Mechanical wrist stiffness control method based on air pressure feedback

Abstract

The present application relates to the technical field of mechanical wrist control, and discloses a mechanical wrist variable stiffness control method based on air pressure feedback. The method comprises the following steps: acquiring real-time air pressure distribution data and joint angle data of a multi-air bag system of a mechanical wrist; calculating an instant air pressure feedback vector of each air bag region according to the difference between the real-time air pressure distribution data and a preset air pressure reference value; obtaining a local stiffness adjustment amount of each air bag region based on the instant air pressure feedback vector and the joint angle data, in combination with a predefined air bag region stiffness influence factor; when it is detected that the real-time moment load of the wrist exceeds a preset load threshold, fusing the local stiffness adjustment amount to generate a comprehensive target stiffness value of the wrist joint; evaluating the priority of the contribution of each air bag region to the target stiffness, and outputting an air bag state switching instruction according to the priority, so as to switch the air bag region with high priority to a high-pressure state and maintain the air bag region with low priority in a low-pressure or pressure-releasing state. The method enhances the operation stability and energy efficiency.

Description

Technical Field

[0001] This invention relates to the field of robotic wrist control technology, specifically a method for variable stiffness control of a robotic wrist based on pneumatic feedback. Background Technology

[0002] In fields such as industrial automation, medical rehabilitation, and service robots, the robotic wrist, as a key component connecting the end effector and the robotic arm, directly affects the overall system's performance through its flexibility and operational stability. With the continuous expansion of application scenarios, robotic wrists need to adapt to diverse operational requirements. They must maintain low stiffness to enhance flexibility during delicate operations, while possessing high stiffness to ensure stability when bearing loads or performing heavy tasks. Therefore, variable stiffness control has become one of the core requirements for the development of robotic wrist technology.

[0003] Existing methods for variable stiffness control in robotic wrists partially rely on the rigidity adjustment of mechanical structures, such as changing the physical parameters of elastic elements like springs and dampers to adjust stiffness. While these methods are structurally simple, they suffer from limited adjustment range and slow response speed. Furthermore, long-term friction of mechanical components can lead to wear, affecting control accuracy and equipment lifespan. Another common method is active adjustment based on motor drive, which changes joint stiffness by controlling the motor's output torque. However, motor drives often result in a significant increase in size and weight, hindering lightweight design of robotic wrists. Moreover, the lag in torque adjustment during dynamic load changes can lead to decreased system stability.

[0004] Pneumatic actuation, with its advantages of high output force / torque, good compliance, and low cost, is gradually being applied to variable stiffness control in robotic wrists. Multi-airbag systems, as a typical structure of pneumatic actuation, achieve stiffness adjustment by changing the inflation state of different airbags. However, existing multi-airbag control methods still have significant limitations: Firstly, they lack real-time and accurate sensing and feedback of airbag pressure, relying heavily on preset relationships between inflation volume and stiffness. When airbags leak, ambient temperature changes, or load fluctuates, the actual air pressure easily deviates from the preset value, leading to reduced stiffness adjustment accuracy. Secondly, there is insufficient coordinated control of the multi-airbag area, failing to dynamically allocate the stiffness contribution of each airbag according to joint angle and real-time load. Under high-load conditions, if the airbag pressure distribution is unreasonable, local overload or insufficient overall stiffness can easily occur, affecting the operational safety and reliability of the robotic wrist. In addition, existing methods have relatively simple stiffness adjustment strategies when the load exceeds the threshold, and mostly adopt the overall pressurization method. They fail to distinguish the contribution priority of different airbag regions to the target stiffness, which can easily lead to energy waste and make it difficult to achieve targeted stiffness enhancement. Summary of the Invention

[0005] The purpose of this invention is to provide a method for controlling the variable stiffness of a mechanical wrist based on pneumatic feedback, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a method for variable stiffness control of a mechanical wrist based on pneumatic feedback, the method comprising:

[0007] Acquire real-time air pressure distribution data and joint angle data of the multi-airbag system of the robotic wrist;

[0008] Based on the difference between the real-time air pressure distribution data and the preset air pressure reference value, calculate the instantaneous air pressure feedback vector of each airbag area;

[0009] Based on the real-time air pressure feedback vector and joint angle data, combined with the predefined airbag area stiffness influence factor, the local stiffness adjustment amount of each airbag area is calculated.

[0010] When the real-time torque load on the wrist exceeds the preset load threshold, the local stiffness adjustment of each airbag region is integrated to generate a comprehensive target stiffness value for the wrist joint.

[0011] Based on the comprehensive target stiffness value, the priority of each airbag region's contribution to the target stiffness is evaluated, and an airbag state switching command is output to the air pressure regulation system accordingly. The airbag state switching command is used to switch the high-priority airbag region to a high-pressure state and maintain the low-priority airbag region in a low-pressure or depressurized state.

