47 results about "Fast path" patented technology

The invention provides a data processing method and device of an intelligent network card, equipment and a storage medium, and the method comprises the steps: carrying out the industrial protocol analysis processing of a received data packet, extracting an equipment identifier and request type information, and distributing the data packet to a fast path or a standard path according to the request type information; performing hotspot detection processing on the data packets in the fast path, performing hotspot prediction in combination with a detection result and a preset equipment association relationship, and generating a hotspot data set; performing dynamic management on multi-level caches of the intelligent network card according to the hotspot data set, and distributing the hotspot data to corresponding cache levels according to timeliness; and when a query request is received, performing response processing according to the request type information and the multi-level cache, and when the query content is hit, constructing a response data packet in the intelligent network card and directly sending the response data packet. According to the method, the data processing task is unloaded through the intelligent network card, CPU resource consumption is reduced, dynamic changes of data hotspots are adapted, and the system response speed is increased.

The invention discloses a video behavior recognition method based on multi-scale time feature fusion, and the method comprises the steps: taking a dual-path network as a basic structure, carrying out the normalization and uniform frame extraction of an input video, and respectively inputting a Slow path and a Fast path for feature extraction. The Slow path focuses on low-frequency background semantic modeling, the Fast path integrates a dynamic Token sparse module (VDTS) constructed based on one-dimensional convolution, and a key action area is adaptively selected through a Gumbel Top-K mechanism. Furthermore, according to the method, a gating bidirectional cross attention module (Bi-CAM) is designed between a fast path and a slow path, dynamic complementary fusion of bidirectional information streams is achieved by introducing a Query-Key-Value fusion structure and a channel attention gating mechanism, and the coupling efficiency and expression ability between multi-scale features are effectively improved. And finally, accurate recognition of behavior categories is completed through a classifier. The method gives consideration to the recognition precision and the calculation efficiency, is suitable for scenes such as behavior monitoring, man-machine interaction and intelligent security and protection, and has a good application prospect.

The invention discloses a language interactive automatic driving method and device based on an adaptive slow-fast system. The method comprises the following steps: generating a feature vector sequence fusing a long-range context and key features of a current frame; constructing a parallel dual-path decision system; cooperating with a dual-path decision system through an adaptive connector system; the dynamic activation controller generates a binary decision according to the complexity of the current scene; if it is judged that LLM intervention is not needed, the trajectory is output completely depending on the fast path; if the LLM needs to be activated, a slow path extracts historical contexts to generate semantic features, a feature fusion controller calculates a weighting coefficient according to confidence, and finally, a trajectory planner fuses the two types of features to generate an optimized trajectory. According to the fast path, a lightweight planner is adopted to realize high-frequency frame-level trajectory prediction, and real-time response of basic driving control is ensured; and the slow path realizes low-frequency semantic reasoning through a large language model, and an adaptive judgment mechanism of LLM aided decision is introduced to realize optimal balance of automatic driving safety and real-time performance.

The invention discloses a message processing method and device, equipment, a storage medium and a product, and the method comprises the steps: receiving a non-first packet message sent by an intelligent network card through a fast path module; performing fast path matching on the non-first packet message through a fast path module; and when the fast path matching of the non-first packet message fails, performing flow size and anomaly analysis on the non-first packet message by a local management and control surface, deciding whether to generate a flow table for guiding the non-first packet message to work or not, and sending the flow table to the intelligent network card. By adopting the embodiment of the invention, the CPU overhead of the host can be reduced, and the forwarding efficiency is improved.

A method for selecting a trajectory for a vehicle includes receiving first perception data from one or more sensors positioned on the vehicle at a first time; generating at least one nominal trajectory based on the first perception data; receiving second perception data from the one or more sensors at a second time after the first time; generating at least one fast-path trajectory based on the second perception data; selecting a trajectory from the at least one nominal trajectory and the at least one fast-path trajectory.

