Device control method, smart device, and storage medium

Abstract

The application provides a device control method, a smart device and a storage medium, and is applied to a smart device. The method comprises the following steps: acquiring first information of a current control period, wherein the first information comprises current state information or a current control residual identifier; determining a first control residual record from a residual memory module based on the first information; optimizing a control strategy of the smart device based on the first control residual record; and performing a control operation based on the optimized control strategy in a next control period. The embodiment of the application can improve the control accuracy of the device.

Description

Technical Field

[0001] This application relates to the field of equipment control technology, and in particular to equipment control methods, intelligent devices, and storage media. Background Technology

[0002] In the field of equipment control, the rationality of the control strategy directly determines the control accuracy of intelligent devices (such as mobile robots, embodied intelligence, and robotic arms). In related technologies, the control strategies of intelligent devices are typically based on a preset, fixed architecture, capable of executing control operations only according to preset logic. Therefore, in complex working conditions or dynamic environments, fixed control strategies are insufficient to meet the demands of high-precision control. Summary of the Invention

[0003] This application provides a device control method, a smart device, and a storage medium, with the aim of improving the control accuracy of the device.

[0004] Firstly, a device control method is provided for use in a smart device, the method comprising: Obtain the first information of the current control cycle, which includes the current status information or the current control residual identifier; Based on the first information, the first control residual record is determined from the residual memory module; Based on the first control residual record, the control strategy of the intelligent device is optimized; In the next control cycle, control operations are executed based on the optimized control strategy.

[0005] In this technical solution, by introducing a residual memory mechanism, the execution error of each control cycle is transformed into the basis for strategy optimization in the next cycle, thus constructing a control closed loop with cross-cycle error learning and adaptive correction capabilities, thereby continuously improving control accuracy.

[0006] In conjunction with the first aspect, in one possible implementation, before obtaining the first information of the current control cycle, the method further includes: Generate a control input sequence based on the control strategy of the intelligent device; Based on the control input sequence, predict the first execution state sequence of the smart device; During the current control cycle, while performing control operations based on the control input sequence, the second execution state sequence of the intelligent device is acquired. The current control residual identifier is determined based on the first execution state sequence and the second execution state sequence.

[0007] In this technical solution, control residual identifiers are generated based on the actual state sequence and the predicted state sequence, which can capture the distribution characteristics of the error throughout the entire control cycle, making the residual identifiers carry richer information, and providing more sufficient basis for the subsequent residual memory module to retrieve and match historical residuals.

[0008] In conjunction with the first aspect, in one possible implementation, determining the current control residual identifier based on the first execution state sequence and the second execution state sequence includes: The first execution state sequence and the second execution state sequence are aligned to obtain the third execution state sequence and the fourth execution state sequence. Determine the common segmentation point between the third execution state sequence and the fourth execution state sequence; Based on the common segmentation point, the third execution state sequence and the fourth execution state sequence are segmented respectively to obtain multiple sequence groups. Each sequence group includes a first sub-sequence segmented from the third execution state sequence and a second sub-sequence segmented from the fourth execution state sequence, and the first sub-sequence and the second sub-sequence are aligned. Determine the control residual identifier for each of the sequence groups, and use the determined control residual identifier as the current control residual identifier.

[0009] In this technical solution, after aligning the predicted state sequence with the measured state sequence, a common segmentation point is determined, and control residual identifiers for each stage are generated in segments. The error information of the complete control cycle is structurally decomposed into multiple stage local residuals, which not only avoids timing misalignment and cross-stage error aliasing, but also achieves fine stage positioning of errors, providing hierarchical residual data for the residual memory module.

[0010] In conjunction with the first aspect, in one possible implementation, for each said sequence group, determining the control residual identifier of the sequence group includes: Based on the first subsequence and the second subsequence included in the sequence group, the control residual information of the sequence group is determined, and the control residual information includes: state change sequence, control input sequence, state error change information, and constraint state change information; The control residual identifier is determined based on the control residual information.

[0011] In this technical solution, multi-dimensional control residual information, such as state changes, control inputs, state error changes, and constraint state changes, is extracted from the predicted and measured subsequence pairs and residual identifiers are generated. This allows the residual identifiers to carry multi-dimensional semantics, such as error magnitude, dynamic characteristics, control background, and constraint activation. This fundamentally improves the residual identifiers' ability to distinguish the causes of errors and provides an information basis for the accurate matching of the residual memory module.

[0012] In conjunction with the first aspect, in one possible implementation, determining the control residual identifier based on the control residual information includes: The control residual information is mapped into a vector representation; The vector representation is converted into a control residual identifier, which is used to represent a specific type of control deviation pattern.