[0012] Preferably, a pressure and torque coupling monitoring unit is deployed inside the pressure regulation system to perform fine-grained synchronous acquisition of pressure fluctuations and externally applied torque in each airbag zone of the robotic wrist. The pressure and torque coupling monitoring unit obtains real-time operating indicators from different airbag zones, including:

[0013] The current air pressure value inside each airbag, the change in air pressure gradient between adjacent airbags, the real-time torque load of the wrist joint and its direction vector.

[0014] Preferably, after collecting the real-time operating indicators, a pressure-torque coupling feature set is formed, and airbag partition stiffness calculation rules are defined to quantify the stiffness support capability that each airbag partition can provide under the current pressure configuration.

[0015] Based on the airbag zonal stiffness calculation rule, a short-term prediction is made on the selected air pressure distribution sequence, and the predicted air pressure stiffness sequence is output.

[0016] Preferably, by fusing the collected real-time operating indicators with the predicted air pressure stiffness sequence, an airbag contribution coefficient is constructed to achieve dynamic sorting of the variable stiffness operation of each airbag zone.

[0017] The stiffness adjustment priority of each airbag partition is quantified and assigned using the airbag contribution coefficient, and all airbag partitions are arranged from high to low priority according to the airbag contribution coefficient to form an airbag partition scheduling sequence table.

[0018] Preferably, based on the current available air pressure capacity of the air pressure regulation system and the real-time torque load, air pressure resources are allocated or reserved for airbag zones with higher priority from top to bottom:

[0019] If the system capacity is insufficient when allocating air pressure resources to a certain airbag zone, it is determined whether the pressurization operation of that zone needs to be delayed or a gradient pressurization strategy should be adopted.

[0020] During the high torque load phase, the pressurization requests of low-priority airbag partitions are cached in an adaptive queue. When the real-time torque load or its predicted value is detected to be lower than a set threshold, the pressurization requests in the adaptive queue are released and a new round of air pressure resource allocation is executed.

[0021] Preferably, a stiffness maintenance value function is defined to measure the necessity of maintaining a high-pressure state in the airbag zoning.

[0022] When the stiffness maintenance value function exceeds a preset value threshold, the corresponding airbag section is maintained in a high-pressure state; otherwise, it can be converted to a low-pressure or depressurized state to release air pressure resources and make room for higher-priority airbag sections.

[0023] Preferably, if it is determined that a certain airbag section can be converted to a low-pressure or depressurized state, the depressurization logic of that airbag section is triggered and the released air pressure capacity is registered in the air pressure resource pool.

[0024] Based on the real-time torque load and the high-priority airbag partition list indicated by the airbag partition scheduling sequence table, air pressure capacity is reserved in advance for airbag partitions that will face high torque load in subsequent periods.

[0025] If the current available capacity of the air pressure resource pool is insufficient to meet the reservation requirements of all candidate airbag partitions, then the allocation will be re-sorted and distributed based on the airbag contribution coefficient.

[0026] Preferably, an airbag state switching overhead is defined and embedded into a pressure adjustment delay function for dynamically adjusting the high-pressure / low-pressure state switching strategy, and a delayed confirmation step is introduced into the pressure state switching queue.

[0027] When multiple airbag zones share a portion of the base pressure pipeline or air source, only the differentiated pressure configuration of each zone is adjusted during the execution state switch.

[0028] By comparing the current status of the air pressure resource pool with that of the shared air pressure pipeline, only the portion of air pressure capacity that differs from that of the shared pipeline is recovered or allocated.

[0029] Preferably, the air pressure control parameters and associated piping structures of each airbag zone are divided into blocks to form several air pressure control units;

[0030] Define a similarity function for air pressure parameters. When the calculated value of the similarity function for air pressure parameters exceeds a set value, the two air pressure control units are considered as potentially shareable units.

[0031] Deduplication is performed on air pressure control units that meet the sharing conditions, and a unified index is established in the shared air pressure parameter library;

[0032] When an airbag zone needs to be depressurized or pressurized, the system checks whether its air pressure control unit is registered in the shared air pressure parameter library to determine the specific differential air pressure parameters that need to be adjusted.

[0033] Preferably, the air pressure configuration of each airbag zone is recorded as a set of parameter segments. If some parameter segments match the shared units in the shared air pressure parameter library, the shared parameters are directly referenced, while the remaining differentiated parameter segments need to be adjusted independently.

[0034] If a shared pressure parameter unit is not referenced by any high-pressure airbag partition for a long time, a cleanup operation for that shared unit is triggered.

[0035] Based on the latest air pressure distribution requirements and sharing status, the mapping relationship between the air pressure parameter segments and the sharing units is periodically updated, and a dynamic air pressure control spectrum is output to the air pressure regulation system.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] This pressure feedback-based variable stiffness control method for robotic wrists effectively improves the performance of robotic wrists under diverse working conditions by constructing a closed-loop control mechanism that integrates real-time pressure sensing and dynamic stiffness adjustment. Its core advantages lie in its refined control and adaptive load adjustment capabilities for multi-airbag systems.

