A data migration method and system based on embodied intelligence hybrid cache, a terminal and a storage medium

Abstract

This invention discloses a data migration method, system, terminal, and storage medium based on embodied intelligent hybrid caching. The method includes: acquiring data access characteristics and type identifiers; performing multi-dimensional identification on the data based on the access characteristics and type identifiers to obtain access temperature values ​​and key data lock status; acquiring the current task stage status of the intelligent system; dynamically adjusting the partitioning strategy based on the key data lock status and access temperature values ​​according to the task stage status to obtain an allocation strategy; acquiring capacitor power information; migrating the data according to the capacitor power information and the allocation strategy to obtain a migration scheduling result. This invention acquires temperature and lock status based on data access characteristics and type identifiers, dynamically adjusts the partitioning strategy based on task stages, migrates data based on capacitor power, and then controls the robot to dynamically adjust the storage strategy according to the task stage to obtain accurate scheduling results.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a data migration method, system, terminal, and computer-readable storage medium based on embodied intelligence hybrid caching. Background Technology

[0002] Currently, embodied AI is a technology that deeply integrates artificial intelligence with physical entities such as robots, enabling intelligent agents to perceive complex environments, understand task intentions, and execute precise physical operations through multimodal sensors. In embodied AI systems, the real-time control loop needs to process data streams from multiple sensors, including vision, touch, force, and IMU, simultaneously, placing extremely stringent requirements on the storage subsystem: millisecond-level response latency to ensure real-time control stability, low-power design to extend the mobile robot's endurance, and a reliable power-loss protection mechanism to ensure that critical states during task execution are not lost due to unexpected power outages.

[0003] Current storage technologies face a dilemma. SRAM (Static Random-Access Memory) boasts nanosecond-level read / write speeds, low dynamic power consumption, and good compatibility with standard logic processes. However, its 6-transistor cell structure results in low storage density and high cost. Furthermore, as a volatile memory, data is immediately lost upon power failure, failing to meet the persistent state requirements of embodied intelligent systems. Using a pure SRAM solution to build a large-capacity cache for storing multimodal sensor data would lead to excessively large chip area and high cost, severely hindering the commercial viability of the system. PRAM (Phase-Change Memory), as a novel non-volatile memory, offers 4-8 times the storage density of SRAM, byte-level addressing capability, and lower cost per bit. Its non-volatile nature effectively solves the data protection problem during power failure. However, PRAM's write latency is more than 10 times slower than read latency (100-500ns vs 20-50ns), write power consumption is high, and it suffers from 10... 8 The system has a durability limit of -10¹² cycles. If PRAM is used as the cache throughout, the frequent write operations in real-time control scenarios will not meet the millisecond-level response requirements, and rapid wear will significantly shorten the system's lifespan.

[0004] Existing hybrid caching architectures mainly include data migration methods based on access frequency statistics and static partition management methods based on general-purpose processors. These traditional methods are limited by rigid strategies due to a lack of task awareness, and "open-loop" control defects such as simple migration decisions and imperfect power-off protection. They cannot effectively support the comprehensive requirements of real-time control, large-capacity storage, and power-off persistence in embodied intelligence scenarios. As a result, using existing methods cannot solve the problems of low-latency guarantee of critical control data and PRAM wear leveling. Furthermore, it is difficult to achieve dynamic allocation of storage resources according to task stages, power consumption adaptive migration combined with battery power, and efficient differential power-off preservation. Consequently, they cannot meet the hybrid caching architecture requirements of millisecond-level response, long battery life, and high reliability of embodied intelligence systems in robots. Therefore, existing data migration technologies based on embodied intelligence hybrid caching still need to be improved and optimized. Summary of the Invention

[0005] The main objective of this invention is to provide a data migration method, system, terminal, and computer-readable storage medium based on embodied intelligent hybrid caching. This invention aims to address the problems that existing hybrid caching architectures cannot guarantee low latency for critical control data and PRAM wear leveling, and are difficult to dynamically allocate storage resources according to the robot's task stage and adapt to power consumption based on battery level, thus preventing the robot from dynamically adjusting its storage strategy according to the task stage.

[0006] To achieve the above objectives, the present invention provides a data migration method based on embodied intelligent hybrid caching, the data migration method based on embodied intelligent hybrid caching comprising the following steps: The access characteristics and type identifiers of the data are obtained, and the data is identified in multiple dimensions based on the access characteristics and type identifiers to obtain the access temperature value and the key data lock status. Obtain the current task stage status of the intelligent system, and dynamically adjust the key data locking status and the access temperature value according to the task stage status to obtain the allocation strategy. Obtain capacitor power information, migrate the data according to the capacitor power information and the allocation strategy, and obtain migration scheduling results.

[0007] Optionally, in the data migration method based on embodied intelligent hybrid caching, the access characteristics include access frequency, access urgency, data correlation, and migration cost; The process of acquiring access features and type identifiers for data, and performing multi-dimensional identification based on these features and identifiers to obtain access temperature and lock status, specifically includes: The access frequency, access urgency, data relevance, and migration cost are obtained. The access temperature value is obtained by weighting the access frequency, access urgency, data relevance, and migration cost. When the type is identified as emergency control data, the data is locked based on the emergency control data to obtain a critical data lock status.