[0013] In this technical solution, by first vectorizing and uniformly representing multidimensional control residual information and then mapping it to specific residual identifiers, heterogeneous residual data is abstracted into patterned knowledge that can be indexed and reused across cycles. This enables the residual memory module to identify historical similar deviation patterns and directly reuse corresponding optimization experience, thereby significantly improving the efficiency and pertinence of control strategy optimization.

[0014] In conjunction with the first aspect, in one possible implementation, determining the common segmentation point of the third execution state sequence and the fourth execution state sequence includes: Monitor the state change characteristics in the third execution state sequence or the fourth execution state sequence; When determining the existence of a target event based on the state change characteristics, the time node where the target event occurs is taken as the common segmentation point. The target event includes at least one of the following: contact state change event, control constraint change event, and dynamic characteristic change event.

[0015] In this technical solution, by monitoring the execution state sequence, the target events existing in the current control cycle are determined, and the event occurrence time node is used as a common segmentation point to achieve adaptive sequence segmentation that is strictly aligned with the physical stage of the control process. This ensures the consistency of the dynamic background and constraints within each sequence group and fundamentally eliminates cross-stage error aliasing.

[0016] In conjunction with the first aspect, in one possible implementation, after determining the current control residual identifier, the method further includes: The second control residual record is stored in the residual memory module. The second control residual record includes the control task information of the current control cycle, environmental change information, control execution result, and the current control residual identifier.

[0017] In this technical solution, by writing control task information, environmental change information, control execution results and control residual identifiers as complete associated records into the residual memory module, a structured residual knowledge base with four-dimensional indexing capabilities of task, environment, result and pattern is constructed, enabling historical residual experience to be accurately retrieved and reused according to multi-dimensional conditions.

[0018] In conjunction with the first aspect, in one possible implementation, optimizing the control strategy of the intelligent device based on the first control residual record includes: Based on the first control residual record, the control law of the control strategy of the intelligent device is optimized; the optimization includes at least one of the following: modifying the control variables, adjusting the control objective function, and adjusting the control constraints.

[0019] In this technical solution, based on historical control residual records, targeted optimization is carried out on the control variables, objective functions, and constraints of the control law, covering the complete operable space of control law optimization. This allows the historical deviation knowledge accumulated in the residual memory to be directly transformed into precise improvements to the control law, achieving cross-cycle adaptive optimization of the control law.

[0020] Secondly, a device control apparatus is provided for use in intelligent devices, comprising: The acquisition module is used to acquire the first information of the current control cycle, the first information including the current status information or the current control residual identifier; The determining module is used to determine the first control residual record from the residual memory module based on the first information; An optimization module is used to optimize the control strategy of the intelligent device based on the first control residual record; The execution module is used to perform control operations based on the optimized control strategy in the next control cycle.

[0021] Thirdly, a smart device is provided, including a memory and a processor, the memory being connected to the processor, the processor being configured to execute one or more computer programs stored in the memory, the processor causing the smart device to implement the device control method of the first aspect described above when executing the one or more computer programs.

[0022] Fourthly, a computer-readable storage medium is provided, which stores a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the device control method of the first aspect.

[0023] Fifthly, a computer program product is provided that, when run on a smart device, causes any of the methods provided in the first aspect to be executed. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic flowchart illustrating a device control method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a device control apparatus provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a smart device provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0027] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. Moreover, the terms "first," "second," and "third" used in this application do not limit the data or execution order, but only distinguish identical or similar items with essentially the same function and effect.

[0028] The technical solution of this application is described in detail below. This technical solution can be applied to intelligent devices. Intelligent devices refer to mechatronic equipment that possesses signal reception, logic operation, decision control, and motion adjustment capabilities, and can independently or integrate sensing, driving, and execution modules to achieve precise control of its own and related systems' position, attitude, trajectory, and other parameters. Its scope includes independent control components such as PLCs and servo drives, as well as integrated intelligent equipment that integrates control logic, sensing units, and motion execution mechanisms, capable of autonomously completing tasks such as navigation, obstacle avoidance, grasping, and trajectory movement, such as mobile robots and multi-degree-of-freedom robotic arms. These are widely used in automated production and intelligent operations, and their core function is to ensure the system operates stably, accurately, and efficiently according to preset instructions.

[0029] See Figure 1 , Figure 1 This is a flowchart illustrating a device control method provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes the following steps: S101, Obtain the first information of the current control cycle, the first information including the current status information or the current control residual identifier.

[0030] In the operation of intelligent devices, the current control cycle refers to the specific time interval during which the control system completes one closed-loop process of "signal acquisition → logic operation → instruction output → status feedback". It is a single cycle unit in the currently executing continuous cyclic control.