[0038] By acquiring real-time air pressure distribution and joint angle data of the multi-airbag system, precise raw data is provided for stiffness adjustment. Unlike traditional control methods that rely on preset parameters, this real-time sensing mechanism can promptly capture subtle changes in airbag pressure, including pressure deviations caused by air leakage, temperature fluctuations, or load disturbances. This ensures that subsequent stiffness adjustments are based on actual working conditions, avoiding a decrease in control accuracy due to parameter mismatch.

[0039] By calculating local stiffness adjustments based on real-time air pressure feedback vectors and joint angle data, targeted adjustments to each airbag region are achieved. Different airbags play different mechanical roles during robotic wrist movement. By introducing predefined stiffness influencing factors, the required stiffness adjustment range can be determined based on the actual air pressure state and joint movement posture of each airbag. This avoids the problems of excessive or insufficient local stiffness that may occur in traditional overall adjustment methods, making the stiffness distribution more aligned with the mechanical needs of the robotic wrist.

[0040] When the real-time torque load exceeds a preset threshold, a comprehensive target stiffness value is generated by integrating local stiffness adjustments. Airbag state switching commands are then output based on contribution priority, significantly enhancing the robotic wrist's adaptability to high-load conditions. Under high load conditions, not all airbags need to be in a high-pressure state. By evaluating the contribution priority of each airbag to the target stiffness, high-priority airbags can be switched to high pressure to provide primary stiffness support, while low-priority airbags remain in a low-pressure or depressurized state. This ensures that the overall stiffness meets load requirements, reduces unnecessary energy consumption, and avoids system redundancy and response lag caused by nationwide high pressure.

[0041] This method achieves stiffness variation through pneumatic pressure regulation, offering higher compliance and lower mechanical wear compared to mechanical structure or motor-driven adjustments. The dynamic response characteristics of the pneumatic system can quickly follow stiffness adjustment commands, enabling the robotic wrist to rapidly adjust its stiffness state during sudden load changes, thus improving operational stability. Simultaneously, the non-rigid contact adjustment method reduces frictional losses between components, helping to extend the robotic wrist's lifespan and lower maintenance costs.

[0042] This control method is highly adaptable. By adjusting the preset air pressure reference value, load threshold, and stiffness influence factor, it can be adapted to multi-airbag mechanical wrists with different structures and diverse operating scenarios. Whether it is the low-stiffness and flexible state required for fine operation or the high-stiffness and stable state required for heavy-duty operation, it can achieve precise control through a unified control logic, thereby improving the versatility and practical value of the technology. Attached Figure Description

[0043] Figure 1 This is a schematic diagram illustrating the working principle of the pneumatic feedback-based variable stiffness control method for mechanical wrists described in this invention.

[0044] Figure 2 This is a flowchart of the pressure-torque coupling feature set and stiffness prediction.

[0045] Figure 3 A flowchart for air pressure resource allocation and priority scheduling;

[0046] Figure 4 This is a flowchart for sharing and managing parameters of the air pressure control unit. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Please see Figure 1 This invention provides a method for variable stiffness control of a mechanical wrist based on pneumatic feedback, the method comprising:

[0049] Adaptive adjustment of wrist joint stiffness is achieved through dynamic air pressure regulation of a multi-airbag system. This method first acquires real-time air pressure distribution data and joint angle data of the multi-airbag system in the robotic wrist. Based on the difference between the air pressure distribution and a preset benchmark value, the instantaneous air pressure feedback vector for each airbag region is calculated. Combining the joint angle data with a predefined airbag region stiffness influence factor, the local stiffness adjustment amount is calculated. When the real-time torque load on the wrist exceeds a preset threshold, the local stiffness adjustment amounts of each region are fused to generate a comprehensive target stiffness value. The contribution priority of each airbag region to the target stiffness is evaluated, and an airbag state switching command is output to the air pressure regulation system. High-priority airbag regions are prioritized for switching to a high-pressure state, while maintaining or reducing the air pressure state of low-priority regions.

[0050] Example 1: See Figure 2 This involves the deployment, operation, and subsequent data processing of a pressure and torque coupling monitoring unit. This monitoring unit, as an embedded component of the pressure regulation system, is directly connected to the multi-airbag zone network of the robotic wrist. Its physical structure includes an array-type pressure sensor network, a three-axis torque sensor interface, and a high-speed data acquisition circuit. The pressure sensors are distributed in a star topology, with at least two redundant sensor nodes installed within each airbag zone, and additional cross-boundary monitoring points deployed at the boundaries of adjacent zones. The three-axis torque sensor is integrated into the wrist joint drive axis, and its sampling frequency is synchronized with the pressure sensing system.

