Real-time interaction system and method for robot and peripheral based on edge computing

Abstract

The application discloses a kind of real-time interaction system and method of robot and peripheral based on edge computing, belong to robot control technical field.System includes regional edge node for determining task hierarchical information according to the semantic feature vector of global perception data, global decision is carried out, and scheduling is carried out to shared backup node in regional redundant resource pool, and regional division is determined based on task throughput, data aggregation demand and energy consumption constraint, and global perception data includes scene state data and micro-sensing data;Ultra near-end edge node is deployed based on task delay limit and communication reliability demand, for collecting scene state data, extracting semantic feature vector to micro-sensing data and scene state data;Robot terminal is used to collect micro-sensing data, and task execution is carried out based on task hierarchical information.The method ensures the determinacy delay of high critical task, and improves the utilization efficiency of overall resource of system.

Description

Technical Field

[0001] This application belongs to the field of robot control technology, and in particular relates to a real-time interaction system and method for robots and peripherals based on edge computing. Background Technology

[0002] In advanced industrial scenarios such as intelligent manufacturing, flexible automation, and human-robot collaboration, robots need to interact with peripherals such as force sensors, vision cameras, and electric grippers at high frequency and with high reliability. These applications place stringent requirements on the system's real-time performance, determinism, and security. How to efficiently integrate global perception information and optimize the utilization of edge resources while ensuring extreme real-time control performance has become a core challenge for current robot intelligent control systems.

[0003] To address these needs, edge computing architecture is typically used to move computing and decision-making capabilities closer to the device.

[0004] However, in complex wireless environments with multiple robots and peripherals, existing edge computing architectures struggle to simultaneously guarantee deterministic latency for highly critical tasks and overall system resource utilization. Summary of the Invention

[0005] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a real-time interaction system and method for robots and peripherals based on edge computing, which ensures deterministic latency for highly critical tasks and improves the overall resource utilization efficiency of the system.

[0006] In a first aspect, this application provides a real-time interaction system between a robot and peripherals based on edge computing. The system includes at least one regional edge node, multiple ultra-near-end edge nodes, and multiple robot terminals. Each regional edge node is connected to a micro-edge cluster. The micro-edge cluster includes a master node and at least one shared backup node. The master node and the shared backup node are divided into ultra-near-end edge nodes based on task classification information.

[0007] The ultra-near-end edge node is connected to the robot terminal based on the coupling of shared memory and time-gated scheduling;

[0008] The regional edge nodes are deployed based on the regional division of the robot terminal. They are used to determine the task classification information based on the semantic feature vector of the global perception data, make global decisions, and schedule the shared backup nodes in the regional redundant resource pool. The regional division is determined based on task throughput, data aggregation requirements, and energy consumption constraints. The global perception data includes scene status data and micro-sensor data.

[0009] The ultra-near-end edge node is deployed based on task latency constraints and communication reliability requirements, and is used to collect the scene state data and extract the semantic feature vector from the micro-sensor data and the scene state data;

[0010] The robot terminal is used to collect the microsensor data and perform task execution based on the task classification information.

[0011] According to one embodiment of this application, a control channel and a data synchronization channel are configured between the robot terminal and the ultra-near-end edge node;

[0012] The ultra-near-end edge node is configured with a gated shared memory manager, which is used to open access permissions to the mapped logical blocks within the activated time slot according to the time slot trigger signal generated by the time gating scheduling mechanism, so that the authorized robot terminal or the ultra-near-end edge node can perform zero-copy read and write to the corresponding logical blocks in the shared memory area.

[0013] The time-gating scheduling mechanism is used to divide the communication cycle into exclusive time slots, periodic time slots, and dynamic time slots according to the time sequence, allocate exclusive time slots to the control channel, and allocate corresponding periodic time slots or dynamic time slots to the data synchronization channel. The priority of the exclusive time slots is higher than that of the periodic time slots and dynamic time slots.

[0014] The shared memory region is divided into multiple logical blocks, each logical block including a control block and at least one data block;

[0015] The control block is used to map the control channel so that the ultra-near-end edge node can write the task classification information and control instructions through kernel-mode memory mapping, and the robot terminal can read them.

[0016] The data block is used to map the data synchronization channel so that the robot terminal can write the microsensor data and the ultra-near-end edge node can read it.

[0017] According to one embodiment of this application, the ultra-near-end edge node is used to adjust the period and duty cycle of the time gates of the control channel and the data synchronization channel based on the task classification information issued by the regional edge node; within the exclusive time slot of the control channel, it issues control instructions containing task classification information to the robot terminal through the control block; and allocates a data synchronization channel with a fixed position and redundant period for data streams with high criticality based on the task classification information, and the mapped data blocks are contiguous in physical memory.

[0018] The robot terminal is used to receive and respond to the control commands, and according to the task classification information, within the authorized and activated time slots, within the mapped data blocks, synchronize the microsensor data through zero-copy.

[0019] The exclusive time slot of the control channel is located at the beginning of each communication cycle, and the communication cycle is synchronized with the control decision cycle of the robot terminal.

[0020] According to one embodiment of this application, the ultra-near-end edge node integrates a joint arbitrator;

[0021] The joint arbitrator is used to adjust the suppression strength of the data transmission bandwidth of the data synchronization channel according to the deviation value between the performance index of the control channel and the service level target corresponding to the task classification information when the deviation between the end-to-end delay of the control channel and the tolerance upper limit of the task classification information is less than a preset safety margin. The deviation value is negatively correlated with the suppression strength.

[0022] The joint arbitrator is equipped with a back pressure feedback unit, which is used to send a flow control signal to the robot terminal to pause or reduce the rate at which the robot terminal writes micro-sensor data to the data block.

[0023] The ultra-near-end edge node is also used to release the bandwidth suppression of the data synchronization channel and restore the transmission rate of the data synchronization channel after the end-to-end delay of the control channel has been continuously lower than the safety recovery threshold for a preset stable period of time.

[0024] According to one embodiment of this application, the ultra-near-end edge node is used for:

[0025] Online differential encoding is performed on the semantic feature vector, and change summary data is generated based on the temporal difference between the semantic feature vector and the historical semantic feature vector.

[0026] If the temporal difference exceeds the difference threshold, the change summary data is asynchronously uploaded to the regional edge node. The change summary data is compressed using a differential encoding format and includes the semantic dimension that has changed and the incremental value of the semantic dimension, which is used for internal status log recording and anomaly detection.

[0027] The region edge node is used to receive the change summary data in order to update the global perception data.

[0028] According to one embodiment of this application, the deployment of the master node and the shared backup node in the micro-edge cluster is determined based on joint optimization of radio frequency propagation characteristics, network hop count, and switching latency;

[0029] The radio frequency propagation characteristics, including path loss, multipath delay spread, and channel coherence time, are used to evaluate the lower limit of delay jitter in wireless links.

[0030] The network hop count is used to characterize the number of intermediate forwarding devices from the ultra-near-end edge node to the regional edge node;

[0031] The switching delay is used to characterize the transmission delay of data packets in the edge network path from the ultra-near-end edge node to the regional edge node due to queuing and processing by intermediate nodes in the network.

[0032] The regional edge node is used to construct a deployment cost function based on the radio frequency propagation characteristics, the network hop count, and the switching delay, in order to determine the optimal deployment coordinate set of the micro-edge cluster.

[0033] According to one embodiment of this application, the regional edge node is used to parse the assistance intention information of each robot terminal in the robot queue. When multiple robot terminals are detected to have collaborative operation needs, the ultra-near-end edge node involving the robot queue and its corresponding shared backup node are grouped into a virtual joint node. The assistance intention information includes task dependency relationship, spatial collaboration needs and peripheral sharing requests.

[0034] All ultra-near-end edge nodes within the virtual federated node share resource views and scheduling policies, and the regional edge nodes coordinate task allocation and load balancing.

[0035] According to one embodiment of this application, the task classification information includes high-criticality tasks, medium-criticality tasks, and low-criticality tasks.

[0036] The operating system kernel layer of the ultra-near-end edge node is configured to construct mutually isolated hardware resource slices based on task classification information;

[0037] The reflection control task in a highly critical task is allocated a dedicated CPU core, a contiguous physical memory region, and a last-level cache partition;

[0038] Slices of limited computing resources are allocated to semantic feature extraction or data aggregation tasks in medium-critical and low-critical tasks.