[0031] Current status information refers to the current state of an intelligent device. Specifically, it refers to the comprehensive data set collected by various sensors within the current control cycle, used to describe the real-time physical state of the system. This data is the core foundation for the intelligent device to understand the current operating status of the system, achieve precise control, and assist in residual retrieval. This information all comes from real-time sensor data collection. For mobile robots, the current status information mainly includes core physical parameters such as current position, movement speed, orientation angle, and battery level. For robotic arms, the current status information focuses on key data such as joint angles, end effector pose, and force / torque sensor readings. Together, these pieces of information accurately reflect the actual physical state of the intelligent device.

[0032] The current control residual token, or residual token, is a specific identifier used to characterize the deviation pattern between the control input of the intelligent device and the actual dynamic response of the system within the current control cycle. The core function of this token is to accurately describe the deviation characteristics between the "current control input" and the "actual dynamic response of the system"—within the current control cycle, after the intelligent device outputs a control command (control input), the system generates a corresponding dynamic response, and the current control residual token (residual token) is specifically used to capture and describe the deviation patterns and characteristics between the two.

[0033] S102, based on the first information, determine the first control residual record from the residual memory module.

[0034] The first control residual record can be one or more.

[0035] The residual memory module stores multiple control residual records, which are the control residual records of the control cycles preceding the current control cycle.

[0036] Control residual records are a systematic retention of core information related to the control process. Its core components include four categories: control residual token, control task information, environmental status information, and control execution results. These four categories of information are interconnected and indispensable, and together they construct a complete experience chain of "context-deviation-result", clearly tracing the key details and causal relationships in the control process.

[0037] The "Context" section consists of control task information and environmental state information, which clearly records "under what circumstances" the control process occurs—that is, the specific requirements and parameters of the current control task, as well as the environmental conditions during control execution; the control residual identifier corresponds to the "Deviation" section, which accurately describes the deviation pattern between the control input and the system dynamic response under this context; the control execution result corresponds to the "Result" section, which records the final execution status and output result of the control process after the deviation occurs.

[0038] In other words, by integrating contextual information of "task + environment", deviation information of "residual token", and final information of "execution result", control residual records completely retain the experience data of each control process, providing comprehensive and coherent experience support for subsequent residual retrieval, control strategy optimization, and anomaly tracing.

[0039] In one embodiment, the first information includes a current control residual identifier, and determining the first control residual record from the residual memory module based on the first information includes: The current control residual identifier is matched with the control residual identifiers included in the control residual records stored in the residual memory module to determine the first control residual record. The similarity between the control residual identifiers included in the first control residual record and the current control residual identifier is greater than or equal to a first set value.

[0040] For example, the settings are as follows: The initial setting is 85%; Current control scenario: The robotic arm is performing a "workpiece gripping" control task. During the current control cycle, a deviation mode of "end position overshoot 0.2mm and torque deviation 5N·m" was detected. The corresponding current control residual is identified as "Token-20240508-01". Residual memory module storage: Three historical control residual records have been retained. The control residual identifiers and corresponding deviation modes of these three records are as follows: Historical record 1: Residual identifier "Token-20240420-03", deviation mode "End position overshoot 0.18mm, torque deviation 4.8N·m"; Historical record 2: Residual identifier "Token-20240501-07", deviation mode "End position overshoot 0.35mm, torque deviation 7.2N·m"; Historical record 3: Residual identifier "Token-20240505-02", deviation mode "End position overshoot 0.22mm, torque deviation 5.3N·m".

[0041] The current control residual identifier "Token-20240508-01" is compared with the residual identifiers of three historical records in the residual memory module for similarity (the core of the comparison is the matching of the quantitative parameters of the deviation mode): Similarity with historical record 1 identifier: 92% (position overshoot, torque deviation parameters are very close), ≥85%; Similarity with historical record 2 identifier: 68% (position overshoot, torque deviation parameters are significantly different), <85%; Similarity with historical record 3 identifier: 88% (position overshoot, torque deviation parameters are close), ≥85%.

[0042] According to the comparison results, the similarity between the control residual identifiers of historical records 1 and 3 and the current control residual identifier is ≥85%. Therefore, these two historical control residual records are the "first control residual records".

[0043] In another embodiment, the first information includes current state information, and determining the first control residual record from the residual memory module based on the first information includes: The current state information is matched with the control task information and environmental state information included in the control residual record stored in the residual memory module to determine the first control residual record. The similarity between the control task information and environmental state information included in the first control residual record and the current control residual identifier is greater than or equal to a second set value.