The invention relates to the technical field of computer network communication, and provides a hierarchical bitmap-based RDMA (Remote Direct Memory Access) adaptive selection retransmission method, which comprises the following steps of: performing initialization and connection configuration, completing hierarchical bitmap pool initialization on a hardware level of an intelligent network card, and dynamically allocating resources and determining a data processing path when RDMA connection is established; fast path data transmission and out-of-order packet processing are carried out, packet loss recovery of flow in a data center is realized through a hierarchical bitmap and software-hardware cooperation logic, and ordered transmission of data is ensured; low-speed path data transmission and out-of-order packet processing are carried out, packet loss recovery of cross-center long-distance traffic is realized through hierarchical bitmap and software-hardware collaborative logic, and ordered transmission of data is ensured; connection destruction and resource recovery: releasing hierarchical bitmap resources and resetting a system state when RDMA connection is finished, and providing available resources for new connection; and ending. The method is especially suitable for efficient data transmission optimization in a long-distance lossy network environment.

The invention discloses a power grid system accident monitoring and early warning system, which is characterized in that through combination of discrete wavelet transform and adaptive reclosing logic, combination of random walk and a Skip-Gram model, and combination of principal component analysis and a variational auto-encoder, dimensionality reduction and fusion are carried out on multiple types of feature vectors, so that low-dimensional efficient expression of multi-dimensional features is realized; the method improves the data processing efficiency, distributes dynamic weights for different types of features through the combination of improved Bayesian reasoning and a Transform model, carries out the weighted splicing, inputs the dynamic weights to a deep learning model for training, achieves the more accurate recognition of the operation state of a power grid, achieves the recognition of the operation state of the power grid through the combination of the construction of a hexagonal network and the improved A * algorithm, and achieves the recognition of the operation state of the power grid. According to the method, rapid path planning of the accident report is realized, and a smooth and stable transmission path is generated through combination of the Bezier curve and the dynamic reachability threshold, so that the real-time performance of data transmission is improved.

The invention relates to a flow table processing method and device, a chip, a network interface card, equipment, a medium and a program product, and is applied to a hardware programmable processor, the hardware programmable processor comprises a forwarding thread and a background thread, and the method comprises the following steps: the forwarding thread, when determining that a data message does not hit a fast path flow table, sends the data message to the hardware programmable processor; generating flow table data, and writing the generated flow table data into a target cache block of a target cache space; informing a table look-up engine of writing completion information through a forwarding thread, wherein the writing completion information is used for indicating the table look-up engine to configure a fast path flow table based on the flow table data written into the target cache block; and the background thread obtains a processing result of the table look-up engine, releases the target cache block based on the processing result, and informs the forwarding thread of the released target cache block. By adopting the method, the flow table processing efficiency can be improved.

The invention relates to a distributed heterogeneous unmanned aerial vehicle time-sensitive task allocation and flight path collaborative planning method, which comprises a fast path cost dynamic estimation module, a distributed task allocation and path planning coupling module and a local reset module for new tasks in a dynamic environment. The method has the advantages that a real-time evaluation mechanism of the path cost is embedded into a task allocation process, deep fusion of task allocation and path reachability is achieved, and decision deviation caused by decoupling processing is effectively avoided. On the basis, a local reset strategy based on a task time window and path cost is introduced, the response capability of the system to dynamic task changes is remarkably improved, and the calculation overhead is reduced. Meanwhile, an improved consensus-based task coordination framework is adopted, and an adaptive communication mechanism is fused, so that the robustness and convergence speed of the algorithm under a time-varying network condition are enhanced.

