dimensional generalized control force vector are calculated by matrix multiplication operation to obtain each joint control moment, and after limiting, it is output to servo driver.In the application, by constructing generalized space, joint torque can be directly output by once matrix multiplication, task and zero space are decoupled, calculation efficiency is improved, and high control precision and motion stability are obtained.

The application provides a robot control method based on Model-Based and RL, and relates to the technical field of automatic control.The method maps a target task into a state node by constructing a high-level task graph model and a predefined task action set, performs graph search in combination with a current world state, plans an optimal task action path, and outputs a first action instruction.In a simulation environment, a baseline strategy is called to generate a control instruction, a feasibility check is performed, a simulation and real environment deviation is compensated through an error prediction model, a real control instruction is corrected, and a robot is driven to execute.The success probability weight of state transition in the graph is dynamically updated based on the execution result, and the state after execution is fed back for subsequent decision-making.The application improves the reliability, adaptability and execution safety of simulation strategy migration to a real robot, and realizes closed-loop optimization and continuous self-adaptation of a complex task.

A computer-implemented method can receive an internal representation of a transformer model which defines one or more repeating blocks, each block including a sequence of cells, and each cell including a set of tasks of the transformer model. The method can search for a plurality of parallel schedules for partitioning devices included in a device cluster for parallel execution of the transformer model. The searching includes determining a number of model replicas, determining a number of stages that divide the one or more repeating blocks, determining a number of cell replicas for each cell in a block, and for each cell replica of a cell, generating a task mapping which maps the set of tasks included in the cell to devices partitioned into the cell replica.

The invention discloses an engineering construction progress management system and method based on BIM and task mapping, a BIM three-dimensional model and sub-item data are stored in a centralized manner through a basic data module, a mapping relationship between a BIM component and a sub-item is established, a task and a sub-item are compulsively associated when the task is created through a task management module, and the task management module is used for establishing a task. Indirectly realizing logic connection between the task and the BIM component; meanwhile, progress visualization and dynamic interaction based on the BIM model are presented through a Web end visualization display module by means of the WebGL technology, complete data link support is provided for progress animation simulation based on the BIM, seamless connection from the management dimension to the visualization dimension is achieved, and the management efficiency and execution force of engineering construction are effectively improved.

This invention discloses a real-time optimization method and system for a robot control board based on TSN and a heterogeneous architecture. First, the task status, resource utilization, and communication status of the robot's heterogeneous computing units are collected to generate a system state matrix. A global resource decision model is then input to generate a collaborative optimization instruction set, including resource mapping adjustment instructions for computing tasks and traffic scheduling configuration instructions for the TSN network. Based on this instruction set, the heterogeneous computing scheduler and TSN configurator are invoked respectively to dynamically adjust the mapping relationship of computing tasks and the traffic scheduling strategy of the network, thereby generating adjusted task mapping relationships and network scheduling strategies. The adjusted strategies are then applied to run the robot control board, and state acquisition is restarted, forming a closed-loop optimization. This application achieves joint optimization of computing and network resources through global collaborative decision-making and scheduling, effectively improving the real-time determinism, resource utilization efficiency, and dynamic environment adaptability of the robot control system.

A computer-implemented method can receive an internal representation of a transformer model which defines one or more repeating blocks, each block including a sequence of cells, and each cell including a set of tasks of the transformer model. The method can search for a plurality of parallel schedules for partitioning devices included in a device cluster for parallel execution of the transformer model. The searching includes determining a number of model replicas, determining a number of stages that divide the one or more repeating blocks, determining a number of cell replicas for each cell in a block, and for each cell replica of a cell, generating a task mapping which maps the set of tasks included in the cell to devices partitioned into the cell replica.

The invention relates to a 2.5 D-based multi-core graphics processor cluster task mapping method, and aims to flexibly schedule computing resources in a multi-core integrated system, improve task execution efficiency and reduce system energy consumption. The invention provides a two-stage task mapping algorithm, the first stage is a scene-aware inter-core-particle mapping method, the scene-aware inter-core-particle mapping method performs scene-based classification management on tasks, and comprehensively considers the computing resource isomerism and inter-core-particle communication overhead in the mapping process, so that a heuristic mapping strategy with low complexity and weak information dependence is realized. In the second stage, a dynamic task mapping method in a core grain based on a proportional-integral-differential controller is provided, and optimal resource allocation in the core grain is guided to be realized in a mode of combining task initial mapping, periodic sampling, control algorithm decision and task remapping, so that the resource utilization rate is improved, and the service quality is guaranteed. The method has the characteristics of good expandability, low delay and low power consumption, and is suitable for a large-scale core particle integrated system.