[0044] For example, the settings are as follows: The second setting is 80%; Current control scenario: The current status information of the mobile robot (collected by sensors) is as follows: performing the "warehouse goods handling" task, current location is warehouse area A (coordinates X100, Y50), ambient temperature is 25℃, there are no obstacles on the ground, and the battery level is 80%. Residual memory module storage: Three historical control residual records have been retained. The control task information and environmental status information of these three records are as follows: Historical record 1: The control task information is "Warehouse cargo handling", and the environmental status information is "Warehouse A area (coordinates X102, Y48), ambient temperature 24℃, no obstacles on the ground, battery level 82%"; Historical record 2: The control task information is "Warehouse goods sorting", and the environmental status information is "Warehouse B area (coordinates X80, Y60), ambient temperature 28℃, slight obstacles on the ground, battery power 75%"; Historical record 3: The control task information is "Warehouse cargo handling", and the environmental status information is "Warehouse A area (coordinates X98, Y52), ambient temperature 26℃, no obstacles on the ground, battery level 78%".

[0045] The current status information is compared with the "control task information + environment status information" of these three historical records in a comprehensive similarity comparison: Overall similarity with historical record 1: 90% (same task, minimal location deviation, similar ambient temperature, no obstacles, similar battery level), ≥80%; Overall similarity with historical record 2: 55% (different tasks, large location differences, and significant deviations in environmental parameters), <80%; Overall similarity with historical record 3: 86% (same task, small positional deviation, similar ambient temperature, no obstacles, similar battery level), ≥80%.

[0046] According to the comparison results, the overall similarity between the "control task information + environmental status information" of historical records 1 and 3 and the current status information is ≥80%. Therefore, these two historical control residual records are the "first control residual records".

[0047] S103, Based on the first control residual record, optimize the control strategy of the intelligent device.

[0048] Among them, the control strategy of intelligent devices refers to a set of preset rules, algorithm logic and decision-making processes adopted by intelligent devices to achieve preset control goals. The core is to solve the core problems of "how intelligent devices act, how to adjust control parameters and how to deal with operational deviations".

[0049] S104, In the next control cycle, execute control operations based on the optimized control strategy.

[0050] In this embodiment, by introducing a residual memory mechanism, the execution error of each control cycle is transformed into the basis for strategy optimization in the next cycle, thus constructing a control closed loop with cross-cycle error learning and adaptive correction capabilities, thereby continuously improving control accuracy.

[0051] In one implementation, before obtaining the first information of the current control cycle, the method further includes: Generate a control input sequence based on the control strategy of the intelligent device; Based on the control input sequence, predict the first execution state sequence of the smart device; During the current control cycle, while performing control operations based on the control input sequence, the second execution state sequence of the intelligent device is acquired. The current control residual identifier is determined based on the first execution state sequence and the second execution state sequence.

[0052] In this context, the control input sequence refers to the ordered set of control commands sent to the actuators (such as motor drivers and joint controllers in intelligent devices) within each control cycle. For example, for a mobile robot, the control input sequence is a sequence of velocity commands (linear velocity, angular velocity) at each moment; for a robotic arm, the control input sequence is a sequence of target torques or target positions for each joint. The actuators, based on the intelligent device's control strategy (generated by a large language model or optimizer), output a series of discrete-time-step control commands, thus forming the control input sequence.

[0053] After receiving the control input sequence, the actuator maps the control input sequence to the expected sequence of state changes using the system's forward dynamics model (or predictive model). For example, for a mobile robot, the forward dynamics model calculates the predicted position and velocity at future moments based on the current velocity command and kinematic equations; for a robotic arm, the forward dynamics model calculates the predicted angles and end-effector poses of each joint at future moments based on the torque commands and dynamic equations for each joint. This predictive model can be an analytical model (such as the Newton-Euler equations of dynamics) or a learning model (such as a forward predictor fitted by a neural network).

[0054] The second execution state sequence refers to the sequence of changes in actual state data over time, collected in real time by sensors (such as encoders, inertial measurement units, force sensors, lidar, etc.) during the execution of the control input sequence by the intelligent device. For example, for a mobile robot, the actual execution state trajectory is the actual position (coordinates), velocity, and orientation at each moment; for a robotic arm, the actual execution state trajectory is the actual angle, angular velocity, and end effector pose of each joint at each moment.

[0055] The difference between the second execution state sequence and the first execution state sequence is that the first execution state sequence is the estimated ideal state change, while the second execution state sequence is the actual state change that occurs in the physical world. The deviation between the two is used to determine the control residual identifier.

[0056] In this embodiment, the control residual identifier is generated based on the actual state sequence and the predicted state sequence, which can capture the distribution characteristics of the error throughout the entire control cycle, making the residual identifier carry more information, and providing a more sufficient basis for the subsequent residual memory module to retrieve and match historical residuals.

[0057] In one implementation, determining the current control residual identifier based on the first execution state sequence and the second execution state sequence includes: The first execution state sequence and the second execution state sequence are aligned to obtain the third execution state sequence and the fourth execution state sequence. Determine the common segmentation point between the third execution state sequence and the fourth execution state sequence; Based on the common segmentation point, the third execution state sequence and the fourth execution state sequence are segmented respectively to obtain multiple sequence groups. Each sequence group includes a first sub-sequence segmented from the third execution state sequence and a second sub-sequence segmented from the fourth execution state sequence, and the first sub-sequence and the second sub-sequence are aligned. Determine the control residual identifier for each of the sequence groups, and use the determined control residual identifier as the current control residual identifier.