[0008] Optionally, the data migration method based on embodied intelligent hybrid caching, wherein obtaining access frequency, access urgency, data relevance, and migration cost, and performing a weighted calculation on the access frequency, access urgency, data relevance, and migration cost to obtain an access temperature value, further includes: Determine whether the access temperature value is greater than a preset thermal data threshold; If the access temperature value is greater than the preset hot data threshold, a migration queue is obtained, and the data is asynchronously migrated according to the priority of the migration queue to obtain migrated data.

[0009] Optionally, in the data migration method based on embodied intelligent hybrid caching, the task stage states include a perception stage, a planning stage, a control stage, and a learning stage. The process of obtaining the current task stage status of the intelligent system, and dynamically adjusting the key data locking status and the access temperature value based on the task stage status to obtain an allocation strategy, specifically includes: The current intelligent system's perception, planning, control, and learning phases are obtained. Based on these phases, the key data locking status and access temperature value are dynamically adjusted using a partitioning strategy to obtain multiple cache weights. Obtain the partitioning strategy table, and reallocate the partitioning strategy table according to all the cache weights to obtain the allocation strategy.

[0010] Optionally, in the data migration method based on embodied intelligent hybrid caching, the caching weights include visual frame caching weights, map caching weights, joint state caching weights, and gradient caching weights. The process involves acquiring the current intelligent system's perception, planning, control, and learning phases, and then dynamically adjusting the key data locking status and access temperature value based on these phases to obtain multiple cache weights. Specifically, this includes: The current intelligent system's perception stage, planning stage, control stage, and learning stage are obtained. Based on the perception conditions of the perception stage and the partition ratio of the partition strategy table, the key data locking status and the access temperature value are judged to obtain the visual frame cache weight. Based on the planning conditions of the planning phase and the partition ratio of the partitioning strategy table, the data locking status and the access temperature value are judged during the planning phase to obtain the map cache weight; Based on the control conditions of the control phase and the partition ratio of the partition strategy table, the control phase judgment is performed on the data locking state and the access temperature value to obtain the joint state cache weight. Based on the learning conditions of the learning phase and the partition ratio of the partitioning strategy table, the data locking status and the access temperature value are judged to obtain the gradient cache weight.

[0011] Optionally, the data migration method based on embodied intelligent hybrid caching, wherein obtaining capacitor power information and migrating the data according to the capacitor power information and the allocation strategy to obtain a migration scheduling result specifically includes: Acquire capacitor power information; if a power failure signal is detected or the capacitor power information is lower than a first preset power level, filter the data according to the type identifier to obtain first data; based on the differential write strategy, write the changed bytes of the first data to the recovery area according to the priority. Acquire safety posture information, current task status, and recovery flag. If the capacitor power information is lower than a second preset power level, then write the safety posture information, current task status, and recovery flag into the data to obtain the second data. Based on the capacitor charge information and the allocation strategy, the first data and the second data are migrated to obtain the migration scheduling result.

[0012] Optionally, the data migration method based on embodied intelligent hybrid caching, wherein the step of migrating the first data and the second data according to the capacitor power information and the allocation strategy to obtain a migration scheduling result specifically includes: A migration strategy is determined based on the capacitor charge information. If the migration frequency of the migration strategy is normal, the first data and the second data are migrated according to the first migration threshold of the migration strategy to obtain a first migration scheduling result. If the migration frequency is low, then the first data and the second data are migrated according to the allocation strategy and the second migration threshold of the migration strategy to obtain the second migration scheduling result; If the migration frequency is low, then the first data and the second data are protected and migrated according to the allocation strategy and the migration strategy to obtain the third migration scheduling result.

[0013] Furthermore, to achieve the above objectives, the present invention also provides a data migration system based on embodied intelligent hybrid caching, wherein the data migration system based on embodied intelligent hybrid caching: A multi-dimensional data identification module is used to obtain the access characteristics and type identifiers of data, and to perform multi-dimensional identification of data based on the access characteristics and type identifiers to obtain the access temperature value and key data lock status. The data allocation module is used to obtain the current task stage status of the intelligent system, and dynamically adjust the key data locking status and the access temperature value according to the task stage status to obtain the allocation strategy. The data migration module is used to acquire capacitor power information, migrate the data according to the capacitor power information and the allocation strategy, and obtain migration scheduling results.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a data migration program based on an embodied intelligent hybrid cache, and the data migration program based on an embodied intelligent hybrid cache, when executed by a processor, implements the steps of the data migration method based on an embodied intelligent hybrid cache as described above.

[0015] In this invention, access characteristics and type identifiers of data are acquired, and the data is multi-dimensionally identified based on these characteristics to obtain access temperature values ​​and key data lock status. The current task stage status of the intelligent system is acquired, and a dynamic partitioning strategy is adjusted based on the task stage status to obtain an allocation strategy. Capacitor charge information is acquired, and the data is migrated based on the capacitor charge information and the allocation strategy to obtain a migration scheduling result. This invention obtains temperature and lock status based on data access characteristics and type identifiers, dynamically adjusts the partitioning strategy according to the task stage, migrates data based on capacitor charge, and then the robot dynamically adjusts the storage strategy according to the task stage to obtain accurate scheduling results. Attached Figure Description