[0051] The real-time operational metrics acquisition process employs a time-division multiplexing mechanism. A complete acquisition cycle is executed every 10 milliseconds: first, the raw voltage signals from all pressure sensors are read, converted by an ADC, and mapped to engineering values ​​ranging from 0 to 500 kPa; then, the pressure gradient change between adjacent airbag zones is calculated, obtained by dividing the pressure difference by the Euclidean distance between the zone centers; simultaneously, the three-dimensional component data of the joint torque load are acquired, and the measured values ​​are transformed to the wrist base coordinate system through coordinate transformation. All raw data are timestamped, achieving microsecond-level accuracy.

[0052] The pressure-torque coupling feature set is constructed based on timestamp alignment. The data processor extracts pressure sensor data from three consecutive sampling periods and integrates them into a structured time series according to partition number. The feature vector of each partition includes: the pressure value at the current sampling point, the pressure change trend value of the previous two historical sampling points, the maximum pressure gradient value with adjacent partitions, and the pressure fluctuation variance of this partition in three axes. Torque features are recorded in the form of quadruplets: instantaneous torque magnitude, azimuth and pitch angles of the torque vector in the base coordinate system, and the rate of change of torque in adjacent sampling periods. The feature set storage adopts a sliding buffer design with a fixed buffer depth of 50 samples, and the oldest sample is automatically discarded when new data is input.

[0053] The definition process of the airbag zone stiffness calculation rules references the principles of mechanics of materials. The system establishes a mapping relationship model between airbag deformation and air pressure, and uses joint angle data to inversely deduce the airbag wall deformation constraints. For each airbag zone, the stiffness support capacity is characterized by the coupling coefficient between the torque change and the air pressure change, which is dynamically fitted by a linear regression model. Specifically, the processor uses the current zone air pressure as the independent variable and the joint support torque component contributed by this zone as the dependent variable, calculating the least squares solution within a sliding window of size 100. The slope obtained from the fitting is the real-time stiffness characteristic value of that zone.

[0054] The generation of the predicted air pressure stiffness sequence relies on a two-layer recurrent network model. The short-term prediction module receives the air pressure distribution sequence within the latest two seconds, which includes historical air pressure data for each zone and the corresponding stiffness characteristic values. During data preprocessing, standardization and outlier filtering are performed, and the normalized data is divided into independent subsequences according to the zone index. The core prediction uses an autoregressive moving average model, with parameters trained online based on the most recent 50 sampled data from each zone. An adaptive learning rate mechanism is introduced during training; if the difference between two adjacent prediction results exceeds a threshold, the model weights are dynamically adjusted by decay. The prediction output includes air pressure stiffness values ​​for three consecutive time steps within the next 30 milliseconds. The data structure is a three-dimensional tensor: the first dimension corresponds to the airbag zone number, the second dimension represents the prediction time step index, and the third dimension contains two entries: the predicted air pressure value and the predicted stiffness value.

[0055] The confidence assessment of the prediction results is performed concurrently. The system calculates the root mean square error between the 10 most recent historical predictions and actual measurements. When the error exceeds a set upper limit, a model retraining process is triggered. During retraining, the historical average value is used instead of the predicted value output until the new model converges. The predicted pressure stiffness sequence is finally encapsulated synchronously with real-time operating indicators and transmitted to downstream processing units via a high-speed bus. The entire process implements hard real-time constraints, controlling the total latency from data acquisition to prediction result output to within 15 milliseconds.

[0056] Example 2: See Figure 3This study focuses on real-time data fusion and dynamic allocation mechanisms for air pressure resources. The processing unit receives a predicted air pressure stiffness sequence and a real-time operational index data packet from Example 1. The data packet includes the air pressure value of each airbag zone at the current moment, the change in air pressure gradient in adjacent areas, and the joint moment load component in Newton-meters. The fusion processing module first performs timestamp calibration, aligning the predicted sequence and real-time data to the same time reference point. The aligned dataset is input into the weighted calculation core, which executes the contribution coefficient generation logic. The contribution coefficient of each airbag zone is determined by two key factors: the azimuth matching degree between the moment load direction and the geometric center of the airbag, and the support strength of the current air pressure change trend for the overall stiffness. The azimuth matching degree is defined as the cosine of the angle between the moment vector and the zone normal vector, and the air pressure change trend is taken from the stiffness change rate of the nearest time step in the predicted sequence. The two factors are synthesized into a single scalar according to the following relationship:

[0057] ;

[0058] In the formula: Indicates the first Contribution coefficient of each airbag zone; This is the azimuth weighting factor, with a default configuration of 0.6; It is the spatial angle between the torque vector and the partition normal vector; It is the stiffness trend coefficient, initialized to 0.4; This represents the magnitude of stiffness change in that partition during the prediction period; This is the maximum allowable stiffness variation preset for the system. All calculations are performed on a dedicated hardware coprocessor, with the coefficient matrix refreshed every 8 milliseconds.