[0039] According to one embodiment of this application, the ultra-near-end edge node is further used to extract a semantic feature vector through progressive semantic feature encoding;

[0040] The progressive semantic feature encoding generates a hierarchical data structure that includes at least a first encoding layer and a second encoding layer;

[0041] The first encoding layer is a critical status code, including a limited number of status identifiers, which are used for transmission within a first time window to characterize the security level of the current operation and the approximate state of the system.

[0042] The second encoding layer is a complete semantic feature description, used for transmission within the second time window, including the complete dimension and precision information of the semantic feature vector;

[0043] The regional edge node sends a feature request instruction to the ultra-near-end edge node according to the current network status and the urgency level corresponding to the task classification information, so as to receive the first coding layer, or the first coding layer and the second coding layer.

[0044] Secondly, this application provides a real-time interaction method between a robot and a peripheral device based on edge computing, applied to the real-time interaction system between a robot and a peripheral device based on edge computing as described in the first aspect. The method includes:

[0045] By using regional edge nodes, task classification information is determined based on the semantic feature vector of the global perception data, global decisions are made, and the shared backup nodes in the regional redundant resource pool are scheduled. The global perception data includes scene status data and micro-sensor data.

[0046] The scene state data is collected through ultra-near-end edge nodes, and the semantic feature vector is extracted from the micro-sensing data and the scene state data;

[0047] The robot terminal collects the microsensor data and performs task execution based on the task classification information.

[0048] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application.

[0049] This application provides a real-time interaction system and method for robots and peripherals based on edge computing, which has the following advantages over existing technologies:

[0050] (1) By deeply coupling the ultra-near-end edge nodes and the robot terminal through shared memory zero copy and time-gated scheduling, the uncertain delay caused by the traditional network protocol stack is avoided, and the deterministic execution of highly critical tasks is guaranteed. The regional edge nodes perform task classification and global scheduling based on semantic feature vectors, and use the cross-cluster shared backup node pool to achieve elastic disaster recovery. This avoids resource redundancy and waste, and ensures high system availability. The system can adapt to environmental changes, reduce interaction latency in complex wireless environments, ensure deterministic latency of highly critical tasks, improve the overall system resource utilization efficiency, and enhance the robot's real-time perception and control capabilities of peripherals.

[0051] (2) Through the dual-channel collaborative mechanism of the robot terminal, the control channel completes the instruction issuance in a zero-copy manner within the exclusive time slot, which can stabilize the end-to-end latency and jitter and meet the real-time requirements of the high-critical tasks of industrial robots; through the data synchronization channel, micro-sensor data is uploaded in the isolated time slot, avoiding competition with the control flow for bus bandwidth. At the same time, the round-robin mechanism of the logic block prevents read-write conflicts, reduces uncertain delay and kernel overhead, and improves the system response speed, resource utilization and operation stability, providing underlying support for high-safety and high-precision real-time robot control. Attached Figure Description

[0052] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0053] Figure 1 This is a schematic diagram of the structure of a real-time interaction system between a robot and peripheral devices based on edge computing, provided in an embodiment of this application.

[0054] Figure 2 This is a flowchart illustrating a real-time interaction method between a robot and peripheral devices based on edge computing, as provided in an embodiment of this application. Detailed Implementation

[0055] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0056] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0057] The following description, in conjunction with the accompanying drawings, details the real-time interaction system and method for robot-peripheral based on edge computing provided in this application, through specific embodiments and application scenarios.

[0058] like Figure 1As shown, the real-time interaction system between the robot and peripherals based on edge computing includes: at least one regional edge node 110, multiple ultra-near-end edge nodes 120 and multiple robot terminals 130. Each regional edge node 110 is connected to a micro-edge cluster. The micro-edge cluster includes a master node and at least one shared backup node. The master node and the shared backup node are divided into ultra-near-end edge nodes 120 based on task classification information.

[0059] The ultra-near-end edge node 120 and the robot terminal 130 are connected based on the coupling of shared memory and time-gated scheduling;

[0060] The regional edge node 110 is deployed based on the regional division of the robot terminal 130. It is used to determine the task classification information according to the semantic feature vector of the global perception data, make global decisions, and schedule the shared backup nodes in the regional redundant resource pool. The regional division is determined based on task throughput, data aggregation requirements and energy consumption constraints. The global perception data includes scene status data and micro-sensor data.

[0061] The ultra-near-end edge node 120 is deployed based on task latency constraints and communication reliability requirements, and is used to collect the scene state data and extract the semantic feature vector from the micro-sensor data and the scene state data;

[0062] The robot terminal 130 is used to collect the microsensor data and perform task execution based on the task classification information.

[0063] It is understandable that a Regional Edge Node (REN) is a high-performance edge server deployed in a workshop or production line, managing a physical area and responsible for global perception, task classification, and resource scheduling across multiple robot units. The REN is configured with a load awareness module and a node management module, used to issue start, suspend, or degrade commands based on the service load status of each near-end edge node, enabling dynamic start and stop of the near-end edge node; it also maintains a node load table and a service priority table. The load awareness module determines whether to perform a start or suspend operation on the near-end edge node based on data in the node load table regarding CPU usage, network bandwidth, number of online robots, and number of real-time tasks.

[0064] Ultra-Proximal Edge Nodes (UPENs) are lightweight edge computing units deployed extremely close to the physical location of one or more robot terminals, within 10 meters of the connected robot terminals. They are used for local real-time control and semantic feature extraction, such as embedded industrial PCs or custom system-on-chips (SoCs). UPENs also include protocol adaptation modules to convert control messages from the robot terminals into control commands recognizable by peripherals such as access control systems, elevator control systems, AGV scheduling systems, and camera trigger signals.

[0065] A micro-edge cluster is a logical computing unit consisting of a primary UPEN and at least one shared standby UPEN, serving a group of geographically proximate robotic terminals to provide highly available local edge services.

[0066] The roles of master nodes and shared standby nodes are defined in micro-edge clusters. Master nodes are responsible for exclusive processing of highly critical tasks, while shared standby nodes are in a low-power standby state and can be reused by multiple clusters to improve resource efficiency.

[0067] A robot terminal is an intelligent device with built-in sensors and actuators used to perform physical actions, such as an industrial robotic arm, an automated guided vehicle (AGV), or a collaborative robot (Cobot). The robot terminal also includes a micro-buffer and a local preprocessing module. These modules are used to temporarily store data and extract features from the collected data when there are network anomalies or increased latency. Only the preprocessed results are sent to the ultra-near-end edge node or the regional edge node. Furthermore, when the data synchronization channel is interrupted, the robot terminal continues to receive control commands from the ultra-near-end edge node through the control channel, and after the data synchronization channel is restored, it sends the data in the micro-buffer back in batches.

[0068] Task classification information is a criticality label that regional edge nodes use to classify robot tasks according to their real-time performance, safety, and severity of consequences based on semantic feature vectors. This information is used to guide resource allocation strategies. For example, emergency stop and force control closed loop are labeled as high criticality, path tracking is labeled as medium criticality, and log upload is labeled as low criticality.

[0069] Shared memory is a common memory area established between UPEN and the robot terminal via the high-speed serial computer extended bus standard (Peripheral Component Interconnect Express, PCIe) or high-speed Ethernet. Both parties can directly read and write through memory mapping (mmap), without kernel copying, thus achieving zero copy.

[0070] Time-gated scheduling is based on the Time-Sensitive Networking (TSN) mechanism, which divides the communication cycle into fixed-time transmission windows (time slots) and only allows specific traffic to be sent within the specified window.

[0071] Regional division involves dividing the physical space into several management regions based on task load, data aggregation requirements, and energy consumption constraints. Each region is governed by a REN.

[0072] Task throughput is the number of control or computation tasks that a robot terminal in a region needs to complete per unit of time.

[0073] Data aggregation requirements refer to whether raw data generated by multiple robot terminals or sensors within a region needs to be fused, compressed, or semantically extracted locally before being uploaded. For example, when multiple AGVs are collaboratively transporting goods, location data needs to be fused to generate a global path. This scenario has high data aggregation requirements and necessitates the deployment of a UPEN cluster with strong collaborative perception capabilities within the region.

[0074] Energy consumption constraints are the energy limitations faced by robot terminals in a region when performing tasks, including maximum power consumption limits, battery life requirements, or charging cycle limitations. In regions with high energy consumption constraints, task offloading strategies need to be optimized, communication overhead reduced, and region boundaries adjusted to shorten wireless transmission distances in order to extend device uptime.