[0016] Figure 1 This is a flowchart of a preferred embodiment of the data migration method based on embodied intelligent hybrid caching of the present invention; Figure 2 This is a flowchart illustrating the overall architecture of a preferred embodiment of the data migration method based on embodied intelligent hybrid caching of the present invention; Figure 3This is a flowchart of an SRAM storage array, representing a preferred embodiment of the data migration method based on an embodied intelligent hybrid cache according to the present invention. Figure 4 This is a flowchart of a PRAM storage array, representing a preferred embodiment of the data migration method based on embodied intelligent hybrid caching according to the present invention. Figure 5 This is a structural diagram of a preferred embodiment of the data migration system based on embodied intelligent hybrid caching of the present invention; Figure 6 This is a structural diagram of a preferred embodiment of the terminal of the device of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] Existing hybrid caching architecture methods mainly include data migration methods based on access frequency statistics and static partition management methods based on general-purpose processors. These traditional methods are limited by rigid strategies due to a lack of task awareness, and "open-loop" control defects such as simple migration decisions and imperfect power-off protection. They cannot effectively support the comprehensive requirements of real-time control, large-capacity storage, and power-off persistence in embodied intelligence scenarios. As a result, existing methods cannot solve the problems of low-latency guarantee of critical control data and PRAM wear leveling. Furthermore, they are difficult to achieve dynamic allocation of storage resources according to task stages, power consumption adaptive migration combined with battery power, and efficient differential power-off preservation. Consequently, they cannot meet the hybrid caching architecture requirements of millisecond-level response, long battery life, and high reliability of embodied intelligence robot systems. Therefore, a data migration method based on embodied intelligence hybrid caching is needed. This method obtains temperature and lock status based on data access characteristics and type identifiers, dynamically adjusts partitioning strategies according to task stages, migrates data based on capacitor power, and then allows the robot to dynamically adjust storage strategies according to task stages to obtain accurate scheduling results.

[0019] The data migration method based on embodied intelligent hybrid caching described in the preferred embodiment of the present invention, such as... Figure 1 and Figure 2 As shown, the data migration method based on embodied intelligent hybrid caching includes the following steps: Step S10: Obtain the access characteristics and type identifier of the data, and perform multi-dimensional identification of the data based on the access characteristics and type identifier to obtain the access temperature value and key data lock status.

[0020] Step S10 includes: Step S11: Obtain access frequency, access urgency, data correlation and migration cost, and perform a weighted calculation on the access frequency, access urgency, data correlation and migration cost to obtain the access temperature value; Step S12: When the type is identified as emergency control data, the data is locked according to the emergency control data to obtain the key data lock status.

[0021] Specifically, after step S10, the method further includes: determining whether the access temperature value is greater than a preset thermal data threshold (if in PRAM (phase change memory), data is read from PRAM, the access temperature value is calculated (T is output by the temperature calculation unit), and it is determined whether T is greater than T_hot (thermal data threshold)); if the access temperature value is greater than the preset thermal data threshold, a migration queue is obtained, and the data is asynchronously migrated according to the priority of the migration queue to obtain migrated data (if T>T_hot: the data is added to the migration queue (PRAM to SRAM (static random access memory) direction), the migration priority is set to T multiplied by 15 and rounded down, the data is first returned to the processor (delayed access), and the migration is performed asynchronously in the background); and obtaining the access frequency (recent access count, weight). Access urgency (real-time requirements (control data > sensing data)) and weighting ), data correlation (access correlation with other hot data, weight) ) and migration costs (migration power consumption and retention power consumption, weighted) The access frequency, access urgency, data correlation and migration cost are weighted and calculated to obtain the access temperature value. When the type is identified as emergency control data, the data is locked according to the emergency control data to obtain the key data lock status.

[0022] As an example, the robot initialization process involves powering on the system, resetting the migration controller, and initializing all registers to their default values. The PRAM Bank7 recovery flag area is read to check for the existence of a valid recovery flag. If the recovery flag (a specific data structure stored at PRAM Bank7 address 0x0000, used to determine whether the system has experienced a valid state saving after an abnormal power outage) is valid, the system enters recovery mode; otherwise, it enters normal startup mode. (Recovery Mode): Read the task context from PRAMBank3; read the safe posture and system configuration from PRAMBank7; write the critical state data (critical state data refers to the minimum necessary set of information that the robot must save at the last moment before power failure for seamless recovery execution. Specifically, it includes: task context: the currently executing task (such as "picking up goods"), task stage (control stage / perception stage), target coordinates (x, y, z), etc. Safe posture: the angle values ​​of all robot joints and the chassis odometry pose) to the corresponding position in SRAM; set the type flag of the corresponding cache_line in SRAM (a cache_line that stores joint state data, whose type flag binary is 11 11 1 1 00 (i.e. 0xFC). This indicates: emergency control data (11) + control stage association (11) + lock (1) + need to persist (1)) as emergency control data; set lock_bit to 1 to disable migration; clear the recovery flag to prevent repeated recovery; jump to the initialization task perception module and set the task stage counter threshold to the default value. (Normal Startup Mode): Initialize all SRAM banks to empty; set valid_bit to 0 and dirty_bit to 0; initialize access counters to 0; load the default partitioning strategy (awareness phase strategy). Initialize the task awareness module and set the task phase counter threshold to the default value. Initialize the power management module, read the battery power ADC value, and set the initial migration frequency value. Enable the migration controller and begin normal operation. Wait for access requests from the processor core.