[0058] The alignment process includes time dimension alignment and state space mapping alignment.

[0059] Specifically, since the first execution state sequence is a predicted state sequence based on the control strategy, and the second execution state sequence is an actual state sequence collected during control execution, there may be inconsistencies in their time bases or coordinate systems. Therefore, alignment is required. Time dimension alignment means unifying the first and second execution state sequences to the same time base, ensuring that the state data of the two sequences strictly correspond at each time step when calculating the deviation between the predicted and actual states, avoiding spurious residuals introduced by time misalignment. State space mapping alignment means unifying the first and second execution state sequences to the same coordinate system, ensuring that the state change characteristics of the predicted and actual state sequences are comparable within the same reference frame when identifying state change events (including contact state change events, control constraint change events, and dynamic characteristic change events) during the control process, thereby accurately determining the common segmentation point. The third and fourth execution state sequences obtained after alignment are consistent in both time base and coordinate system, providing an accurate data foundation for subsequent sequence segmentation based on the common segmentation point and the generation of residual identifiers.

[0060] In this embodiment, by aligning the predicted state sequence with the measured state sequence to determine the common segmentation point and generating control residual identifiers for each stage in segments, the error information of the complete control cycle is structurally decomposed into multiple stage local residuals. This avoids timing misalignment and cross-stage error aliasing, and also achieves fine stage positioning of the error, providing hierarchical residual data for the residual memory module.

[0061] In one implementation, determining the common segmentation point of the third execution state sequence and the fourth execution state sequence includes: Monitor the state change characteristics in the third execution state sequence or the fourth execution state sequence; When determining the existence of a target event based on the state change characteristics, the time node where the target event occurs is taken as the common segmentation point. The target event includes at least one of the following: contact state change event, control constraint change event, and dynamic characteristic change event.

[0062] The system aligns the first and second execution state sequences to obtain the third and fourth execution state sequences. Then, it monitors the state change characteristics within either the third or fourth execution state sequence in real time. Since the third and fourth execution state sequences are based on a unified time reference and coordinate system, the event time points determined by monitoring either sequence are equally applicable to both sequences. Specifically, the system continuously detects state change characteristics within the sequence based on preset detection thresholds to determine whether triggering conditions for the target event exist.

[0063] A contact state change event refers to a change in the contact relationship between a smart device and its external environment. For example, when a robotic arm's end effector moves from free space to contact an object's surface (from no contact to contact), or when a gripper releases an object (from contact to no contact), the event is triggered when the contact force exceeds a preset detection threshold. Control constraint change events refer to the activation or deactivation of constraints during the control process. For example, when a mobile robot approaches an obstacle while performing an obstacle avoidance task, the safety distance constraint is activated, or when a robotic arm reaches a joint limit, the joint angle constraint is activated. The event is determined to be triggered when the corresponding constraint state changes. A dynamic characteristic change event refers to a significant change in the dynamic behavior of the controlled object. For example, a robotic arm switching from no-load motion to loaded motion causes a sudden increase in inertia, or a mobile robot moving from flat ground to a slope or soft ground causes a change in the ground friction coefficient. The event is determined to be triggered when the amount of dynamic characteristic change exceeds a preset detection threshold.

[0064] When it is determined that any of the above target events exist based on the state change characteristics, the time node where the target event occurs is taken as the common segmentation point. At this time node, segmentation marks are inserted into the third execution state sequence and the fourth execution state sequence to ensure that the predicted state trajectory and the actual execution state trajectory are cut at the same position, and the correspondence between the two sequences in each segment interval remains intact.

[0065] For example, taking a robotic arm performing a grasping task, the complete execution state sequence can be divided into three segments: the first segment (approach segment): the robotic arm moves from its initial position to the vicinity of the target object. In this segment, the robotic arm is in free space, without contact constraints, and exhibits consistent free-motion dynamics. The second segment (contact segment): the end effector contacts the object's surface and applies a grasping force. In this segment, it is constrained by contact forces, and its dynamics differ from the first segment. The third segment (grasping and transporting segment): the grippers close, clamp the object, and transport it to the target location. In this segment, it is constrained by load gravity, friction, etc., and its dynamics change again. Each segment corresponds to an independent control process interval, meaning that within each segment, the control strategy can employ different control parameters, constraints, and optimization objectives, thereby making the control more refined and adaptable.

[0066] In this embodiment of the application, by monitoring the execution state sequence, the target event existing in the current control cycle is determined, and the event occurrence time node is used as a common segmentation point to achieve adaptive sequence segmentation that is strictly aligned with the physical stage of the control process. This ensures the consistency of the dynamic background and constraints within each sequence group and fundamentally eliminates cross-stage error aliasing.