[0075] Task latency limits are the maximum end-to-end time allowed for a certain type of robot task from instruction issuance to completion. They are determined by the criticality level of the task. For example, the latency limit for high-criticality tasks is usually 1-5 milliseconds, for medium-criticality tasks it is 10-20 milliseconds, and for low-criticality tasks it is 50-100 milliseconds.

[0076] Communication reliability requirements are the task's requirements for the probability of successful data delivery and anti-interference capability. They are quantified by indicators such as Packet Loss Rate (PLR) and Bit Error Rate (BER). These requirements determine the type of physical link to be used between UPEN and the robot terminal, whether to enable redundant transmission, and whether hardware-level error checking is required. This affects the deployment distance of UPEN, network topology, and hardware selection.

[0077] The global perception data covers the environmental and equipment status information of the entire area. In the global perception data, scene status data includes environmental perception data such as vision, lidar, and infrared collected by UPEN, and micro-sensor data includes low-level sensor data such as joint torque, encoder position, current, and temperature collected by the robot terminal.

[0078] The semantic feature vector is a structured high-level semantic representation extracted from the original global perception data through the edge lightweight convolutional neural network (CNN) model built into the ultra-near edge node. It is represented as a fixed-dimensional floating-point or integer array, for example, [object type = glass, position = (x, y, z), risk level = high].

[0079] In actual implementation, RENs are installed in the factory workshop according to the physical layout, for example, one per 500㎡. UPENs are deployed in the control cabinet within 1-10 meters of each robot. The robot terminal is connected to the actuator through motor driver and sensor bus.

[0080] When REN starts, it scans all UPENs in the region. Based on the computing power, network latency, and health status of each UPEN, it evaluates the performance of each UPEN, designates the UPEN with the best performance as the master node, and marks the remaining UPENs as shared standby nodes and adds them to the regional redundancy resource pool. The roles of the master node and the shared standby node can be dynamically switched through the Google Remote Procedure Call (gRPC) heartbeat protocol.

[0081] The UPEN and robot terminal are connected via a 10Gbe TSN network card or a PCIe Gen4 direct connection. The UPEN's operating system is configured with a Portable Operating System Interface X (POSIX) shared memory. At the same time, the 802.1Qbv function of the TSN switch is enabled, and time synchronization and time slot tables are configured.

[0082] REN collects historical data and, based on the task throughput corresponding to the number of control cycles per unit time, the data aggregation requirements corresponding to the multi-machine collaboration frequency, and the energy consumption constraints corresponding to the battery capacity / power limit, runs the K-means clustering algorithm to group the robot terminals, with each group forming a management area.

[0083] UPEN fuses micro-sensor data and global sensor data in the global perception data, calls the pre-trained MobileNetV3-Tiny model to infer the fused global perception data, and outputs a 128-dimensional semantic feature vector in JSON format for local control or uploading to REN.

[0084] REN receives semantic feature vectors from each UPEN and inputs these semantic feature vectors into a pre-trained lightweight Long Short-Term Memory (LSTM) classifier to generate task classification information.

[0085] Based on the task classification results, REN reallocates tasks of high-load UPENs to idle standby nodes; in the event of a heartbeat timeout of any primary UPEN, it selects the optimal standby UPEN from the regional redundant resource pool for activation and achieves seamless switching through Intellectual Property (IP) drift.

[0086] When deploying UPEN, REN decomposes the end-to-end latency constraints of the task into specific latency targets for the wireless transmission segment using a latency budget decomposition method, and quantifies reliability requirements into minimum received signal strength (RSSI) and maximum packet loss rate threshold. Based on on-site RF probe sampling and ray tracing model simulation, a digital map of the RF environment integrating measured and simulated data is constructed. Simultaneously, network planning determines the topology hop count and switching latency between candidate deployment points and the robot terminal. A multi-objective optimization algorithm is used to solve the cost function integrating task weights, predicted latency, and link quality, outputting the optimal deployment coordinate set in the solution space that satisfies all task QoS constraints. The physical deployment of UPEN is completed based on this set, and through continuous performance monitoring and adaptive triggering mechanisms, a re-optimization process is automatically initiated when performance degrades or task distribution changes, achieving dynamic optimization of the deployment location. UPEN connects to a camera via a GigEVision interface and collects scene status data via a CAN bus; it also receives micro-sensor data uploaded by the robot terminal via TSN.

[0087] The robot terminal reads data from micro-sensors such as six-dimensional force sensors and encoders. After receiving control commands from UPEN, it executes high-critical tasks and immediately interrupts other operations based on task classification information, while low-critical tasks are queued for execution.

[0088] The real-time interaction system between robots and peripherals based on edge computing provided in this application embodiment achieves this by deeply coupling ultra-near-end edge nodes and robot terminals through shared memory zero-copy and time-gated scheduling. This avoids the uncertain latency caused by traditional network protocol stacks and ensures deterministic execution of highly critical tasks. Regional edge nodes perform task classification and global scheduling based on semantic feature vectors and utilize a cross-cluster shared backup node pool to achieve elastic disaster recovery. This avoids resource redundancy and waste while ensuring high system availability. The system can adapt to environmental changes, reduce interaction latency in complex wireless environments, ensure deterministic latency for highly critical tasks, improve the overall system resource utilization efficiency, and enhance the robot's real-time perception and control capabilities of peripherals.

[0089] In some embodiments, a control channel and a data synchronization channel are configured between the robot terminal and the ultra-near-end edge node;

[0090] The ultra-near-end edge node is configured with a gated shared memory manager, which is used to open access permissions to the mapped logical blocks within the activated time slot according to the time slot trigger signal generated by the time gating scheduling mechanism, so that the authorized robot terminal or the ultra-near-end edge node can perform zero-copy read and write to the corresponding logical blocks in the shared memory area.

[0091] The time-gating scheduling mechanism is used to divide the communication cycle into exclusive time slots, periodic time slots, and dynamic time slots according to the time sequence, allocate exclusive time slots to the control channel, and allocate corresponding periodic time slots or dynamic time slots to the data synchronization channel. The priority of the exclusive time slots is higher than that of the periodic time slots and dynamic time slots.

[0092] The shared memory region is divided into multiple logical blocks, each logical block including a control block and at least one data block;

[0093] The control block is used to map the control channel so that the ultra-near-end edge node can write the task classification information and control instructions through kernel-mode memory mapping, and the robot terminal can read them.

[0094] The data block is used to map the data synchronization channel so that the robot terminal can write the microsensor data and the ultra-near-end edge node can read it.

[0095] Understandably, the control channel is a high-priority communication path used to transmit highly critical real-time control commands. The data content in the control channel includes task classification information, motion control commands, safety emergency stop signals, etc. It requires low end-to-end latency, low jitter, and high reliability, and maintains a long connection with ultra-near-end edge nodes based on the low-latency profile of DDS.

[0096] The data synchronization channel is a communication path used to upload non-critical or low-to-medium priority business data. It mainly carries micro-sensor data collected by the robot terminal, such as joint current, encoder values, temperature, etc., as well as uploaded task results, alarms, and video thumbnails, allowing for a certain delay and bandwidth fluctuation.

[0097] The gated shared memory manager is a module deployed in the operating system kernel of the ultra-near-end edge node (UPEN). It is responsible for dynamically controlling the access permissions of the shared memory region within a predefined communication cycle based on the precise time slot trigger signal generated by the time-gating scheduling mechanism.

[0098] The time-gated scheduling mechanism is implemented based on the IEEE 802.1Qbv standard. Through time-sensitive networking (TSN) switches or smart network cards, each fixed-length communication cycle is divided into multiple transmission windows that are strictly arranged in time sequence, including exclusive slots, cyclic slots, and dynamic slots.

[0099] Among them, the exclusive time slot is a high-priority time window reserved for the control channel that cannot be preempted, and only control commands are allowed to be transmitted within this window; the periodic time slot is allocated to the regular data stream at a fixed period, and the dynamic time slot is flexibly allocated by the scheduler according to the real-time load.

[0100] The shared memory region is a physical memory space established between UPEN and the robot terminal through a direct PCIe connection. Both parties can access it directly through memory mapping, avoiding the kernel copy overhead of traditional socket communication.

[0101] A logical block is a logical unit of division of the shared memory region, which is isolated from each other to prevent read / write conflicts.