The invention discloses a sequence data hierarchical storage method and system based on a data processing unit, and the method comprises the steps: dividing sequence data into stock data stored in a persistent storage device and incremental data cached in a host memory; the DPU performs high-speed matching and identification on request types, and classifies the request types into additional requests and query requests. And for the additional request, the DPU executes a fast path forwarding strategy and directly routes the data to an incremental data area of a host memory, so that low-delay writing is realized. For the query request, the DPU executes data preprocessing of hardware acceleration, carries out push filtering, mode matching or decompression on stock data, and only transmits filtered effective result data to a host CPU for subsequent processing; intelligent shunting and preprocessing of read-write requests are realized by utilizing the DPU at a system entrance, and a physically isolated read-write path is constructed, so that the overhead of a host CPU (Central Processing Unit) is remarkably reduced, and the query performance and the write-in throughput of large-scale sequence data are effectively improved.

The application discloses a mower path planning method and belongs to the field of agricultural robots. A full-coverage path planning algorithm based on a DARP+ simulated annealing algorithm+ geometric method is provided. The full-coverage path planning algorithm based on the application has the advantages of small overall cost, less turning times of the mower, high coverage rate, small operation amount of the generated path, fast path generation speed, obstacle avoidance, consideration of the vehicle size when setting the safety distance, high safety of the generated path, avoidance of repeated work and omission in the mowing operation, high mowing efficiency and greatly reduced cost. The method can be used for full-coverage path planning of the mower in any land and can avoid obstacles.

Techniques are provided for routing qubit data. For example, one or more embodiments described herein can include a computer-implemented method for training a quantum controller fast path interface that can control the routing of the qubit data. The computer-implemented method can include training, by a system operably coupled to a processor, the quantum controller fast path interface for routing qubit data bits between a quantum controller and a conditional engine by adjusting a delay value such that an equalized clock domain is characterized by a direct register-to-register transfer mode.

The invention discloses a power transmission line dynamic first-aid repair scheduling and path planning method and system under disasters, and relates to the technical field of power emergency first-aid repair, and the method comprises the steps: sensing the topological dynamic change of a power transmission line under the disasters in real time, fusing the geographic position priority of important users and the distribution information of first-aid repair resources, and carrying out the first-aid repair scheduling and path planning. Constructing a dynamic resource scheduling model under various constraint conditions; and generating a first-aid repair path based on a fastest path algorithm, establishing a recoverability evaluation system including dimensions such as resource matching degree and topological connectivity, and triggering a first-aid repair action when an evaluation value reaches 80% or above. According to the method, the problems that the traditional emergency repair scheduling response is lagged, the path planning is separated from the disaster real-time scene, and the resource configuration and the recovery demand are not matched are solved, the accuracy, the dynamic property and the high efficiency of the emergency repair of the power transmission line under the disaster condition are realized, and the post-disaster recovery efficiency of the power grid is remarkably improved.

Systems and methods for reducing read application in a virtual storage system are provided. According to one embodiment, read amplification is reduced when AZCS compression is being utilized by avoiding restarting of a read process via a slow path via a RAID layer of the virtual storage system when a data block associated with a read request and obtained via a first fast path read has been found not to be compressed. Instead, a second fast path read may be performed to obtain the corresponding checksum. Alternatively, or additionally, heuristics may be used to predict the odds of the data block being compressed. For example, when information encoded within a PVBN of the data block that identifies the PVBN as being within a compressed AZCS zone has shown to be sufficiently / insufficiently predictive of the data block being compressed, then a flag may be set to enable / disable fast path reads.

The method for recording the memory allocation information comprises the following steps of: setting the state of a switch by adopting a dynamic mechanism under a hosting language; wherein the dynamic mechanism comprises setting a value of a variable, calling an interface and / or sending a signal; compiling the application program to generate an executable program; the executable program comprises information of a memory allocation path, and the information of the memory allocation path comprises a fast path and a slow path; the state of the switch is used for indicating a memory allocation path; before the executable program is executed or in the execution process of the executable program, the state of the switch is obtained; and recording memory allocation information according to a memory allocation path indicated by the state of the switch. According to the method for recording the memory allocation information, before the program is started and in the program execution process, the memory allocation log function is dynamically opened, and the memory allocation information is obtained; a user can be helped to quickly position the source code position associated with the memory allocation problem, the memory problem positioning time is shortened, and the analysis efficiency of memory problems is greatly improved.