[0067] In one implementation, for each sequence group, determining the control residual identifier of the sequence group includes: Based on the first subsequence and the second subsequence included in the sequence group, the control residual information of the sequence group is determined, and the control residual information includes: state change sequence, control input sequence, state error change information, and constraint state change information; The control residual identifier is determined based on the control residual information.

[0068] Among them, the state change sequence refers to the sequence of changes in the system state over time within the time interval corresponding to the sequence group. For example, in the contact segment of the robotic arm, the sequence of changes in the position, speed and torque sensor readings of the end effector over time records the state change process experienced by the system within this interval.

[0069] The control input sequence refers to the sequence of control commands sent to the actuator within the time interval corresponding to the sequence group. For example, the sequence of target torque commands sent to the joint motor at each moment in the contact segment. The control commands given by the control strategy within this interval are recorded.

[0070] State error change information refers to the trend of the deviation between the first subsequence and the second subsequence within the time interval corresponding to the sequence group. For example, in contact segmentation, the deviation between the predicted end position and the actual end position changes over time, reflecting the prediction accuracy of the prediction model and the impact of external disturbances.

[0071] Constraint state change information refers to the state changes of control constraints within the time interval corresponding to the sequence group. For example, in the contact segment, the contact force constraint changes from none to one, and the upper limit of torque constraint is activated. This records the changes in the constraint environment within the interval.

[0072] The above four types of information are combined to form structured control residual information, which is used to describe the deviation structure between the control input and the system dynamic response within the time interval corresponding to the sequence group.

[0073] In this embodiment, by extracting multi-dimensional control residual information such as state changes, control inputs, state error changes, and constraint state changes from the predicted and measured subsequence pairs and generating residual identifiers, the residual identifiers simultaneously carry multi-dimensional semantics such as error magnitude, dynamic features, control background, and constraint activation. This fundamentally improves the residual identifiers' ability to distinguish the causes of errors and provides an information basis for the accurate matching of the residual memory module.

[0074] In one implementation, determining the control residual identifier based on the control residual information includes: The control residual information is mapped into a vector representation; The vector representation is converted into a control residual identifier, which is used to represent a specific type of control deviation pattern.

[0075] First, an encoder maps control residual information into a continuous vector representation. Multi-dimensional heterogeneous residual information, including state change sequences, control input sequences, state error change information, and constraint state change information, is uniformly encoded into the same vector space, resulting in the residual vector representation of the sequence group. Then, vector quantization technology is used to convert this residual vector representation into a discrete control residual identifier. Specifically, the residual vector representation is compared with all candidate codewords in a pre-trained residual codebook for similarity. The codeword most similar to the residual vector representation is selected, and its corresponding discrete index number is used as the control residual identifier for the sequence group. This control residual identifier represents a specific type of control deviation pattern within the corresponding segment interval of the sequence group. Through this method, the continuous residual information of each sequence group is abstracted into indexable and matchable discrete deviation pattern identifiers, providing a structured residual representation foundation for efficient retrieval by the residual memory module and subsequent control strategy optimization.

[0076] In this embodiment, by first vectorizing and uniformly representing multidimensional control residual information and then mapping it to specific residual identifiers, heterogeneous residual data is abstracted into patterned knowledge that can be indexed and reused across cycles. This enables the residual memory module to identify historical similar deviation patterns and directly reuse the corresponding optimization experience, thereby significantly improving the efficiency and pertinence of control strategy optimization.

[0077] In this embodiment of the application, after determining the current control residual identifier, the method further includes: The second control residual record is stored in the residual memory module. The second control residual record includes the control task information of the current control cycle, environmental change information, control execution result, and the current control residual identifier.

[0078] The control task information refers to the task description and target setting for the current control cycle. For example, for a mobile robot navigation task, the control task information includes the starting position, target position, path points, and task type (such as point-to-point navigation, patrol, etc.); for a robotic arm operation task, the control task information includes the operation type (such as grasping, placing, assembling, etc.), target object position, and target pose. This control task information is used to provide a basis for task relevance matching when retrieving historical residual records in subsequent control cycles, ensuring that the retrieval results match the current task objective.

[0079] Environmental change information refers to the state description of the environment in which the intelligent device is located during the current control cycle. This includes environmental factors that affect the mechanical characteristics of control execution, such as the location and shape of obstacles, ground type (flat ground, slope, steps, etc.), ambient temperature and humidity, and lighting conditions. This environmental change information is used to provide a basis for matching environmental conditions during subsequent retrievals, supports the residual memory module in identifying the differences in performance of the same deviation pattern under different environmental conditions, and provides a basis for corrective environmental perception for control strategy optimization.