[0102] Within the logic blocks, the control block is a dedicated logic block for the control channel. UPEN performs memory mapping in kernel mode through the UIO (Userspace I / O) driver, writing task classification information and control instructions into it. After authorization, the robot terminal's application or real-time control loop can directly read the contents of this block through the user-mode pointer, achieving zero-copy read and write, thereby forming a deterministic interaction channel with dual spatial and temporal isolation.

[0103] A data block is one or more logical blocks used for data synchronization channels. The robot terminal writes microsensor data in a structured format within a specified time slot, and UPEN reads and processes it in subsequent time slots.

[0104] In actual execution, a connection is established between UPEN and the robot terminal through a 10GbE physical link that supports TSN, and a shared memory region is created using the POSIX shared memory interface.

[0105] A custom gated shared memory manager driver module is loaded into the UPEN kernel. This module subscribes to the periodic clock interrupts provided by the TSN time synchronization protocol and generates time slot trigger signals according to the pre-configured time slot table.

[0106] When the system starts, the shared memory is divided into a control block and several data blocks, and access permissions are set through memory protection keys (MPK) or page table attributes. The time-gated scheduling mechanism is configured by the TSN scheduler on UPEN, which divides each communication cycle to obtain exclusive time slots, periodic time slots and dynamic time slots.

[0107] For example, the communication period is 1ms, of which 0–50μs is a dedicated time slot for the control channel, 50–900μs is a periodic time slot for the regular sensor data of the data synchronization channel, and 900–1000μs is a dynamic time slot for the burst log of the data synchronization channel.

[0108] When entering an exclusive time slot, the gated shared memory manager automatically removes write protection from the control block. The UPEN kernel thread immediately writes the latest control instructions and task hierarchy information into the block. Simultaneously, the robot terminal's real-time control task, upon detecting time slot activation, atomically reads and executes instructions through a pre-mapped user-mode pointer. Within periodic or dynamic time slots, the robot terminal writes packaged microsensor data into one of the rotating data blocks. UPEN completes the reading and releases the block before the time slot ends for use in the next round. The entire process requires no central processing unit (CPU) involvement in data copying or context switching; all operations are completed within the pre-allocated time slot window.

[0109] In addition, the robot terminal is equipped with a local micro-buffer of 1-3 seconds. When the data synchronization channel is interrupted, the control channel can still receive access control from the ultra-near end, and the robot terminal will complete the action first. After the data synchronization channel is restored, the missing status will be retransmitted in batches.

[0110] In this embodiment, through the dual-channel collaborative mechanism of the robot terminal, the control channel completes the instruction issuance in a zero-copy manner within a dedicated time slot, which can stabilize end-to-end latency and jitter and meet the real-time requirements of highly critical tasks of industrial robots. Through the data synchronization channel, micro-sensor data is uploaded in an isolated time slot, avoiding competition for bus bandwidth with the control traffic. At the same time, the round-robin mechanism of the logic block prevents read-write conflicts, reduces uncertain latency and kernel overhead, and improves system response speed, resource utilization and operational stability, providing underlying support for high-safety and high-precision real-time robot control.

[0111] In some embodiments, the ultra-near-end edge node is configured to adjust the period and duty cycle of the time gates of the control channel and the data synchronization channel according to the task classification information issued by the regional edge node; and to issue control instructions containing task classification information to the robot terminal through the control block within the exclusive time slot of the control channel; and to allocate data synchronization channels with fixed positions and redundant periods for data streams with high criticality based on the task classification information, and the mapped data blocks are contiguous in physical memory.

[0112] The robot terminal is used to receive and respond to the control commands, and according to the task classification information, within the authorized and activated time slots, within the mapped data blocks, synchronize the microsensor data through zero-copy.

[0113] The exclusive time slot of the control channel is located at the beginning of each communication cycle, and the communication cycle is synchronized with the control decision cycle of the robot terminal.

[0114] It is understandable that adjusting the period and duty cycle of the time gate is to dynamically modify the repetition interval and effective transmission window length of each communication slot in a time-sensitive network (TSN).

[0115] Fixed position refers to the transmission time slot allocated to highly critical data streams within the communication cycle defined by the time-gated scheduling mechanism. It has a definite and unchanging time offset. The relative time interval between the start time of the time slot and the start time of the cycle is constant within each cycle, ensuring that the receiver can accurately predict the arrival time of the data and avoid synchronization failure caused by scheduling jitter.

[0116] Redundancy cycles are multiple repeated transmission opportunities configured for highly critical data streams to improve reliability;

[0117] The contiguous nature of physical memory is used to characterize the fact that data blocks allocated to highly critical data streams are arranged contiguously in the address space of Dynamic Random Access Memory (DRAM), avoiding access latency caused by page table jumps;

[0118] The control decision cycle is the time interval between the robot's end effector executing a control algorithm once.

[0119] Authorized and activated time slots are time slots in which the robot terminal can only read and write shared memory within time slots that are explicitly open by the time gating mechanism and in which it has the access permission.

[0120] Zero-copy synchronization refers to reading and writing data directly through shared memory without the CPU's involvement in copying or protocol stack processing.

[0121] A shared memory pool is set up within the ultra-near-end edge node, and each data frame is written only once. Multiple consumer terminals, such as access control drivers, video analysis plugins, and task coordinators, use read-only pointers and do not copy them.

[0122] The robot terminal is equipped with a shared memory data exchange module and a priority queue scheduling module, which are used to realize low-latency data interaction between the robot and peripherals locally. For example, control messages such as opening a door, the elevator going to the 3rd floor, and obstacle avoidance pause are put into the high-priority queue, while large data frames such as video or point cloud are put into the low-priority queue or multicast distribution, without blocking control messages.

[0123] In actual execution, after completing task classification, the Region Edge Node (REN) sends the task classification information to the corresponding Upper Nearest Edge Node (UPEN) via a reliable message queue. Upon receiving this information, the UPEN calls the TSN scheduler interface to dynamically reconfigure the time gating table: if the task is highly critical, the exclusive time slot period of the control channel is set to match the robot control decision cycle and placed at the beginning of each cycle to ensure it executes first. Simultaneously, the periodic time slot of the data synchronization channel is adjusted to a shorter period, and two backup time slots at the same position are allocated within the same communication cycle to increase redundancy and prevent single-transmission failures. For data streams corresponding to highly critical tasks, the UPEN initializes shared memory... At that time, mmap is used in conjunction with HugeTLB to request a large page of physical memory with contiguous physical addresses, which is divided into multiple contiguous data blocks and mapped to the robot terminal. The robot terminal loads a predefined time slot grant table at startup, and the control task running in the real-time kernel continuously listens for time slot interrupt signals generated by the TSN hardware. When it detects that it has entered an authorized exclusive time slot, it immediately reads the control instructions containing task classification information from the control block and executes the corresponding actions. In the allocated data synchronization time slot, the robot terminal directly writes the packaged microsensor data into the continuously used data block. UPEN completes the reading before the end of the time slot, thus completing a deterministic interaction cycle from control to data synchronization.

[0124] It should be noted that in the above process, the communication cycle is strictly aligned with the control decision cycle by precise synchronization of hardware timestamps and TSN gating.

[0125] In this embodiment, by configuring exclusive time slots for priority execution of highly critical tasks, the corresponding data stream also achieves higher transmission reliability and lower access latency due to redundant cycles and physical contiguous memory. Control commands are issued at the beginning of each cycle, ensuring the timeliness of the control loop. Zero copy and precise time slot collaboration eliminate queuing, copying and scheduling jitter in traditional communication, while supporting multi-level tasks to allocate bandwidth on demand.

[0126] In some embodiments, the ultra-near-end edge node integrates a joint arbitrator;

[0127] The joint arbitrator is used to adjust the suppression strength of the data transmission bandwidth of the data synchronization channel according to the deviation value between the performance index of the control channel and the service level target corresponding to the task classification information when the deviation between the end-to-end delay of the control channel and the tolerance upper limit of the task classification information is less than a preset safety margin. The deviation value is negatively correlated with the suppression strength.

[0128] The joint arbitrator is equipped with a back pressure feedback unit, which is used to send a flow control signal to the robot terminal to pause or reduce the rate at which the robot terminal writes micro-sensor data to the data block.

[0129] The ultra-near-end edge node is also used to release the bandwidth suppression of the data synchronization channel and restore the transmission rate of the data synchronization channel after the end-to-end delay of the control channel has been continuously lower than the safety recovery threshold for a preset stable period of time.