[0023] In this embodiment, the robot data access process involves the migration controller receiving a data access request from the processor core and parsing the address. It then queries the SRAM Tag array to determine if the target data is in SRAM. If it is in SRAM (hit): the data is directly read from SRAM and returned to the processor, the corresponding access counter is incremented, the access temperature is updated, and the data is returned, ending the process. If it is not in SRAM (miss), the PRAM metadata is queried to determine if the target data is in PRAM. If it is in PRAM, the data is read from PRAM, the access temperature T is calculated (T is output by the temperature calculation unit), and it is determined whether T is greater than T_hot (the hot data threshold). If T > T_hot: the data is added to the migration queue (PRAM to SRAM direction), the migration priority is set to T multiplied by 15 and rounded down, the data is returned to the processor first (delayed access), and the migration is executed asynchronously in the background. If T ≤ T_hot: the data is directly returned from PRAM to the processor without triggering migration. If the target data is neither in SRAM nor PRAM, it is loaded from main memory (DDR). The initial storage location is determined based on the data type: control data (data used for real-time control loops, with a latency requirement of <1ms. Examples: joint angle commands (updated every millisecond), motor PWM duty cycle (affecting robot motion safety), force / torque sensor readings (used for impedance control)) and emergency data (the highest priority subset of control data, the loss of which may lead to robot malfunction or damage. Examples: emergency stop status flags (0 / 1), collision detection results (whether a collision occurred), safe posture reference values ​​(used to prevent tipping)) are written to SRAM. Historical data and weighted data are written to PRAM, and the corresponding type flags are set. Access statistics are updated (referring to updating the following hardware counters or registers for subsequent migration decisions and performance analysis), and the data is returned to the processor.

[0024] The weighted calculation formula is as follows: ; in, Indicates the access temperature value. Indicates access frequency. Indicates the urgency of access. Indicates the degree of data correlation. This represents migration costs.

[0025] The formula for calculating access frequency is as follows: ; The formula for calculating access urgency is as follows: ; The formula for calculating data correlation is as follows: ; The formula for calculating migration costs is as follows: .

[0026] In this embodiment, the PRAM-SRAM hybrid caching system is applied to the embedded main control platform of the indoor inspection mobile robot. A full-process on-site scenario example is provided: the robot completes its entire lifecycle including power outage shutdown, sudden power failure, power-on restart, and daily task operation, all of which are adapted to this process. Power-on initialization corresponds to the robot's self-test during startup; recovery mode corresponds to the safe posture restoration after an unexpected power outage during restart; normal startup mode corresponds to the standard startup process for a brand-new power-on of the robot with no abnormal power outage records; subsequent data access and migration processes correspond to the entire real-time business read / write process of robot motion control, environmental perception, path planning, and data storage.

[0027] Step S20: Obtain the current task stage status of the intelligent system, and adjust the key data locking status and the access temperature value dynamically according to the task stage status to obtain the allocation strategy.

[0028] Step S20 includes: Step S21: Obtain the current perception stage, planning stage, control stage, and learning stage of the intelligent system; adjust the key data locking status and access temperature value dynamically according to the perception stage, planning stage, control stage, and learning stage to obtain multiple cache weights. Step S22: Obtain the partitioning strategy table, and reallocate the partitioning strategy table according to all the cache weights to obtain the allocation strategy.

[0029] Specifically, the current intelligent system's perception, planning, control, and learning phases are obtained. Based on these phases, the key data locking status and access temperature value are dynamically adjusted using a partitioning strategy to obtain multiple cache weights. A partitioning strategy table is then obtained, and the partitioning strategy table is reallocated based on all the cache weights to obtain an allocation strategy. Furthermore, when multiple task phases are activated simultaneously (e.g., perception and control occur concurrently), the current strategy is determined according to the following rules: The control phase has the highest priority (related to robot safety). If the control phase is activated, the control phase strategy is adopted regardless of the status of other phases. If there is no control phase, the perception phase takes precedence over the planning and learning phases. This mechanism is implemented through a hardware priority encoder, with a decision latency of <10ns.

[0030] For example, such as Figure 3 and Figure 4As shown, the robot data migration process is as follows: the migration decision engine triggers migration when it detects any of the following conditions: SRAM space utilization is greater than 80%; there is data with access temperature T less than T_cold; task phase switching causes a change in partitioning strategy; power optimization requires batch migration. The process involves scanning all SRAM cache lines to find all data where T is less than T_cold and lock_bit is equal to 0. Candidate data is sorted in ascending order of access temperature, with the lowest temperature data migrated first. The available space in the PRAM target bank is checked, and a bank with free space is selected. If PRAM space is insufficient, data eviction (LRU policy) is executed within the PRAM. A migration entry is generated and added to the migration queue: source_addr is the physical address of the SRAM cache line; dest_addr is the free address of the PRAM bank; size is 64B (the size of one cache line); direction is 00 (SRAM to PRAM); priority is floor ((1 minus T) multiplied by 15). The DMA controller retrieves the highest priority entry from the migration queue. DMA performs data transfer: Reads 64B of data from SRAM into the DMA buffer; writes it to the PRAM target address; writes PRAM metadata (Tag, valid bit, etc.). Updates SRAM status: Clears valid_bit; clears dirty_bit; resets the access counter. Updates migration statistics (migration count, cumulative power consumption). Checks if the migration queue is empty; if not, returns to the DMA controller to retrieve the highest priority entry from the migration queue for further processing. Migration complete, releases SRAM space.

[0031] Step S21 includes: Step S211: Obtain the current intelligent system's perception stage, planning stage, control stage, and learning stage; make perception stage judgments on the key data locking status and access temperature value based on the perception conditions of the perception stage and the partition ratio of the partition strategy table to obtain the visual frame cache weight. Step S212: Based on the planning conditions of the planning stage and the partition ratio of the partition strategy table, make a planning stage judgment on the data locking status and the access temperature value to obtain the map cache weight; Step S213: Based on the control conditions of the control phase and the partition ratio of the partition strategy table, the data locking state and the access temperature value are judged in the control phase to obtain the joint state cache weight. Step S214: Based on the learning conditions of the learning stage and the partition ratio of the partition strategy table, determine the data locking status and the access temperature value to obtain the gradient cache weight.