The application provides a memory exception processing method and an electronic device; the method comprises the following steps: detecting that an exception occurs in a link between a computing high-speed link memory pool and the link, and acquiring link exception information; based on the link exception information, scheduling a fast path for accessing the computing high-speed link memory pool to a slow path for accessing a local memory space; acquiring a memory operation log of the computing high-speed link memory pool, and based on the memory operation log, restoring memory data between a latest checkpoint and the exception to the local memory space; the memory state before the checkpoint has been persisted; and based on the slow path and the memory data, continuing to perform read-write operations on the local memory space.

The invention relates to a layered fast path planning method for a large-scale complex environment, and the method comprises the steps: carrying out the connectivity segmentation of a tile-type map, taking each subregion obtained through segmentation as a node, and carrying out the long-range planning through a depth-first search algorithm; sampling strategies of RRT and RRT * algorithms are improved, a heuristic exploration mechanism is added in the sampling process, the improved RRT algorithm is firstly called to sample the surrounding environment during path planning, the improved RRT algorithm is directly used for a sub-region with a relatively simple obstacle environment, and the improved RRT * algorithm is switched for a sub-region with a relatively complex obstacle environment; the historical planning path is cached, key path points in the current environment and the historical environment are matched, and if matching succeeds, the historical planning path is fused into a new planning task. The problems of low planning efficiency, poor path quality, search space explosion, large influence of frequent re-planning and the like caused by tile-type maps and meter-level or below high-precision coordinate requirements are solved.

The invention provides a memory processing method and device and a computer storage medium. The memory processing method comprises the steps of obtaining a read-write request; determining whether target data accessed by the read-write request is hot data or not; if yes, a hot data memory cache pool is used for completing the read-write request, and the hot data memory cache pool is located in a block device driving layer. By means of the mode, read access is completely completed in the memory cache pool of the driving layer, and a fast path which does not need to access physical storage equipment is formed. According to the path, the physical I / O operation is thoroughly eliminated, and the performance bottleneck caused by read delay and track seeking time in the prior art is solved.

The invention belongs to the technical field of action recognition, and particularly relates to an unmanned aerial vehicle simulation training method and system based on action recognition, and the method comprises the steps: obtaining a low-frame-rate spatial stream image sequence and a high-frame-rate time stream image sequence which are synchronized in parallel; generating a fast path suppression coefficient based on the brightness homogeneity and the motion space dispersion of the image, and performing feature gating modulation on the high-frame-rate time stream; and extracting features through a double-flow network, performing dual verification in combination with an environmental noise index, and finally generating a control instruction. According to the method, an asymmetric double-flow framework is combined with a physical feature gating mechanism, real limb movement and environment illumination sudden change can be distinguished, the problem of mistaken touch in a complex illumination environment is avoided, and the interaction accuracy and robustness of a simulation training system are improved while calculation redundancy is reduced.

The present invention belongs to the field of artificial intelligence, and discloses a meta-heuristic method for rapid search and obstacle avoidance of a robot path, in order to solve the problems of slow convergence and long planning time caused by the random sampling strategy of the existing robot obstacle avoidance algorithm. The present invention integrates the improved Gaussian mixture distribution sampling strategy and the map complexity constraint strategy to implement a hybrid sampling strategy. The present invention reduces the random sampling points in the obstacle-free area and increases the random sampling points in the area close to the obstacle, thereby improving the generation speed and sampling efficiency of the planned path, and the obstacle avoidance search process requires the joint space to be fully controllable, without breaking the physical limits of the joint control position and speed, thereby ensuring the safe operation of the robot arm.