[0130] Understandably, the joint arbitrator is a hardware-assisted or kernel-level software module integrated into the ultra-near-end edge node (UPEN) to monitor control channel performance in real time and dynamically coordinate resource allocation between control and data channels.

[0131] The end-to-end delay of the control channel is used to describe the complete closed-loop time from the generation of control commands by UPEN to the completion of execution feedback by the robot terminal. It is calculated by timestamping in shared memory and by synchronizing the clocks of both parties.

[0132] The tolerance limit corresponding to the task classification information is the maximum latency threshold allowed by the system for each task level. For example, it is 5 milliseconds for highly critical tasks and 20 milliseconds for medium-critical tasks.

[0133] The preset safety margin is a conservative buffer value used to trigger the protection mechanism in advance and prevent the latency from approaching the limit.

[0134] Performance metrics are specific measurable parameters used to quantify the current operating quality of the control channel. They are collected in real time through hardware timestamps, shared memory marking, or TSN probes. For example, performance metrics include end-to-end latency, latency jitter, and packet loss rate. Latency jitter is used to characterize the standard deviation or peak-to-peak fluctuation of latency.

[0135] The Service Level Objective (SLO) corresponding to the task classification information is a set of indicators predefined by the system for each level of task, including latency, jitter, reliability, etc.

[0136] The suppression strength of data transmission bandwidth is used to characterize the degree of restriction on the available bandwidth of the data synchronization channel. It can be achieved by reducing the token bucket rate, pausing transmission, reducing the sampling frequency, or compressing the data volume. The higher the suppression strength, the less network or computing resources the data synchronization channel occupies.

[0137] Among them, the smaller the deviation value, the closer the actual time delay is to the tolerance limit, and the stronger the suppression.

[0138] Flow control signals are instructions sent via shared memory flags, TSN Pause frames, or custom User Datagram Protocol (UDP) control packets to instruct the robot terminal to slow down or pause data uploads.

[0139] The safe recovery threshold is a more lenient recovery criterion than the tolerance limit, used to avoid frequent start-stop cycles;

[0140] The preset stabilization period is the time window required for the latency to remain below the recovery threshold, ensuring that the suppression is lifted only after the system has truly entered a stable state.

[0141] In actual execution, UPEN continuously monitors the end-to-end latency of the control channel during runtime. Before each instruction is written to the control block via the Precision Time Protocol (PTP) hardware clock, a high-precision timestamp is recorded. After the robot terminal completes the execution, it writes back the confirmation timestamp, which UPEN reads and calculates the closed-loop latency.

[0142] The joint arbitrator samples the latency at fixed intervals and compares it with the tolerance limit corresponding to the current task level. If the deviation is less than the preset safety margin, a risk is identified. At this time, the joint arbitrator calculates the suppression strength using a linear function based on the deviation to ensure that the control loop of highly critical tasks is not disturbed and reduces the token bucket rate of the data synchronization channel. Simultaneously, the backpressure feedback unit sets a 1-byte flow control flag in shared memory. The robot terminal's driver periodically polls this flow control flag. Once a pause signal is detected, it immediately stops writing new data to the data block or switches to a low-frequency sampling mode. When subsequent monitoring shows that the control channel latency is below the safety recovery threshold for 10 consecutive communication cycles, the joint arbitrator determines that the system has stabilized, clears the flow control flag, and gradually restores the token bucket rate to its original value, thus completing the removal of bandwidth suppression.

[0143] For example, the suppression strength based on the safety margin nonlinear mapping is:

[0144]

[0145] in, The bandwidth suppression strength applied to the data synchronization channel, with a value range of [value range missing]. A larger value indicates a stronger restriction; The system has a preset upper limit for the maximum suppression strength; It is a safety attenuation factor used to control the nonlinear steepness of the suppression intensity as the risk increases; It is the safety margin between the current performance of the control channel and the tolerance limit, defined as , To control the current measured end-to-end delay of the channel, This represents the tolerance limit for the k-th level task; As a preset safety margin threshold, when When this occurs, the inhibition mechanism is triggered.

[0146] The backpressure triggering condition for introducing service level deviation is:

[0147]

[0148] in, It is the Service Level Objective (SLO) corresponding to the k-th level task, i.e., the ideal expected latency; This is a preset safety margin ratio used to trigger back pressure in advance, preventing the delay from approaching the limit.

[0149] In this embodiment, when the control channel faces the risk of latency degradation, the bandwidth occupation of non-critical data streams is actively suppressed to prioritize the deterministic execution of highly critical tasks; the suppression intensity is dynamically adjusted according to the risk level to avoid data interruption caused by a one-size-fits-all approach; the back pressure mechanism ensures that the suppression command can quickly reach the source of the robot terminal to reduce the load from the generation end; and the recovery condition introduces a stable duration judgment to prevent frequent oscillations caused by instantaneous fluctuations.

[0150] In some embodiments, the near-end edge node is used for:

[0151] Online differential encoding is performed on the semantic feature vector, and change summary data is generated based on the temporal difference between the semantic feature vector and the historical semantic feature vector.

[0152] If the temporal difference exceeds the difference threshold, the change summary data is asynchronously uploaded to the regional edge node. The change summary data is compressed using a differential encoding format and includes the semantic dimension that has changed and the incremental value of the semantic dimension, which is used for internal status log recording and anomaly detection.

[0153] The region edge node is used to receive the change summary data in order to update the global perception data.

[0154] It is understandable that online differential coding is a compression method that calculates the difference between current features and historical features in real time while the data is being generated and only retains the changed parts;

[0155] Historical semantic feature vectors are snapshots of semantic features cached by UPEN in the previous cycle or the most recent valid state;

[0156] Temporal dissimilarity is a quantitative indicator that measures the degree of change between current and historical semantic feature vectors. It can be calculated using Manhattan distance or weighted Hamming distance, for example, by assigning higher weights to key dimensions such as risk level.

[0157] Change summary data is a compact data structure that contains only the semantic dimensions that have changed and the incremental values ​​of those semantic dimensions;

[0158] Semantic dimension is a structured attribute component in the semantic feature vector that has an independent physical or logical meaning, used to characterize the high-level state or environmental understanding of the robot system at a specific moment.

[0159] The difference threshold is a preset sensitivity threshold used to determine whether a change has semantic meaning;

[0160] Asynchronous upload sends data to the Region Edge Node (REN) immediately after the event is triggered in a non-blocking manner, without relying on a fixed period.

[0161] Differential encoding is a custom binary protocol that transmits only the identity document (ID), old value, new value, or increment, significantly reducing data volume.

[0162] Internal status logging involves UPEN writing a summary of changes to a local circular buffer for debugging or auditing purposes.

[0163] Anomaly detection triggers local alerts or security policies through mutation patterns.

[0164] In actual execution, at the end of each control cycle, UPEN uses its built-in AI model, optimized by the TensorRT inference framework, to infer the fused micro-sensor and scene data, outputting the current semantic feature vector. It then calls the kernel-state differential engine to compare this semantic feature vector with historical semantic feature vectors cached in Double Data Rate (DDR) memory dimension by dimension, calculating a weighted L1 difference. For example, the weight for the emergency stop signal dimension is set to 10, and the weight for the normal position dimension is 1. If the difference exceeds a preset threshold, it iterates through all dimensions, recording only fields whose changes exceed the minimum resolution, generating a compact change summary data. This change summary data is serialized into a custom binary format and asynchronously sent to REN via a low-priority TSN queue. Simultaneously, UPEN writes this change summary data to its local circular log buffer and sends it to a lightweight anomaly detection module based on 1D-CNN to determine whether a local safety action needs to be triggered. REN runs a listening service; upon receiving the change summary data, it parses the change dimensions and incremental values, directly updating its in-memory global awareness state table, replacing a full refresh without needing to re-aggregate all the original data.

[0165] For example, the temporal dissimilarity degree based on semantically perceived weighting is:

[0166]

[0167] in, The weighted temporal difference between the current and historical semantic feature vectors is used to determine whether to trigger the upload of the change summary. The semantic feature vector extracted at the current time t; For the previous valid time t 1. Historical semantic feature vector; This is an element-wise (Hadamard) product; For the state-dependent weight matrix, each element Adjust dynamically based on historical data.

[0168] Weight of the i-th semantic dimension for:

[0169]

[0170] in, The risk level dimension in the historical vector; .