[0032] Specifically, the system acquires the current intelligent system's perception, planning, control, and learning phases. Based on the perception conditions of the perception phase and the partition ratio of the partitioning strategy table, it performs perception phase judgments on the key data locking status and the access temperature value to obtain the visual frame cache weight (perception phase judgment: if (camera data rate is greater than threshold 1) and (perception kernel activation equals 1), then current_stage equals PERCEPTION). Based on the planning conditions of the planning phase and the partition ratio of the partitioning strategy table, it performs planning phase judgments on the data locking status and the access temperature value to obtain the map cache weight (planning phase judgment: if (path planning kernel activation equals 1) and (map access frequency is greater than threshold 2), then current_stage equals PLANNING). Based on the control conditions of the control phase and the partition ratio of the partitioning strategy table, it performs perception phase judgments on the key data locking status and the access temperature value to obtain the visual frame cache weight (planning phase judgment: if (path planning kernel activation equals 1) and (map access frequency is greater than threshold 2), then current_stage equals PLANNING). The lock state and the access temperature value are used for control phase judgment to obtain the joint state cache weight (control phase judgment: if (joint command frequency is greater than threshold 3) or (force sensor access frequency is greater than threshold 4), then current_stage equals CONTROL). Based on the learning conditions of the learning phase and the partitioning policy table (partitioning policy table: a configuration table used to manage the allocation rules of memory resources (such as SRAM and PRAM) under different task phases. It defines how the size, priority, or lock state of various cache regions should be adjusted in each phase to match the computation and data access characteristics of that phase), the data lock state and the access temperature value are used for learning phase judgment to obtain the gradient cache weight (learning phase judgment: if (gradient calculation kernel activation equals 1) and (weight update frequency is greater than threshold 5), then current_stage equals LEARNING).

[0033] As an example, in the robot task switching process, the task perception module continuously monitors the following hardware counters: camera data inflow rate counter; path planning kernel activation counter; joint control command output frequency counter; gradient calculation kernel activation counter. It detects the current task stage. Perception stage judgment: If (camera data rate greater than threshold 1) and (perception kernel activation equals 1), then current_stage equals PERCEPTION. Planning stage judgment: If (path planning kernel activation equals 1) and (map access frequency greater than threshold 2), then current_stage equals PLANNING. Control stage judgment: If (joint command frequency greater than threshold 3) or (force sensor access frequency greater than threshold 4), then current_stage equals CONTROL. Learning stage judgment: If (gradient calculation kernel activation equals 1) and (weight update frequency greater than threshold 5), then current_stage equals LEARNING. It compares current_stage with previous_stage to determine if a change has occurred. If the stage has changed and the time since the last switch is greater than 100ms (anti-shake), a new strategy is loaded from the partitioning strategy table; the required SRAM space adjustment is calculated; and data reallocation is triggered. The data distribution is adjusted according to the new strategy: In the perception phase, a visual frame cache is added, and the historical frame cache is reduced; in the planning phase, a map cache is added, and the visual frame cache is reduced; in the control phase, a joint state cache is added, and urgent data is locked; in the learning phase, a gradient cache is added, and the perception cache is reduced. Corresponding data migration is triggered (between SRAM and PRAM). `previous_stage` is updated to equal `current_stage`, and the switching timestamp is recorded. Monitoring continues.

[0034] Specifically, the dynamic partitioning strategy adjustments are shown in Table 1: Table 1: Adjustments to Dynamic Partitioning Strategy

[0035] As an example, when the system detects a switch from the "perception phase" to the "control phase" in the task phase, it loads the control phase strategy entries from the partitioning strategy table, adjusts the SRAM ratio from 60% to 70%, shifts the SRAM allocation focus to "joint state cache," and sets the locked data type to "emergency control data." Then, it triggers the data reallocation in S05. A portion of the visual frame data is migrated from SRAM to PRAM, and joint state data is prefetched from PRAM to SRAM, with lock_bit=1 set. Through this partitioning strategy table, the system can dynamically optimize cache resources based on the robot's current task phase, ensuring real-time performance while improving overall energy efficiency.

[0036] Step S30: Obtain capacitor power information, migrate the data according to the capacitor power information and the allocation strategy, and obtain migration scheduling results.

[0037] Step S30 includes: Step S31: Obtain capacitor power information. If a power failure signal is detected or the capacitor power information is lower than the first preset power, filter the data according to the type identifier to obtain the first data. Based on the differential writing strategy, write the changed bytes of the first data to the recovery area according to the priority. Step S32: Obtain safety posture information, current task status, and recovery flag. If the capacitor power information is lower than the second preset power, write the safety posture information, the current task status, and the recovery flag into the data to obtain the second data. Step S33: Perform data migration on the first data and the second data according to the capacitor power information and the allocation strategy to obtain the migration scheduling result.