A system for intelligent resource planning in a multiprocessor computing environment, consisting of: a multitude of processing units arranged in a computing machine and configured to perform computing tasks simultaneously; a non-volatile memory that is operationally connected to the multitude of processing units and is configured to store task descriptions, processor state data, and historical execution logs; a task intake unit that is operationally connected to the non-volatile memory and configured to receive task descriptions corresponding to executable workloads, each task description containing characteristics of the workload, indicators of memory access behavior, information about execution dependency, and parameters for time constraints; a processor state monitoring unit that is operationally connected to the multitude of processing units and is configured to collect processor-specific state data during runtime, including utilization level, cache occupancy state, execution throughput behavior, power consumption state, and thermal operating state; a predictive task allocation unit executed by one or more processors and operationally coupled to the non-volatile memory, wherein the predictive task allocation unit is configured to analyze the task descriptions together with the processor-specific state data and historical execution records to generate predicted execution suitability scores corresponding to the assignment of each task to each processing unit prior to task execution; a scheduling control unit executed by one or more processors, which is operationally coupled to the Predictive Task Mapping Unit, wherein the scheduling control unit is configured to assign each received task to a selected processing unit based on the predicted execution suitability values, while suppressing unnecessary task migrations; and a feedback adjustment unit that is operationally coupled with the planning control unit and is configured to monitor the actual execution results of the distributed tasks and update the historical execution records stored in non-volatile memory to refine subsequent predictive task assignment operations.

The invention relates to the technical field of service management, in particular to an automobile aftermarket service method and system based on model fusion, and the method comprises the following steps: obtaining a service request and aggregating and analyzing data, matching historical records to generate mapping, fusing a label to reconstruct a task template, and outputting a service result. According to the method, the operation content and the accessory information in the service request are collected and associated, time sequence aggregation and synchronous marking are achieved in combination with the time label, the structured degree of the service data is improved, matching and sequence offset analysis are carried out on historical service records, unified confirmation of multiple judgment results is completed through a label fusion strategy, and the service quality is improved. Reconstruction of a service template is guided through a structure fusion label, the suitability of task combination and the rationality of an execution path are ensured, the configuration process of service tasks is optimized through a task mapping structure, service requests and execution track information are integrated, and the pertinence of automobile post-service recommendation and the real-time performance of response are effectively improved.

The embodiment of the invention provides a GPU load balancing method and system applied to a server cluster, and relates to the technical field of computers.The method comprises the steps that firstly, a mapping relation between GPU load fingerprints (covering computing unit loads, video memory occupancy and load change rate data) and task characteristics (including task data amount, computing complexity and data dependence attributes) is modeled; generating a load task mapping data set; constructing a dynamic load compensation pool based on the data set, wherein the dynamic load compensation pool comprises a GPU node group adaptive to load characteristics, task migration link parameters and a load allocation rule; splitting the task data according to rules and transmitting the task data to corresponding node groups for execution to generate task migration execution data; generating a calibration result based on the execution data calibration compensation pool parameters and rules; and a result is fed back to the mapping data set, configuration is updated to form a cyclic adaptation process, server cluster GPU load balancing is achieved, and the calculation efficiency and the resource utilization rate are improved.

The application provides a Hawk algorithm acceleration method based on GPU multi-thread parallelism. The Hawk algorithm acceleration method based on GPU multi-thread parallelism comprises the following steps: (1) thread resource division and task mapping; (2) kernel fusion and throughput peak detection; (3) NTT / iNTT parallelization reconstruction and memory access optimization; (4) FFT / iFFT structured parallelization optimization and butterfly operator fusion. The application realizes a coarse-grained parallel butterfly operator execution mode without complex address calculation, without shared memory synchronization, and without competition between threads, greatly improving the throughput efficiency on the GPU; meanwhile, the application unifies the parallel execution framework in the forward and inverse transformations of FFT / NTT, wherein the inverse transformation only needs to use the corresponding inverse rotation factor and perform simple normalization processing at the end to complete the overall recovery.