[0080] Control execution results refer to the actual state and effect evaluation achieved at the end of the current control cycle. Examples include the distance deviation between the actual position reached by the mobile robot and the target position, whether the task was successfully completed, whether the robotic arm successfully grasped the target, and the end-effector pose error. These control execution results are used to evaluate the effectiveness of the control strategy within the current control cycle and are stored in association with the current control residual identifier. This ensures that the records accumulated in the residual memory module simultaneously include deviation patterns and their corresponding control effects, providing a result-oriented reference for subsequent control strategy optimization.

[0081] The current control residual identifier, which is the discrete deviation pattern identifier obtained by vectorizing and quantizing the control residual information of each segment interval of the current control cycle, is stored as the core index field of the second control residual record, supporting the residual memory module to perform efficient retrieval in subsequent control cycles using the deviation pattern as the key.

[0082] In this embodiment, by writing control task information, environmental change information, control execution results, and control residual identifiers as complete associated records into the residual memory module, a structured residual knowledge base with four-dimensional indexing capabilities of task, environment, result, and pattern is constructed, enabling historical residual experience to be accurately retrieved and reused according to multi-dimensional conditions.

[0083] In one implementation, optimizing the control strategy of the intelligent device based on the first control residual record includes: Based on the first control residual record, the control law of the control strategy of the intelligent device is optimized; the optimization includes at least one of the following: modifying the control variables, adjusting the control objective function, and adjusting the control constraints.

[0084] Specifically, the control law of the intelligent device's control strategy is optimized based on the first control residual record retrieved from the residual memory module. The first control residual record contains deviation pattern identifiers, task information, environmental information, and control execution results from historical control cycles matching the current control scenario. Based on this, the control law is subject to targeted correction without replacing the overall architecture of the control strategy. The optimization includes at least one of the following: Correcting control variables refers to making directional corrections to control variables (such as gain parameters, weighting coefficients, feedforward compensation amounts, etc.) in the control law based on the historical deviation patterns and corresponding control execution results recorded in the first control residual record. The magnitude and direction of the correction are directly driven by the deviation data in the historical residual record. Under the premise of keeping the control law structure unchanged, known systematic deviations are compensated, so that the correction of control variables has a clear residual basis. Adjusting the control objective function refers to adjusting the optimization weights or optimization indicators of the objective function in the control law based on the systematic deviation patterns revealed by the first control residual record. When the historical residual record shows that the setting of the current objective function leads to the continuous sacrifice of a certain type of state, the trade-off relationship between multiple objectives is rebalanced to correct the systematic bias of the control law from the perspective of optimization direction. Adjusting the control constraints refers to adjusting the constraint boundaries or constraint forms in the control law based on the constraint state change information in the first control residual record, identifying and correcting control deviations caused by unreasonable constraint settings, so that the control law can obtain a larger effective optimization space while satisfying the safety boundary.

[0085] Furthermore, in the next control cycle, a control operation is performed based on the optimized control strategy, including: in the next control cycle, generating a control input sequence based on the optimized control strategy, and performing a control operation based on the control input sequence.

[0086] After performing the control operation, the process also includes: determining the control residual identifier for the next control cycle. The method for determining the control residual identifier for the next control cycle is the same as the method for determining the "current control residual identifier" described above; please refer to the above content.

[0087] As can be seen, after optimizing the control strategy, a new control operation is executed based on this new control strategy, thereby starting a new round of control residual identification, retrieval and control strategy optimization process, realizing a continuous iterative closed loop at the control strategy level, so that the control strategy continues to converge to a better state as the control cycle accumulates.

[0088] In this embodiment, based on historical control residual records, targeted optimization is performed on the control variables, objective function, and constraints of the control law, covering the complete operable space of control law optimization. This allows the historical deviation knowledge accumulated in the residual memory to be directly transformed into precise improvements to the control law, achieving cross-cycle adaptive optimization of the control law.

[0089] The method of this application has been described above; the apparatus of this application will be described below.

[0090] See Figure 2 , Figure 2 This is a schematic diagram of a device control apparatus provided in an embodiment of this application, which is applied to smart devices, such as... Figure 2 As shown, the equipment control device 20 includes: The acquisition module 201 is used to acquire the first information of the current control cycle, the first information including the current status information or the current control residual identifier; The determining module 202 is used to determine the first control residual record from the residual memory module based on the first information; Optimization module 203 is used to optimize the control strategy of the smart device based on the first control residual record; The execution module 204 is used to perform control operations based on the optimized control strategy in the next control cycle.

[0091] It should be noted that the aforementioned equipment control device 20 can execute the equipment control method provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in the embodiments can be found in the aforementioned equipment control method provided in the embodiments of this application.