[0038] Specifically, the capacitor charge information is obtained. If a power outage signal is detected or the capacitor charge information is lower than a first preset charge level, the data is filtered according to the type identifier to obtain the first data (traversing all cache lines in SRAM; finding data in the type marker where the persistent flag is equal to 1 and there is already a copy in PRAM; these data are already in PRAM and do not need to be written separately, thus obtaining the first data). Based on the differential write strategy, the changed bytes of the first data are written to the recovery area according to the priority (finding data in all type markers where the persistent flag is equal to 1 and the lock_bit is equal to 0; sorting by priority (higher urgency is prioritized); enabling differential write: only writing changed bytes; writing to PRAM). (Bank7 recovery area), obtain safety posture information, current task status and recovery flag. If the capacitor power information is lower than the second preset power, then write the safety posture information, the current task status and the recovery flag into the data to obtain the second data (write current task status (4B); write safety posture information (16B); write recovery flag (magic number plus checksum plus timestamp); the total data volume is less than 32B, which can be completed within 1ms to obtain the second data). According to the capacitor power information and the allocation strategy, perform data migration on the first data and the second data to obtain the migration scheduling result.

[0039] As an example, the robot's power-off protection process is as follows: The power-off protection module continuously monitors the GPIO interrupt signal and the capacitor's ADC value. Upon detecting a power-off signal (triggered by a GPIO interrupt) or if the battery level is less than 20%, the power-off protection process begins. The power-off protection module sends an NMI (Non-Maskable Interrupt) to the processor, stopping all tasks. Capacitor discharge power is initiated (maintaining system operation for approximately 50ms). Level 1 protection checks are performed: all cache lines in the SRAM are traversed; data with a persistent flag equal to 1 in the type flag and already copied in the PRAM are identified; this data already in the PRAM does not require additional writing. Level 2 protection processes are performed: data with a persistent flag equal to 1 and a lock_bit equal to 0 in all type flags are identified; data is sorted by priority (higher urgency takes precedence); differential writing is enabled: only changed bytes are written; data is written to the PRAM Bank7 recovery area. The remaining capacitor battery level is checked: if the battery level is greater than 15%, Level 2 writing continues; otherwise, jump to S09 to execute Level 3. This process is repeated until all Level 2 data is written or the battery is depleted. Execute Level 3 emergency save (must be performed regardless of battery level): Write current task status (4B); write safe posture information (16B); write recovery flag (magic number plus checksum plus timestamp); total data volume is less than 32B, can be completed within 1ms. Set the recovery flag at PRAM Bank7 address 0x0000 to valid. Power failure protection complete, waiting for power failure. When the system loses power, the data is safely saved in PRAM.

[0040] Step S33 includes: Step S331: Determine a migration strategy based on the capacitor charge information. If the migration frequency of the migration strategy is normal, then perform data migration on the first data and the second data according to the first migration threshold of the migration strategy in the allocation strategy to obtain a first migration scheduling result. Step S332: If the migration frequency is low, then the first data and the second data are migrated according to the allocation strategy and the second migration threshold of the migration strategy to obtain the second migration scheduling result; Step S333: If the migration frequency is low, then the first data and the second data are protected and migrated according to the allocation strategy and the migration strategy to obtain the third migration scheduling result.

[0041] Specifically, a migration strategy is determined based on the capacitor charge information. If the migration frequency of the migration strategy is normal, the first data and the second data are migrated according to the first migration threshold of the migration strategy, resulting in a first migration scheduling result (migration threshold equal to T_cold; batch size equal to 4KB; migration threshold equal to T_cold plus 0.1 (reducing migration amount); batch size equal to 8KB). If the migration frequency is low, the first data and the second data are migrated according to the second migration threshold of the allocation strategy and the migration strategy, resulting in a second migration scheduling result (migration frequency equal to low frequency; migration threshold equal to T_cold plus 0.2; batch size equal to 16KB; PRAM to SRAM migration is prohibited). If the migration frequency is low, the first data and the second data are protected and migrated according to the allocation strategy and the migration strategy, resulting in a third migration scheduling result (migration frequency equal to minimum; only the most necessary SRAM data is retained; PRAM usage is maximized; preheating and power-off protection).

[0042] Furthermore, such as Figure 5 As shown, based on the above-described data migration method based on embodied intelligent hybrid caching, the present invention also provides a data migration system based on embodied intelligent hybrid caching, wherein the data migration system based on embodied intelligent hybrid caching includes: The multi-dimensional data identification module 51 is used to obtain the access characteristics and type identifier of the data, and to perform multi-dimensional identification of the data based on the access characteristics and type identifier to obtain the access temperature value and key data lock status. Data allocation module 52 is used to obtain the current task stage status of the intelligent system, and dynamically adjust the key data locking status and the access temperature value according to the task stage status to obtain an allocation strategy; The data migration module 53 is used to acquire capacitor power information, migrate the data according to the capacitor power information and the allocation strategy, and obtain migration scheduling results.

[0043] Furthermore, such as Figure 6 As shown, based on the above-mentioned data migration method and system based on embodied intelligent hybrid caching, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 6 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0044] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a data migration program 40 based on an embodied intelligent hybrid cache, which can be executed by the processor 10 to implement the data migration method based on an embodied intelligent hybrid cache in this application.

[0045] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the data migration method based on embodied intelligent hybrid cache.

[0046] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The terminals communicate with each other via a system bus.

[0047] In one embodiment, when the processor 10 executes the data migration procedure 40 based on the embodied intelligent hybrid cache in the memory 20, the following steps are performed: The access characteristics and type identifiers of the data are obtained, and the data is identified in multiple dimensions based on the access characteristics and type identifiers to obtain the access temperature value and the key data lock status. Obtain the current task stage status of the intelligent system, and dynamically adjust the key data locking status and the access temperature value according to the task stage status to obtain the allocation strategy. Obtain capacitor power information, migrate the data according to the capacitor power information and the allocation strategy, and obtain migration scheduling results.