[0059] After coefficient calculation, the normalization engine projects the original coefficients to the 0-1 interval. The projection process employs a non-uniform scaling strategy, using logarithmic compression for partitions with coefficients 30% higher than the historical average and linear mapping for partitions below the average. The resulting data is written to a priority sorting buffer, using a modified radix sort algorithm, processing 128 partitions in less than 1.2 milliseconds. The generated airbag partition scheduling sequence list is stored using a doubly linked data structure, with each node simultaneously recording the original contribution value, normalized value, and partition physical address pointer. A sequence list maintenance thread continuously monitors the list's integrity, automatically reorganizing the sorted sequence upon detecting newly added or removed partition nodes.

[0060] The air pressure resource allocation subsystem comprises a capacity monitor and an allocation decision engine. The capacity monitor scans the real-time available capacity of the air pressure supply network every 5 milliseconds, dividing the total system capacity into 256 discrete allocation units. When the joint torque load detection value exceeds a preset threshold, the allocation decision engine activates a high-priority allocation mode. In this mode, the engine sequentially reads the scheduling sequence head node and executes dynamic allocation logic: it checks whether the current remaining capacity meets the target air pressure increment for the partition; if so, it directly generates a pressurization command. If the capacity is insufficient but the partition contribution coefficient is higher than the critical value of 0.72, a three-stage gradient pressurization procedure is initiated: the first stage allocates 50% of the target increment, the second stage allocates 30%, and the third stage allocates the remaining 20%, with each stage spaced twice the valve response time. For partition requests with a contribution coefficient lower than 0.72, the system generates a delayed allocation flag and injects it into an adaptive queue.

[0061] The adaptive queue design employs a priority-time dual-dimensional management strategy. Queue members include three main fields: request timestamp, original priority score, and air pressure demand value. The queue release mechanism is triggered by torque load monitoring signals: when the real-time torque load value is below 80% of the threshold benchmark for three consecutive samples, or when the short-term prediction module from Example 1 outputs a downward trend signal, the queue processor initiates the release process. During the release process, the system re-verifies the currently available capacity and attempts to allocate resources starting from the head of the queue. Specifically, when the system detects a significant decrease in the air pressure pipeline network load fluctuation rate, it allows merging partition requests from adjacent physical locations into batch operations, reducing the number of valve openings and closings through pipeline paralleling technology.

[0062] The resource allocation module incorporates a dynamic rollback mechanism to aid decision-making. After a high-pressure partition completes its mission, the system assesses the necessity of its continued occupancy based on the latest sequence list. When a higher-priority partition is detected as being unable to activate due to insufficient capacity, the resource reclaimer automatically calculates the benefits of releasing the pressure resources of that partition. The reclamation operation employs a tiered release strategy: rigid connection areas undergo segmented depressurization, with each reduction not exceeding 25% of the original pressure; flexible connection areas are allowed to undergo single depressurization down to the baseline maintenance pressure. The released capacity is immediately transferred to the allocable resource pool, prioritizing the needs of the first partition in the current sequence list awaiting activation. The entire process is physically executed within 17 milliseconds, during which the system synchronously updates the capacity distribution map to the shared memory area.

[0063] Example 3

[0064] Example 3 involves a stiffness maintenance decision-making and air pressure resource recycling mechanism. After receiving the airbag zoning scheduling sequence list generated in Example 2, the system initiates a value assessment process. The stiffness maintenance value function for each zoning zone consists of a vector projection and a contribution factor, and its calculation formula is expressed as follows:

[0065] ;

[0066] in: Represents the real-time value assessment of the i-th airbag partition; It is the current three-dimensional torque load vector of the wrist joint; This represents the normalized version of the airbag partition normal vector; The contribution coefficient is derived from that calculated in Example 2. The calculation result is processed by a 32-bit floating-point coprocessor and then sent to a comparator unit, which compares the output value with a preset threshold parameter of 0.5. When the value assessment value is higher than the threshold for three consecutive sampling periods, the corresponding partition enters a high-pressure maintenance state; conversely, when any two consecutive sampling values ​​are lower than the threshold, the system triggers a pressure relief state transition protocol.

[0067] The pressure relief operation engine includes two mode switching logics. Mode selection is based on the absolute difference between the current zone pressure and the target baseline: if the difference exceeds a 50kPa threshold, a high-speed pressure relief mode is activated, opening the direct-flow electromagnetic pressure relief valve and using a vortex reducer to control the airflow speed; otherwise, a gradual pressure relief mode is entered, executing staged pressure relief through a proportional regulating valve. The stage division adopts a proportionally decreasing strategy, with each pressure reduction step set at 20% of the current pressure value, and adjacent steps maintained at an interval of three times the valve response time. The pressure release process is monitored in real time; when the actual pressure reduction curve deviates from the expected trajectory by more than 10%, a dynamic compensation airflow is injected.