[0092] See Figure 3 , Figure 3 This is a schematic diagram of the structure of a smart device provided in an embodiment of this application. The smart device 30 includes a processor 301 and a memory 302. The memory 302 is connected to the processor 301, for example, via a bus.

[0093] Processor 301 is configured to support the smart device 30 in performing the corresponding functions in the methods described in the above method embodiments. Processor 301 may be a central processing unit (CPU), a network processor (NP), a hardware chip, or any combination thereof. The aforementioned hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The aforementioned PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0094] Memory 302 is used to store program code, etc. Memory 302 may include volatile memory (VM), such as random access memory (RAM); memory 302 may also include non-volatile memory (NVM), such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD); memory 302 may also include combinations of the above types of memory.

[0095] The memory 302 is used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the device control method in the embodiments of this application. The processor executes various functional applications and data processing of the device control method by running the non-volatile software programs, instructions, and modules stored in the memory, thereby realizing the functions of the device control method provided in the above-described method embodiments.

[0096] The memory 302 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and application programs required for at least one function. The data storage area may store data created based on the use of the 3D simulation circuit display device. In some embodiments, the memory may include memory remotely located relative to the processor, which can be connected to the 3D simulation circuit display device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0097] The one or more modules are stored in the memory. When executed by the one or more processors, they perform the device control method in any of the above method embodiments. For example, they perform the method steps described in the above method embodiments to realize the functions of the modules described in the above device embodiments.

[0098] This application also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the method described in the foregoing embodiments.

[0099] This application also provides a computer program product that, when run on a smart device, causes any of the above methods to be executed.

[0100] 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. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0101] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A device control method, characterized in that, Applied to smart devices, the method includes: Obtain the first information of the current control cycle, which includes the current status information or the current control residual identifier; Based on the first information, the first control residual record is determined from the residual memory module; Based on the first control residual record, the control strategy of the intelligent device is optimized; In the next control cycle, control operations are executed based on the optimized control strategy.

2. The method according to claim 1, characterized in that, Before obtaining the first information of the current control cycle, the method further includes: Generate a control input sequence based on the control strategy of the intelligent device; Based on the control input sequence, predict the first execution state sequence of the smart device; During the current control cycle, while performing control operations based on the control input sequence, the second execution state sequence of the intelligent device is acquired. The current control residual identifier is determined based on the first execution state sequence and the second execution state sequence.

3. The method according to claim 2, characterized in that, The step of determining the current control residual identifier based on the first execution state sequence and the second execution state sequence includes: The first execution state sequence and the second execution state sequence are aligned to obtain the third execution state sequence and the fourth execution state sequence. Determine the common segmentation point between the third execution state sequence and the fourth execution state sequence; Based on the common segmentation point, the third execution state sequence and the fourth execution state sequence are segmented respectively to obtain multiple sequence groups. Each sequence group includes a first sub-sequence segmented from the third execution state sequence and a second sub-sequence segmented from the fourth execution state sequence, and the first sub-sequence and the second sub-sequence are aligned. Determine the control residual identifier for each of the sequence groups, and use the determined control residual identifier as the current control residual identifier.

4. The method according to claim 3, characterized in that, For each of the sequence groups, a control residual identifier for the sequence group is determined, including: Based on the first subsequence and the second subsequence included in the sequence group, the control residual information of the sequence group is determined, and the control residual information includes state change sequence, control input sequence, state error change information, and constraint state change information; The control residual identifier is determined based on the control residual information.

5. The method according to claim 4, characterized in that, The step of determining the control residual identifier based on the control residual information includes: The control residual information is mapped into a vector representation; The vector representation is converted into a control residual identifier, which is used to represent a specific type of control deviation pattern.

6. The method according to claim 3, characterized in that, Determining the common segmentation point of the third execution state sequence and the fourth execution state sequence includes: Monitor the state change characteristics in the third execution state sequence or the fourth execution state sequence; When determining the existence of a target event based on the state change characteristics, the time node where the target event occurs is taken as the common segmentation point. The target event includes at least one of the following: contact state change event, control constraint change event, and dynamic characteristic change event.

7. The method according to any one of claims 2-6, characterized in that, After determining the current control residual identifier, the method further includes: The second control residual record is stored in the residual memory module. The second control residual record includes the control task information of the current control cycle, environmental change information, control execution result, and the current control residual identifier.

8. The method according to claim 1, characterized in that, The optimization of the control strategy for the intelligent device based on the first control residual record includes: Based on the first control residual record, the control law of the control strategy of the intelligent device is optimized; the optimization includes at least one of the following: modifying the control variables, adjusting the control objective function, and adjusting the control constraints.

9. A smart device, characterized in that, The device includes a memory and a processor, the memory being connected to the processor, the processor being configured to execute one or more computer programs stored in the memory, the processor causing the smart device to implement the method as described in any one of claims 1-8 when executing the one or more computer programs.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1-8.

Images