[0048] The access characteristics mentioned above include access frequency, access urgency, data correlation, and migration cost; The process of acquiring access features and type identifiers for data, and performing multi-dimensional identification based on these features and identifiers to obtain access temperature and lock status, specifically includes: The access frequency, access urgency, data relevance, and migration cost are obtained. The access temperature value is obtained by weighting the access frequency, access urgency, data relevance, and migration cost. When the type is identified as emergency control data, the data is locked based on the emergency control data to obtain a critical data lock status.

[0049] The step of obtaining access frequency, access urgency, data relevance, and migration cost involves weighting these factors to obtain an access temperature value, and then further includes: Determine whether the access temperature value is greater than a preset thermal data threshold; If the access temperature value is greater than the preset hot data threshold, a migration queue is obtained, and the data is asynchronously migrated according to the priority of the migration queue to obtain migrated data.

[0050] The task phase states include the perception phase, planning phase, control phase, and learning phase. The process of obtaining the current task stage status of the intelligent system, and dynamically adjusting the key data locking status and the access temperature value based on the task stage status to obtain an allocation strategy, specifically includes: The current intelligent system's perception, planning, control, and learning phases are obtained. Based on these phases, the key data locking status and access temperature value are dynamically adjusted using a partitioning strategy to obtain multiple cache weights. Obtain the partitioning strategy table, and reallocate the partitioning strategy table according to all the cache weights to obtain the allocation strategy.

[0051] The cache weights include visual frame cache weights, map cache weights, joint state cache weights, and gradient cache weights. The process involves acquiring the current intelligent system's perception, planning, control, and learning phases, and then dynamically adjusting the key data locking status and access temperature value based on these phases to obtain multiple cache weights. Specifically, this includes: The current intelligent system's perception stage, planning stage, control stage, and learning stage are obtained. Based on the perception conditions of the perception stage and the partition ratio of the partition strategy table, the key data locking status and the access temperature value are judged to obtain the visual frame cache weight. Based on the planning conditions of the planning phase and the partition ratio of the partitioning strategy table, the data locking status and the access temperature value are judged during the planning phase to obtain the map cache weight; Based on the control conditions of the control phase and the partition ratio of the partition strategy table, the control phase judgment is performed on the data locking state and the access temperature value to obtain the joint state cache weight. Based on the learning conditions of the learning phase and the partition ratio of the partitioning strategy table, the data locking status and the access temperature value are judged to obtain the gradient cache weight.

[0052] The step of obtaining capacitor charge information and migrating the data based on the capacitor charge information and the allocation strategy to obtain a migration scheduling result specifically includes: Acquire capacitor power information; if a power failure signal is detected or the capacitor power information is lower than a first preset power level, filter the data according to the type identifier to obtain first data; based on the differential write strategy, write the changed bytes of the first data to the recovery area according to the priority. Acquire safety posture information, current task status, and recovery flag. If the capacitor power information is lower than a second preset power level, then write the safety posture information, current task status, and recovery flag into the data to obtain the second data. Based on the capacitor charge information and the allocation strategy, the first data and the second data are migrated to obtain the migration scheduling result.

[0053] Specifically, the step of migrating the first data and the second data according to the capacitor charge information and the allocation strategy to obtain the migration scheduling result includes: A migration strategy is determined based on the capacitor charge information. If the migration frequency of the migration strategy is normal, the first data and the second data are migrated according to the first migration threshold of the migration strategy to obtain a first migration scheduling result. If the migration frequency is low, then the first data and the second data are migrated according to the allocation strategy and the second migration threshold of the migration strategy to obtain the second migration scheduling result; If the migration frequency is low, then the first data and the second data are protected and migrated according to the allocation strategy and the migration strategy to obtain the third migration scheduling result.

[0054] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a data migration program based on an embodied intelligent hybrid cache, and the data migration program based on an embodied intelligent hybrid cache, when executed by a processor, implements the steps of the data migration method based on an embodied intelligent hybrid cache as described above.

[0055] In summary, this invention provides a data migration method, system, terminal, and storage medium based on embodied intelligent hybrid caching. The method includes: acquiring data access characteristics and type identifiers; performing multi-dimensional identification on the data based on the access characteristics and type identifiers to obtain access temperature values ​​and key data lock status; acquiring the current task stage status of the intelligent system; dynamically adjusting the partitioning strategy based on the task stage status and the key data lock status and access temperature values ​​to obtain an allocation strategy; acquiring capacitor power information; migrating the data based on the capacitor power information and the allocation strategy to obtain a migration scheduling result. This invention obtains temperature and lock status based on data access characteristics and type identifiers, dynamically adjusts the partitioning strategy based on task stages, migrates data based on capacitor power, and then the robot dynamically adjusts the storage strategy based on the task stage to obtain accurate scheduling results.

[0056] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal system that includes that element.

[0057] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0058] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A data migration method based on embodied intelligent hybrid caching, characterized in that, The data migration method based on embodied intelligent hybrid caching includes: The access characteristics and type identifiers of the data are obtained, and the data is identified in multiple dimensions based on the access characteristics and type identifiers to obtain the access temperature value and the key data lock status. Obtain the current task stage status of the intelligent system, and dynamically adjust the key data locking status and the access temperature value according to the task stage status to obtain the allocation strategy. Obtain capacitor power information, migrate the data according to the capacitor power information and the allocation strategy, and obtain migration scheduling results.