[0068] The air pressure resource pool maintenance structure adopts a paged memory design, with each page recording three fields: the timestamp of the released capacity, the partition identifier, and the air pressure value. The pool manager performs four core operations: the new capacity registration module receives the confirmation signal of depressurization completion and updates the idle capacity counter; the capacity reservation module scans the top 15% of high-priority partitions in the scheduling sequence table based on the real-time torque load vector direction; the reservation request processor assigns virtual capacity tags to candidate partitions, which contain three attributes: partition number, reserved air pressure, and validity period. When the resource contention arbitrator detects multiple partitions requesting reservations, it performs a secondary sorting based on contribution coefficients, with the selection rule being: priority is given to fulfilling these requests. For partitions with a priority greater than 0.85, the remaining capacity is allocated to the next lower priority partitions according to the sorting coefficient ratio.

[0069] The physical implementation of reserved capacity employs spatial time-sharing multiplexing technology. When a target partition reaches the activation window, the reservation manager performs three operations: unlocking the virtual capacity marker; submitting a formal allocation request to the resource pool; and starting the partition boosting timer. If no activation instruction is received within the reservation validity period, the system automatically releases the reservation status and reclaims the capacity. The real-time guarantee mechanism for boosting response employs a two-level timeout handling: the basic timeout threshold is 100 milliseconds; if boosting is not completed within the timeout, a pressure compensation pump is activated; when the timeout reaches 300 milliseconds, boosting is considered a failure, and the existing capacity of the partition is re-registered and released.

[0070] The dynamic capacity reallocation strategy continues to operate during the high-voltage maintenance phase. The value assessment function is recalculated every 8 milliseconds when a fault is detected. When the value drops below 0.4 and other partitions have resources waiting, a preemptive capacity transfer is triggered. The transfer process includes three sub-processes: partition status freezing, pressure release, and target partition preheating. The release process performs stepped pressure reduction, with each reduction being 15% of the current value. The pressure reduction interval is dynamically adjusted according to the system load: a fixed 20-millisecond interval is used under high torque load conditions, while an adaptive interval algorithm is enabled under low load conditions to match the pressure reduction rate with the time required by the target partition.

[0071] The resource pool anomaly handling unit monitors two conflict scenarios: in the case of overlapping capacity allocation, timestamp priority is used for adjudication, and subsequent requests are automatically queued; in the case of capacity leakage, a three-stage recovery protocol is initiated, first freezing the operation permissions of the suspected partition, then executing the capacity audit process, and finally supplementing the missing capacity through the backup electric pump. All operation records are written to non-volatile memory, and the data packet contains four dimensions of information: operation type code, partition number, timestamp, and air pressure value change, forming a complete air pressure resource cycle tracking chain.

[0072] Example 4: See Figure 4 The specific operations revolve around optimizing air pressure state switching and managing shared pipelines. When the stiffness requirements of the A7 airbag zone in the robotic wrist change, the system initiates a switching process. This zone connects the main air supply pipeline PL-207 and the auxiliary regulating pipeline AL-49, with PL-207 shared with three adjacent zones. The state switching cost assessment module first searches the pipeline topology database and extracts the following parameters: PL-207 pipeline length 1.2 meters, AL-49 length 0.3 meters, average solenoid valve response time 12 milliseconds, and current air pressure difference 85 kPa. Based on these parameters, the delay prediction algorithm outputs a switching time cost of 58 milliseconds, triggering a step-by-step execution strategy.

[0073] The pressure adjustment delay function decomposes the operation into three stages: the first stage adjusts the independent parameters of the AL-49 pipeline, taking 8 milliseconds; the second stage coordinates the pressure balance of the shared pipeline PL-207, taking 35 milliseconds; and the third stage fine-tunes the final pressure value of the A7 partition, taking 15 milliseconds. After each stage is completed, the verification module checks the deviation between the actual delay and the estimated value. When the deviation exceeds 15%, the time allocation of subsequent steps is dynamically adjusted. The state coordination of the shared pipeline PL-207 involves special handling; the system only modifies the opening parameters of the regulating valve specific to the A7 partition, keeping the base pressure value unchanged. Adjacent partitions B3 and C5 enter a parameter freeze state during this period to avoid cross-interference. The block partitioning engine performs cluster analysis on the relevant parameters of the A7 partition to generate a configuration fragment table, see Table 1.

[0074] Table 1: Configuration Fragment Table.

[0075]

[0076] The similarity calculation module compared the parameter configurations of A7 with those of adjacent partitions and found that the base air pressure and safety threshold perfectly matched the S-7721 unit in the shared library. The system automatically established a reference relationship, skipping the repeated adjustment of these two items in subsequent operations. The deduplication processor scanned all associated partitions, reducing the six copies of shared parameters in memory to a single primary copy, and achieving data synchronization through pointer mapping.