[0171] In this embodiment, data is uploaded only when there are meaningful changes in the environment or task status, avoiding the bandwidth waste caused by traditional periodic full reporting; differential coding reduces the amount of data uploaded at one time; asynchronous transmission does not affect the local control closed loop; the global state of REN can be continuously and accurately updated, while UPEN can also use change summaries to achieve rapid response to local anomalies. While ensuring the timeliness of perception, it reduces the uplink load and extends the battery life of the mobile robot, enabling semantically driven sparse and efficient communication.

[0172] In some embodiments, the deployment of the master node and the shared backup node in the micro-edge cluster is determined based on joint optimization of radio frequency propagation characteristics, network hop count, and switching latency;

[0173] The radio frequency propagation characteristics, including path loss, multipath delay spread, and channel coherence time, are used to evaluate the lower limit of delay jitter in wireless links.

[0174] The network hop count is used to characterize the number of intermediate forwarding devices from the ultra-near-end edge node to the regional edge node;

[0175] The switching delay is used to characterize the transmission delay of data packets in the edge network path from the ultra-near-end edge node to the regional edge node due to queuing and processing by intermediate nodes in the network.

[0176] The regional edge node is used to construct a deployment cost function based on the radio frequency propagation characteristics, the network hop count, and the switching delay, in order to determine the optimal deployment coordinate set of the micro-edge cluster.

[0177] Understandably, the master node is the UPEN currently undertaking the main computing and control tasks, responsible for handling highly critical tasks. The end-to-end communication latency between the master node and the robot terminals under its jurisdiction meets the tolerance limit for highly critical tasks. The shared backup node is a UPEN in a low-power standby state, which can be reused by multiple micro-edge clusters and take over its tasks when the master node fails.

[0178] Radio frequency (RF) propagation characteristics are used to describe the physical behavior of wireless signals transmitted in space. Among the RF propagation characteristics, path loss characterizes the degree to which signal strength attenuates with distance, multipath delay spread characterizes the delay widening caused by multiple signal arrivals due to reflection and refraction, and channel coherence time characterizes the length of time the channel remains stable, reflecting how fast the channel changes time. These parameters together determine the lower limit of delay jitter that a wireless link can achieve, that is, the theoretically optimal delay stability.

[0179] Network hop count is the number of intermediate forwarding devices that data passes through when traveling from the UPEN to the Region Edge Node (REN). These intermediate forwarding devices can be switches, routers, or MEC access points. Each additional hop introduces additional latency.

[0180] Switching delay refers to the cumulative delay of data packets in each intermediate network device due to queuing, buffer processing, scheduling decisions, etc., and can be collected through in-band network telemetry (INT).

[0181] The deployment cost function is used to weight and fuse radio frequency propagation characteristics, network hop count, and switching delay into a single optimization objective. The objective function is to minimize the weighted sum of the network hop count and the switching delay. The optimization model is constrained by the radio frequency propagation characteristics not being lower than a preset communication quality threshold. The optimization algorithm finds the combination of node deployment locations that minimizes the deployment cost function value. The specific installation coordinates of the primary node and the shared backup node in physical space are finally determined as the optimal deployment coordinate set.

[0182] In actual implementation, REN obtains candidate deployment points through digital twin models of the plant area or on-site surveys. It uses WinProp RF simulation tools or measured data from Wi-Fi / 5G channel probing to calculate path loss, multipath delay spread, and channel coherence time between each candidate point. Simultaneously, it determines the path from any UPEN to REN and the corresponding hop count based on the network topology map, and actively probes and measures the end-to-end switching delay of each path by sending timestamp probe packets. REN substitutes the RF propagation characteristics, network hop count, and switching delay into a preset deployment cost function, and uses an integer linear programming optimization algorithm to traverse all feasible primary / backup node combinations to calculate the total cost of each combination. Finally, it selects the combination with the lowest cost, uses its corresponding physical coordinates as the deployment scheme for the micro-edge cluster, and issues configuration commands to the relevant UPEN to complete role assignment and network binding.

[0183] For example, the deployment cost function for a deployment task-aware type is:

[0184]

[0185] in, For micro-edge clusters in deployment locations The lower the overall deployment cost, the better the deployment plan. Includes the coordinates of each UPEN; Used to characterize the lower limit of wireless jitter; Used to characterize end-to-end deterministic latency; A hierarchical set of tasks; Let the traffic weight of the k-th level task in this region satisfy the following condition: This information was obtained by REN through statistical analysis of historical task distributions. These are normalized weighting coefficients that reflect the relative importance of wireless link quality and network transmission delay in the total cost, satisfying... ; For deployment location set Below, the wireless path loss required to support the k-th level task is obtained by ray tracing or actual measurement. The larger the value, the more severe the signal attenuation. For reference path loss, such as free space path loss, it is used to... Normalize to eliminate the influence of dimensions; For deployment location set Below, the number of network hops that the data stream corresponding to the k-th level task traverses from UPEN to REN, i.e., the number of intermediate forwarding devices; The single-hop average switching latency is the average delay introduced by queuing and processing of data packets in a single switch / router, which is obtained through actual measurement using a network probe. For deployment location set Below, the standard deviation of end-to-end delay jitter corresponding to the k-th level task is used to reflect the delay uncertainty caused by wireless multipath and network congestion, and is a key indicator for measuring control stability.

[0186] In this embodiment, by considering the computational load, the physical characteristics of the wireless channel and the transmission performance of the wired network are incorporated into a unified evaluation framework to ensure that the uplink between the master node and the REN has sufficiently low latency jitter and high reliability. In hybrid wired-wireless deployment scenarios, control command delays caused by wireless multipath or switching congestion are avoided, the success rate of fault switching is improved, the service quality assurance capability of the edge cluster in complex industrial environments is enhanced, and the micro-edge cluster deployment is transformed from experience-based site selection to performance-driven deployment.

[0187] In some embodiments, the regional edge node is used to parse the assistance intent information of each robot terminal in the robot queue. When multiple robot terminals are detected to have collaborative operation needs, the ultra-near-end edge node involving the robot queue and its corresponding shared backup node are grouped into a virtual joint node. The assistance intent information includes task dependencies, spatial collaboration needs and peripheral sharing requests.

[0188] All ultra-near-end edge nodes within the virtual federated node share resource views and scheduling policies, and the regional edge nodes coordinate task allocation and load balancing.

[0189] Understandably, a robot queue is a list of tasks to be executed in the system, ordered by priority, containing requests from multiple robot terminals.

[0190] Assistance intent information is collaborative request data proactively reported by the robot terminal, including task dependencies, spatial collaboration needs, and peripheral sharing requests.

[0191] Virtual federated nodes are logically grouped sets of UPENs by REN to support collaborative tasks.

[0192] The resource view includes real-time resource status of UPEN, such as CPU utilization, memory usage, and network bandwidth.

[0193] Scheduling policies are the rules for task allocation, which include parameters such as priority and load threshold.

[0194] Collaborative task allocation involves splitting and distributing tasks from multiple robots to different UPENs; load balancing dynamically adjusts the workload of each UPEN to avoid overload.

[0195] In actual execution, REN periodically collects assistance intent information from each robot terminal in the robot queue, receives JSON format data via the gRPC protocol, and parses the task dependencies, spatial collaboration requirements, and peripheral sharing requests. REN runs a collaboration detection module based on the Drools rule engine. When it detects that multiple robots have common needs, such as two robots simultaneously requesting spatial collaboration and having peripheral sharing conflicts, REN configures the involved master nodes and their corresponding shared backup nodes through logical groups in memory, and groups them into virtual federated nodes by creating virtual node IDs. REN aggregates all UPENs within this virtual federated node to generate a global view, and distributes a unified resource view and scheduling strategy, such as a weighted allocation algorithm based on task urgency. REN uses a dynamic weighted round-robin load balancing algorithm based on the current load to split collaborative tasks into subtasks, allocate them to UPENs within the virtual federated node, and monitors load changes in real time. When the load of a certain UPEN exceeds a threshold, it automatically adjusts the allocation strategy and migrates some tasks to backup nodes.

[0196] In addition, when task dependency conflicts or spatial constraint overlaps are detected, the conflict resolution module can be activated to sort the DAG topology and calculate a secure cooperative path. The involved ultra-near-end edge nodes and their shared backup nodes are grouped into virtual federated nodes through a hash table structure to share a real-time resource view, which includes CPU utilization, memory usage and network status.

[0197] The regional edge nodes dynamically allocate collaborative tasks based on real-time resource views and trigger a task migration mechanism when resources are overloaded to ensure the continuity of task execution.