This paper proposes a fast path generation algorithm for a given set of waypoints with initialized start and end heading angles. This algorithm considers the kinematic constraints of fixed-wing drones and the directional constraints of the start and end points, generating a smooth trajectory using the Dubins curve. We decouple a set of waypoints into subpaths consisting of paths between two points. We mathematically prove that the shortest path type is CS (where C represents a curve and S represents a straight line) when the end heading angle is not fixed. Furthermore, we expand the two-point approach to three points, fixing the start and end positions and heading angles, and randomly generating a waypoint as the intermediate point. Since the shortest path between two points exists and is unique, the optimal trajectory for the three points can be determined, demonstrating that RSR outperforms CS. Finally, multiple points are inserted between the start and end points. Using the above conclusions, an iterative method is used to determine the heading of the current task point based on the headings of the previous and next task points, effectively shortening the total path length. This shortened solution is then used as the initial solution, and a neighborhood search algorithm is used to iteratively optimize the solution until convergence. Because of the excellent initial solution, our algorithm can quickly converge to the optimal value.

The application relates to a flow table processing method and device, a chip, a network interface card, equipment, a medium and a program product, which are applied to a hardware programmable processor, the hardware programmable processor comprising a forwarding thread and a background thread, and the method comprises the following steps: the forwarding thread generates flow table data and writes the generated flow table data into a target cache block of a target cache space under the condition that it is determined that a data packet does not hit a fast path flow table; the forwarding thread informs a lookup engine of information about writing completion, wherein the information about writing completion is used to instruct the lookup engine to configure a fast path flow table based on the flow table data written into the target cache block; the background thread acquires a processing result of the lookup engine, releases the target cache block based on the processing result, and informs the forwarding thread of the released target cache block. The method can improve the flow table processing efficiency.

The invention discloses a two-stage collaborative planning and control method and system for motion grabbing of a composite robot, and aims to solve the problems that the existing composite robot mostly adopts a method of'stopping first and then grabbing 'during a grabbing task, and the motion grabbing of a chassis to a target in the motion process is still faced with huge challenges, and the grabbing efficiency is poor. The method comprises the following steps: a first stage, remote approaching: providing a spatial constraint visual servo method based on a chassis path point as a reference, taking the chassis path point as the reference, establishing a dynamic constraint space for a mechanical arm, and ensuring continuous tracking of a target when the mechanical arm moves on a chassis; in the second stage, in order to grasp the short grabbing time, a fast path planning algorithm is provided, the sampling range is constrained, the path planning time is shortened, it is ensured that the grabbing target is not missed when the chassis continuously moves, different contrast experiments are set, and the contrast experiments show that the success rate of the method reaches 96% at the speed of 0.3 m / s.

To provide a relay device, a relay method and a relay program for reducing the load of arithmetic processing and the restriction of a setting value relating to communication connection management.SOLUTION: A relay device 1 that relays communication between a first communication net WAN4 and a second communication net LAN5 includes a slow path transfer processing unit that performs slow path transfer of the communication, slow path entry information including connection information and a timer value of the communication subjected to the slow path transfer, a first management unit that manages the slow path entry information, a fast path transfer processing unit that performs fast path transfer of the communication at a higher speed than the slow path transfer processing unit, fast path entry information including connection information and a timer value of the communication subjected to the fast path transfer, and a second management unit that manages the fast path entry information. When the timer value of the slow path entry information expires, the first management unit updates the timer value of the slow path entry information by using the timer value of the fast path entry information.SELECTED DRAWING: Figure 1

The invention discloses an AI accelerator optimization method based on a multipath kernel software stack and a system thereof, which are characterized in that a zero-copy, automatic reverse address mapping and dynamic load balancing method is adopted, high-compatibility accelerator access support is realized, efficient reasoning is realized through a double-path architecture, and the stability of the system is improved. The slow path calls a user mode accelerator library through a user mode daemon process to provide wide compatibility, the fast path realizes direct access of a kernel to an accelerator through a hijacking working environment, and zero-copy data transmission, automatic reverse address mapping, dynamic load balancing and a self-adaptive water level mechanism are introduced to further optimize memory management and task scheduling. Compared with the prior art, the method has the advantages that inherent delay and end-to-end delay of kernel machine learning reasoning are reduced, kernel-user space communication overhead is eliminated, system performance is remarkably improved, the problems of high reasoning delay and poor compatibility caused by the fact that an operating system kernel cannot directly access a hardware accelerator are solved, and the method has good application prospects.