[0077] When zone A7 needs to switch from high pressure to low pressure, the differential adjustment module performs precise control. The pressure relief command targets only the control valve opening parameter unique to this zone, reducing it from 68% to 32% while maintaining the stable base pressure value of the shared pipeline. The released capacity is calculated by the resource reclaimer, only including the portion of the pressure resource independently occupied by this zone (approximately 35 kPa); the shared portion is not included in the capacity statistics. During the reclamation process, the pipeline status monitor continuously tracks the fluctuations of PL-207 to ensure that the shared parameters are not affected.

[0078] The shared unit cleanup daemon periodically checks the usage status of the S-7721 unit. If no partition references the unit for three consecutive maintenance cycles (5 seconds per cycle), the system initiates a cleanup preparation process: first, it verifies the current status of all associated partitions to confirm there are no ongoing adjustments; then, it backs up the parameter history to non-volatile memory; finally, it releases the memory space occupied by the shared unit. If a new partition requests to reference the shared unit before cleanup is complete, the cleanup process is immediately terminated and access permissions are restored.

[0079] The dynamic pressure control map update mechanism is triggered after each state transition. The map generator collects the following information: a list of currently active shared unit IDs, version numbers of independent parameters for each zone, and pipeline topology change markers. For the current operation of zone A7, the map recorded two types of updates: the version of the control valve opening parameter was upgraded from v3.1 to v3.2; the reference counter of shared unit S-7721 was decremented. These changes are transmitted to all relevant nodes via differential encoding, updating only the changed parts rather than retransmitting the entire data.

[0080] The pipeline conflict resolution mechanism handles typical scenarios in actual operation. When partition A7 is performing a depressurization operation, if the adjacent partition C5 suddenly needs to be pressurized, the conflict detector identifies that both partitions share the PL-207 pipeline. The arbitration logic immediately suspends the depressurization process of A7 and prioritizes the pressurization request of C5. After C5 completes the baseline pressure adjustment, the system recalculates the remaining depressurization amount for A7 and uses a gentler gradient to complete subsequent operations. The entire process is recorded in the operation log, including detailed information such as the pause time, conflict type code, and final completion status.

[0081] State switching queue management employs a priority interpolation algorithm. When switching requests from multiple partitions arrive simultaneously, the queue manager considers not only preset priorities but also analyzes the shared resource usage of each request. For requests involving the same shared pipeline, the system automatically staggers their execution time windows to minimize mutual interference. Each request is assigned a time tolerance parameter, allowing for flexible adjustment of execution time within a specified time range, thereby improving overall pipeline utilization.

[0082] The parameter version control system ensures synchronized updates of shared and private configurations. Each time an individual parameter is modified, such as changing the control valve response curve of partition A7 from linear to accelerated mode, the system automatically generates a new version number and updates local storage. Simultaneously, the parameter fingerprint information for that partition is registered with the shared parameter library for subsequent similarity calculations. When configuration changes in other partitions cause the similarity to exceed the 0.85 threshold, the system prompts that a new shared unit may be created, which is then confirmed by an engineer before parameter merging is performed.

[0083] Example 5: Built on a dynamic control system based on a shared parameter library. The air pressure configuration of each airbag partition within the system is stored in a fragmented data structure. The fragment linked list contains two types of entities: shared parameter reference identifiers marked as read-only, with their storage area pointing to the registered unit address in the shared library; and independent parameter entities residing in the partition's local cache, recording complete air pressure settings, tolerance ranges, and adjustment history. A lazy initialization strategy is adopted during the configuration loading phase. When a partition is first activated, the parameter matching engine scans the linked list structure. The matching process checks the compatibility between local parameter items and the shared library. If compatible items account for more than 80% of the total, the engine only loads the remaining non-shared parameters into execution memory, while the shared parts achieve zero-copy access through pointer mapping.

[0084] The shared unit validity daemon runs periodically, with a period of three seconds. The process first retrieves the shared unit access records of all currently high-pressure partitions, generating a snapshot with a recent access timestamp. Snapshot analysis uses a sliding window detection method: for shared units that haven't been referenced by any active partitions for more than sixty seconds, the daemon moves them to a cleanup queue and sets a countdown flag. During the cleanup preparation phase, secondary verification is performed, sending verification queries to all potential partitions that reference the unit, with a query response window of 100 milliseconds. If any partition reports a dependency within the window period, the cleanup plan for that unit is immediately canceled; units that do not receive a response release their memory resources after the countdown ends, and the shared library directory index is updated.