[0198] For example, collaborative sensing load metrics:

[0199]

[0200] in, This is the collaborative perception comprehensive load index for the j-th UPEN, used for task scheduling within the virtual federated node; Let be the CPU utilization of the j-th UPEN; Let be the memory usage of the j-th UPEN; Used for collaborative distance penalty Let j be the set of robot terminals that collaborate based on the assistance intention information of the j-th UPEN. Let be the Euclidean distance between the j-th UPEN and the i-th robot terminal, used to reflect the physical cooperation tightness; These are weighting coefficients for resources, memory, and collaboration distance, respectively, satisfying... .

[0201] In this embodiment, the originally scattered UPEN resources are logically integrated through virtual joint nodes, reducing the latency of collaborative task switching, improving resource utilization and execution efficiency, and avoiding control interruptions caused by single-point overload through dynamic load balancing; real-time parsing of peripheral sharing requests can support complex collaborative scenarios, realizing efficient and adaptive scheduling of multi-robot collaborative tasks.

[0202] In some embodiments, the task classification information includes high-criticality tasks, medium-criticality tasks, and low-criticality tasks.

[0203] The operating system kernel layer of the ultra-near-end edge node is configured to construct mutually isolated hardware resource slices based on task classification information;

[0204] The reflection control task in a highly critical task is allocated a dedicated CPU core, a contiguous physical memory region, and a last-level cache partition;

[0205] Slices of limited computing resources are allocated to semantic feature extraction or data aggregation tasks in medium-critical and low-critical tasks.

[0206] It is understandable that the operating system kernel layer is the underlying module of the real-time kernel such as Linux PREEMPT_RT or Xenomai in ultra-near-end edge nodes.

[0207] Hardware resource slicing is a computing resource partitioning achieved through physical isolation. Resource slices are statically reserved at startup and cross-slice memory access and interrupt preemption are prohibited during operation to prevent control channel latency jitter caused by resource contention.

[0208] An exclusive CPU core is a dedicated processor core allocated to a high-critical task, which avoids sharing through CPU affinity binding;

[0209] A contiguous physical memory region refers to a block of memory allocated consecutively in DRAM, which is achieved through mmap in conjunction with HugeTLB page tables.

[0210] The final cache partition is a dedicated area of ​​the CPU's L3 cache or similar high-speed cache, configured through Intel Cache Allocation Technology (CAT) or ARM Memory Partitioning and Monitoring (MPAM).

[0211] Reflective control tasks are real-time control operations in highly critical tasks, such as closed-loop feedback.

[0212] A restricted compute resource slice is a pool of resources shared by medium- or low-critical tasks, with CPU utilization and memory bandwidth limited by control groups (CGroups).

[0213] In actual execution, upon UPEN startup, it receives task classification information from the region edge nodes, and the operating system kernel loads a customized resource slicing configuration module. For highly critical tasks, the kernel calls CPU affinity APIs, such as `sched_setaffinity`, to bind a specified core to the task. Simultaneously, it allocates contiguous memory blocks via `mmap`, utilizes HugeTLB page tables to avoid page table flips, and leverages Intel CAT's Allocation Technology to set cache partitions. For medium-critical tasks such as semantic feature extraction, the kernel creates resource control groups using CGroup v2 to limit CPU utilization and memory bandwidth. For low-critical tasks such as data aggregation, it allocates a shared computing resource pool and controls bandwidth through a token bucket mechanism. During task execution, the kernel monitors resource usage in real time; if a highly critical task detects cache contention, it automatically adjusts the cache partitions. The entire configuration is loaded at startup via kernel modules such as `kmod-cpu-alloc`, requiring no application-layer intervention.

[0214] In this embodiment, by dedicating CPU cores, contiguous memory, and cache partitions to highly critical tasks, jitter caused by shared resource contention is eliminated, achieving hardware-level resource isolation, reducing end-to-end latency jitter of control commands, and improving the safety level; medium and low critical tasks run in restricted slices to avoid interfering with highly critical tasks; this improves control reliability and resource utilization in industrial robot scenarios, providing physical guarantees for high safety and high real-time control.

[0215] In some embodiments, the ultra-near-end edge node is further used to extract a semantic feature vector through progressive semantic feature encoding;

[0216] The progressive semantic feature encoding generates a hierarchical data structure that includes at least a first encoding layer and a second encoding layer;

[0217] The first encoding layer is a critical status code, including a limited number of status identifiers, which are used for transmission within a first time window to characterize the security level of the current operation and the approximate state of the system.

[0218] The second encoding layer is a complete semantic feature description, used for transmission within the second time window, including the complete dimension and precision information of the semantic feature vector;

[0219] The regional edge node sends a feature request instruction to the ultra-near-end edge node according to the current network status and the urgency level corresponding to the task classification information, so as to receive the first coding layer, or the first coding layer and the second coding layer.

[0220] Understandably, progressive semantic feature encoding is used for hierarchical data compression, splitting the semantic feature vector into two layers: key states and a complete description. The first encoding layer is the key state code, which contains the security level identifier and the system overview state, and is used for transmission within a very short time window. For example, in the security level, 0 = safe, 1 = warning, and 2 = emergency. The system overview state can be running / paused. The second encoding layer is the complete semantic feature description, which contains all dimensions and precision information of the semantic feature vector.

[0221] The Feature Request command is a control command sent by REN that specifies the layer encoding corresponding to the UPEN transmission.

[0222] In the first encoding layer, a fixed 2-byte key status code is used to encode the security level and system status through a bitmask; the second encoding layer is a 128-dimensional floating-point vector.

[0223] In actual execution, UPEN runs a lightweight MobileNetV3 optimized with TensorRT to perform real-time inference on micro-sensor data and scene state data within each control cycle, outputting a 128-dimensional semantic feature vector. The progressive semantic feature encoding module divides this vector into a first encoding layer and a second encoding layer. UPEN uses a time-gated scheduling mechanism to write the first encoding layer into the shared memory control block within the exclusive time slot of the control channel. In subsequent periodic time slots, the second encoding layer is written into the data block. The regional edge nodes monitor network status and task classification information such as bandwidth utilization and end-to-end latency in real time. Due to the priority response required for highly critical tasks, when network congestion or urgent tasks are detected, a feature request instruction is sent to UPEN, requesting the transmission of only the first encoding layer. When the network is idle, the instruction requests the transmission of both the first and second encoding layers simultaneously. Upon receiving the instruction, UPEN immediately executes the corresponding transmission operation within the specified time slot without requiring additional data copying.

[0224] For example, the feature layer request probability is:

[0225]

[0226] in, The probability of requesting the second coding layer for a region edge node; It is the Sigmoid activation function. , used to map the input to the (0, 1) interval; The urgency level of the current task is determined by task hierarchy mapping; The current network normalized load; This is an urgency sensitivity coefficient used for amplification. The impact; This is the network load suppression coefficient, reflecting the constraint of bandwidth on upload decisions.

[0227] In this embodiment, by prioritizing the transmission of key status codes, the decision-making response time of highly critical tasks is shortened. The second encoding layer is only uploaded when bandwidth is sufficient, avoiding redundant data occupation and reducing uplink traffic. The regional edge nodes dynamically adjust their request strategies according to the network to ensure that key information is delivered first, while maintaining the overall bandwidth efficiency of the system. This reduces the delay in task decision-making in industrial robot control scenarios and achieves adaptive transmission of semantic information.

[0228] This application also provides a real-time interaction method between a robot and a peripheral device based on edge computing, which is applied to the real-time interaction system between a robot and a peripheral device based on edge computing as described in any of the above embodiments.

[0229] like Figure 2 As shown, this edge computing-based real-time interaction method between the robot and peripherals includes:

[0230] Step 210: Through the regional edge nodes, determine the task classification information based on the semantic feature vector of the global perception data, make global decisions, and schedule the shared backup nodes in the regional redundant resource pool. The global perception data includes scene status data and micro-sensor data.

[0231] Step 220: Collect the scene state data through ultra-near-end edge nodes, and extract the semantic feature vector from the micro-sensing data and the scene state data;

[0232] Step 230: Collect the microsensor data through the robot terminal, and execute the task based on the task classification information.