This disclosure provides a communication system and a method, a traffic splitting controller, and a switch implemented by the communication system. The communication system includes a switch and a traffic splitting controller, the traffic splitting controller storing a fast path table, determining a subset of the fast path table, and sending an instruction to the switch to insert the subset into a connection table stored at the switch. The switch, when forwarding network traffic at least partially based on entries in the connection table, forwards the network traffic to a virtual machine (VM) host at least partially based on the subset, bypassing a gateway. The fast path table contains information about fast paths used by the switch to forward the network traffic such that the network traffic bypasses the gateway.

The invention discloses an efficient path planning method based on improved JPS, which dynamically adjusts the priority and search range of jump points according to the global and local characteristics of a current node through a dynamic jump point selection strategy, reduces redundant search and improves the path planning efficiency. And in combination with a periodic directed edge storage structure, the storage space requirement is greatly reduced, and meanwhile, quick query of path information is supported. A multi-thread parallel processing technology is utilized, path planning tasks are dynamically distributed to multiple threads to be executed, and efficient utilization of computing resources is achieved through a load balancing strategy. Path extension nodes are reduced through hop point search optimization, memory occupation is reduced through storage structure optimization, the operation speed is increased through parallel processing, and all links form a synergistic effect, so that the overall algorithm performance is remarkably improved, local optimal traps are avoided, and a globally optimal solution is found more efficiently.

Processing commands received from a host computing device by a storage device can require a large amount of processing overhead. This demand for ever greater processing power increases as the size of storage devices increase. Traditional methods have added an increasing number of processors or CPUs to handle these requirements. However, by utilizing a fast path accelerated processing pipeline, additional processors may not be necessary. An accelerated processing pipeline can be configured to bypass one or more steps that are required by non-priority processing pipelines. Each received command can be parsed to determine if it is suitable for accelerated processing. The command can be required to access data in a limited region of the memory device, or to have any data necessary to process the command already in a cache memory. Upon completion of verifications, commands can be placed in a priority queue that is processed before a non-priority queue.

The invention relates to the field of mobile robot path planning, in particular to a robot fast path optimization method based on multi-strategy fusion, which comprises the following steps of: 1, initializing parameters of an iterative greedy algorithm, and generating an initial path by using a bidirectional fast expansion random tree method; the method comprises the following steps of: 1, carrying out damage reconstruction on a path and expanding a path exploration space by using a difference-guided path optimization mechanism, 3, selecting to execute path pruning, node insertion and node mutation operations through a self-adaptive adjustment strategy and carrying out local search on a current path, 4, receiving the path according to a simulated annealing criterion, and 5, judging whether algorithm termination time is reached or not, and if the algorithm termination time is reached, ending the algorithm, and outputting the current optimal path, otherwise, returning to the step 2. The application of the method effectively solves the problem of minimizing the path length and the path smoothness in robot path planning, and has the positive effects of improving the path planning efficiency and optimizing the path quality.

The application discloses a dynamic gesture recognition method and device, equipment and storage medium, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring a to-be-recognized video containing a dynamic gesture, performing key frame extraction on the to-be-recognized video to obtain a target key frame; inputting the target key frame into a target double-path network to obtain first feature information output by a slow path in the target double-path network and second feature information output by a fast path in the target double-path network; wherein a convolution layer of the slow path is constructed based on cascaded small convolution kernels, and a convolution layer of the fast path is constructed based on a 3D deep separable convolution; and performing feature weight distribution and feature fusion on the first feature information and the second feature information by using an attention module to obtain fused features, so that a classifier performs gesture type recognition according to the fused features. The method can reduce the calculation cost and improve the recognition speed and accuracy.