2. The data migration method based on embodied intelligent hybrid caching according to claim 1, characterized in that, The access characteristics include access frequency, access urgency, data correlation, and migration cost; The process of acquiring access features and type identifiers for data, and performing multi-dimensional identification based on these features and identifiers to obtain access temperature and lock status, specifically includes: The access frequency, access urgency, data relevance, and migration cost are obtained. The access temperature value is obtained by weighting the access frequency, access urgency, data relevance, and migration cost. When the type is identified as emergency control data, the data is locked based on the emergency control data to obtain a critical data lock status.

3. The data migration method based on embodied intelligent hybrid caching according to claim 2, characterized in that, The process of obtaining access frequency, access urgency, data relevance, and migration cost involves weighting these factors to obtain an access temperature value, and then further includes: Determine whether the access temperature value is greater than a preset thermal data threshold; If the access temperature value is greater than the preset hot data threshold, a migration queue is obtained, and the data is asynchronously migrated according to the priority of the migration queue to obtain migrated data.

4. The data migration method based on embodied intelligent hybrid caching according to claim 3, characterized in that, The task phase states include the perception phase, planning phase, control phase, and learning phase; The process of obtaining the current task stage status of the intelligent system, and dynamically adjusting the key data locking status and the access temperature value based on the task stage status to obtain an allocation strategy, specifically includes: The current intelligent system's perception, planning, control, and learning phases are obtained. Based on these phases, the key data locking status and access temperature value are dynamically adjusted using a partitioning strategy to obtain multiple cache weights. Obtain the partitioning strategy table, and reallocate the partitioning strategy table according to all the cache weights to obtain the allocation strategy.

5. The data migration method based on embodied intelligent hybrid caching according to claim 4, characterized in that, The cache weights include visual frame cache weights, map cache weights, joint state cache weights, and gradient cache weights; The process involves acquiring the current intelligent system's perception, planning, control, and learning phases, and then dynamically adjusting the key data locking status and access temperature value based on these phases to obtain multiple cache weights. Specifically, this includes: The current intelligent system's perception stage, planning stage, control stage, and learning stage are obtained. Based on the perception conditions of the perception stage and the partition ratio of the partition strategy table, the key data locking status and the access temperature value are judged to obtain the visual frame cache weight. Based on the planning conditions of the planning phase and the partition ratio of the partitioning strategy table, the data locking status and the access temperature value are judged during the planning phase to obtain the map cache weight; Based on the control conditions of the control phase and the partition ratio of the partition strategy table, the control phase judgment is performed on the data locking state and the access temperature value to obtain the joint state cache weight. Based on the learning conditions of the learning phase and the partition ratio of the partitioning strategy table, the data locking status and the access temperature value are judged to obtain the gradient cache weight.

6. The data migration method based on embodied intelligent hybrid caching according to claim 5, characterized in that, The process of acquiring capacitor charge information and migrating the data based on the capacitor charge information and the allocation strategy to obtain a migration scheduling result specifically includes: Acquire capacitor power information; if a power failure signal is detected or the capacitor power information is lower than a first preset power level, filter the data according to the type identifier to obtain first data; based on the differential write strategy, write the changed bytes of the first data to the recovery area according to the priority. Acquire safety posture information, current task status, and recovery flag. If the capacitor power information is lower than a second preset power level, then write the safety posture information, current task status, and recovery flag into the data to obtain the second data. Based on the capacitor charge information and the allocation strategy, the first data and the second data are migrated to obtain the migration scheduling result.

7. The data migration method based on embodied intelligent hybrid caching according to claim 6, characterized in that, The step of migrating the first data and the second data according to the capacitor charge information and the allocation strategy to obtain the migration scheduling result specifically includes: A migration strategy is determined based on the capacitor charge information. If the migration frequency of the migration strategy is normal, the first data and the second data are migrated according to the first migration threshold of the migration strategy to obtain a first migration scheduling result. If the migration frequency is low, then the first data and the second data are migrated according to the allocation strategy and the second migration threshold of the migration strategy to obtain the second migration scheduling result; If the migration frequency is low, then the first data and the second data are protected and migrated according to the allocation strategy and the migration strategy to obtain the third migration scheduling result.

8. A data migration system based on embodied intelligent hybrid caching, characterized in that, The data migration system based on embodied intelligent hybrid caching includes: A multi-dimensional data identification module is used to obtain the access characteristics and type identifiers of data, and to perform multi-dimensional identification of data based on the access characteristics and type identifiers to obtain the access temperature value and key data lock status. The data allocation module is used to obtain the current task stage status of the intelligent system, and dynamically adjust the key data locking status and the access temperature value according to the task stage status to obtain the allocation strategy. The data migration module is used to acquire capacitor power information, migrate the data according to the capacitor power information and the allocation strategy, and obtain migration scheduling results.

9. A terminal, characterized in that, The terminal includes: a memory, a processor, and a data migration program based on an embodied intelligent hybrid cache stored in the memory and executable on the processor. When the data migration program based on the embodied intelligent hybrid cache is executed by the processor, it implements the steps of the data migration method based on an embodied intelligent hybrid cache as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a data migration program based on an embodied intelligent hybrid cache, which, when executed by a processor, implements the steps of the data migration method based on an embodied intelligent hybrid cache as described in any one of claims 1-7.

Images