[0233] The real-time interaction method between robots and peripherals based on edge computing provided in this application avoids uncertain latency caused by traditional network protocol stacks by deeply coupling ultra-near-end edge nodes and robot terminals through shared memory zero-copy and time-gated scheduling, ensuring deterministic execution of highly critical tasks. Regional edge nodes perform task classification and global scheduling based on semantic feature vectors, and utilize a cross-cluster shared backup node pool to achieve elastic disaster recovery, which avoids resource redundancy and waste, ensures high system availability, enables the system to adapt to environmental changes, reduces interaction latency in complex wireless environments, ensures deterministic latency of highly critical tasks, improves the overall system resource utilization efficiency, and enhances the robot's real-time perception and control capabilities of peripherals.

[0234] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0235] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0236] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A real-time interaction system between a robot and peripherals based on edge computing, characterized in that, It includes at least one regional edge node, multiple ultra-near-end edge nodes, and multiple robot terminals. Each regional edge node is connected to a micro-edge cluster. The micro-edge cluster includes a master node and at least one shared backup node. The master node and the shared backup node are divided based on task classification information. The ultra-near-end edge node is connected to the robot terminal based on the coupling of shared memory and time-gated scheduling; The regional edge nodes are deployed based on the regional division of the robot terminal. They are used to determine the task classification information based on the semantic feature vector of the global perception data, make global decisions, and schedule the shared backup nodes in the regional redundant resource pool. The regional division is determined based on task throughput, data aggregation requirements, and energy consumption constraints. The global perception data includes scene status data and micro-sensor data. The ultra-near-end edge node is deployed based on task latency constraints and communication reliability requirements, and is used to collect the scene state data and extract the semantic feature vector from the micro-sensor data and the scene state data; The robot terminal is used to collect the microsensor data and perform task execution based on the task classification information.

2. The real-time interaction system between a robot and peripherals based on edge computing according to claim 1, characterized in that, A control channel and a data synchronization channel are configured between the robot terminal and the ultra-near-end edge node; The ultra-near-end edge node is configured with a gated shared memory manager, which is used to open access permissions to the mapped logical blocks within the activated time slot according to the time slot trigger signal generated by the time gating scheduling mechanism, so that the authorized robot terminal or the ultra-near-end edge node can perform zero-copy read and write to the corresponding logical blocks in the shared memory area. The time-gating scheduling mechanism is used to divide the communication cycle into exclusive time slots, periodic time slots, and dynamic time slots according to the time sequence, allocate exclusive time slots to the control channel, and allocate corresponding periodic time slots or dynamic time slots to the data synchronization channel. The priority of the exclusive time slots is higher than that of the periodic time slots and dynamic time slots. The shared memory region is divided into multiple logical blocks, each logical block including a control block and at least one data block; The control block is used to map the control channel so that the ultra-near-end edge node can write the task classification information and control instructions through kernel-mode memory mapping, and the robot terminal can read them. The data block is used to map the data synchronization channel so that the robot terminal can write the microsensor data and the ultra-near-end edge node can read it.

3. The real-time interaction system between a robot and peripherals based on edge computing according to claim 2, characterized in that, The ultra-near-end edge node is used to adjust the period and duty cycle of the time gate of the control channel and the data synchronization channel according to the task classification information issued by the regional edge node. Within the exclusive time slot of the control channel, it issues control instructions containing task classification information to the robot terminal through the control block. It allocates a data synchronization channel with a fixed position and redundant period for data streams with high task classification information, and the mapped data blocks are contiguous in physical memory. The robot terminal is used to receive and respond to the control commands, and according to the task classification information, within the authorized and activated time slots, within the mapped data blocks, synchronize the microsensor data through zero-copy. The exclusive time slot of the control channel is located at the beginning of each communication cycle, and the communication cycle is synchronized with the control decision cycle of the robot terminal.

4. The real-time interaction system between a robot and peripherals based on edge computing according to claim 2, characterized in that, The ultra-near-end edge node integrates a joint arbitrator; The joint arbitrator is used to adjust the suppression strength of the data transmission bandwidth of the data synchronization channel according to the deviation value between the performance index of the control channel and the service level target corresponding to the task classification information when the deviation between the end-to-end delay of the control channel and the tolerance upper limit of the task classification information is less than a preset safety margin. The deviation value is negatively correlated with the suppression strength. The joint arbitrator is equipped with a back pressure feedback unit, which is used to send a flow control signal to the robot terminal to pause or reduce the rate at which the robot terminal writes micro-sensor data to the data block. The ultra-near-end edge node is also used to release the bandwidth suppression of the data synchronization channel and restore the transmission rate of the data synchronization channel after the end-to-end delay of the control channel has been continuously lower than the safety recovery threshold for a preset stable period of time.

5. The real-time interaction system between a robot and peripherals based on edge computing according to claim 1, characterized in that, The ultra-near-end edge node is used for: Online differential encoding is performed on the semantic feature vector, and change summary data is generated based on the temporal difference between the semantic feature vector and the historical semantic feature vector. If the temporal difference exceeds the difference threshold, the change summary data is asynchronously uploaded to the regional edge node. The change summary data is compressed using a differential encoding format and includes the semantic dimension that has changed and the incremental value of the semantic dimension, which is used for internal status log recording and anomaly detection. The region edge node is used to receive the change summary data in order to update the global perception data.

6. The real-time interaction system between a robot and peripherals based on edge computing according to claim 1, characterized in that, The deployment of the master node and the shared backup node in the micro-edge cluster is determined based on joint optimization of radio frequency propagation characteristics, network hop count, and switching latency; The radio frequency propagation characteristics, including path loss, multipath delay spread, and channel coherence time, are used to evaluate the lower limit of delay jitter in wireless links. The network hop count is used to characterize the number of intermediate forwarding devices from the ultra-near-end edge node to the regional edge node; The switching delay is used to characterize the transmission delay of data packets in the edge network path from the ultra-near-end edge node to the regional edge node due to queuing and processing by intermediate nodes in the network. The regional edge node is used to construct a deployment cost function based on the radio frequency propagation characteristics, the network hop count, and the switching delay, in order to determine the optimal deployment coordinate set of the micro-edge cluster.

7. The real-time interaction system between a robot and peripherals based on edge computing according to claim 1, characterized in that, The region edge node is used to parse the assistance intention information of each robot terminal in the robot queue. When multiple robot terminals are detected to have collaborative operation needs, the ultra-near-end edge node involving the robot queue and its corresponding shared backup node are grouped into a virtual joint node. The assistance intention information includes task dependency relationship, spatial collaboration needs and peripheral sharing requests. All ultra-near-end edge nodes within the virtual federated node share resource views and scheduling policies, and the regional edge nodes coordinate task allocation and load balancing.

8. The real-time interaction system between a robot and peripherals based on edge computing according to claim 1, characterized in that, The task classification information includes high-criticality tasks, medium-criticality tasks, and low-criticality tasks. The operating system kernel layer of the ultra-near-end edge node is configured to construct mutually isolated hardware resource slices based on task classification information; The reflection control task in a highly critical task is allocated a dedicated CPU core, a contiguous physical memory region, and a last-level cache partition; Slices of limited computing resources are allocated to semantic feature extraction or data aggregation tasks in medium-critical and low-critical tasks.

9. The real-time interaction system between a robot and peripherals based on edge computing according to claim 1, characterized in that, The ultra-near-end edge node is further used to extract semantic feature vectors through progressive semantic feature encoding; The progressive semantic feature encoding generates a hierarchical data structure that includes at least a first encoding layer and a second encoding layer; The first encoding layer is a critical status code, including a limited number of status identifiers, which are used for transmission within a first time window to characterize the security level of the current operation and the approximate state of the system. The second encoding layer is a complete semantic feature description, used for transmission within the second time window, including the complete dimension and precision information of the semantic feature vector; The regional edge node sends a feature request instruction to the ultra-near-end edge node according to the current network status and the urgency level corresponding to the task classification information, so as to receive the first coding layer, or the first coding layer and the second coding layer.

10. A real-time interaction method between a robot and peripheral devices based on edge computing, characterized in that, The method, applied to a real-time interaction system between a robot and a peripheral device based on edge computing as described in any one of claims 1-9, comprises: By using regional edge nodes, task classification information is determined based on the semantic feature vector of the global perception data, global decisions are made, and the shared backup nodes in the regional redundant resource pool are scheduled. The global perception data includes scene status data and micro-sensor data. The scene state data is collected through ultra-near-end edge nodes, and the semantic feature vector is extracted from the micro-sensing data and the scene state data; The robot terminal collects the microsensor data and performs task execution based on the task classification information.

Images