[0085] The core generation process of the dynamic pressure control map is triggered by the torque load change rate. The monitor continuously calculates the standard deviation of joint torques within the last 500 milliseconds, and the map update frequency automatically increases when the change rate exceeds a set threshold. The generation module executes five operation steps in sequence: reconstructing the mapping relationship table between parameter fragments and shared units; recalculating the resource pool capacity distribution topology; optimizing the gas supply network structure model; compiling the partition state change difference set; and encapsulating the output triple instruction sequence. The mapping relationship reconstruction stage adopts the proximity similarity principle to rematch the current parameter configuration with shared library entries. If the similarity between the new partition parameters and shared units is found to be above 95%, a reference relationship is automatically established; if the similarity drops below 60%, the association is terminated.

[0086] The output instruction triplet uses fixed-byte encoding: the first field stores the shared unit identifier, occupying two bytes of unsigned integer; the second field identifies the storage location of independent parameters, containing the memory page number and offset address, totaling four bytes; the last field records the relative adjustment range, expressed as an 8-bit signed integer representing -128 to 127 barometric pressure adjustment units. The transmission protocol is optimized for high-frequency update scenarios, allowing incremental transmission of differences. When the map update frequency reaches the upper limit of 10 Hz, a compression algorithm is enabled to remove duplicate triplet entries in consecutive frames, achieving a compression redundancy ratio of up to 70%.

[0087] The parameter matching engine maintains a dedicated cache to handle frequent queries. The cache structure employs a three-tier design: the bottom layer stores basic parameter templates, the middle layer caches recently active shared unit configurations, and the top layer stores currently detected independent parameter fragments. Matching operations prioritize similarity comparisons between the top and middle layers, accelerating the core algorithm's efficiency. After each matching cycle, the engine automatically promotes frequently accessed shared unit configurations to the middle cache layer, while entries not referenced in the last three cycles are demoted to the bottom storage area. Cache hit rate logs serve as system optimization metrics, allowing the operations module to dynamically adjust the capacity allocation of the three-tier structure.

[0088] The shared resource repository cleanup mechanism includes a conflict prevention system. When a shared unit enters the cleanup countdown phase, the repository manager immediately locks access permissions for that unit. If a new partition requests a reference at this time, the system provides two response options: for urgent requests under high pressure, temporary access is allowed and the partition is marked as pending; for regular requests, the partition is guided to directly copy the unit's content and convert it into an independent configuration. The historical version tracker records all cleanup records, including the cleanup time, unit identifier, number of referenced partitions, etc. These records serve as a reference benchmark when creating new shared units, preventing the repeated creation and destruction of homogeneous shared units.

[0089] The physical execution phase of the dynamic control map employs a tiered verification strategy. After the instruction set is transmitted to the regional execution unit, simulation verification is first performed: the execution result is estimated based on the current pipeline status, and boundary safety is checked. If the verification passes, the physical execution logic is triggered; if the predicted pressure change exceeds the safety threshold, the execution unit automatically switches to step mode to complete the operation in segments. The execution process is accompanied by real-time status feedback, collecting the actual pressure change trajectory every five milliseconds and comparing it with the expected instruction. When the deviation value continues to accumulate and exceeds the tolerance limit, the system records an exception code and suspends subsequent instruction processing, resuming operation only after manual intervention and confirmation.

[0090] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0091] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for variable stiffness control of a mechanical wrist based on pneumatic feedback, characterized in that, include: Acquire real-time air pressure distribution data and joint angle data of the multi-airbag system of the robotic wrist; Based on the difference between the real-time air pressure distribution data and the preset air pressure reference value, calculate the instantaneous air pressure feedback vector of each airbag area; Based on the real-time air pressure feedback vector and joint angle data, combined with the predefined airbag area stiffness influence factor, the local stiffness adjustment amount of each airbag area is calculated. When the real-time torque load on the wrist exceeds the preset load threshold, the local stiffness adjustment of each airbag region is integrated to generate a comprehensive target stiffness value for the wrist joint. Based on the comprehensive target stiffness value, the priority of each airbag region's contribution to the target stiffness is evaluated, and an airbag state switching command is output to the air pressure regulation system accordingly. The airbag state switching command is used to switch the high-priority airbag region to a high-pressure state and maintain the low-priority airbag region in a low-pressure or depressurized state.

2. The method for variable stiffness control of a mechanical wrist based on pneumatic feedback according to claim 1, characterized in that, An air pressure and torque coupling monitoring unit is deployed within the air pressure regulation system to perform fine-grained synchronous acquisition of pressure fluctuations and externally applied torque in each airbag zone of the robotic wrist. The air pressure and torque coupling monitoring unit obtains real-time operating indicators from different airbag zones, including: The current air pressure value inside each airbag, the change in air pressure gradient between adjacent airbags, the real-time torque load of the wrist joint and its direction vector.

3. The method for variable stiffness control of a mechanical wrist based on pneumatic feedback according to claim 2, characterized in that, After collecting the real-time operating indicators, a pressure-torque coupling feature set is formed, and the airbag partition stiffness calculation rules are defined to quantify the stiffness support capability that each airbag partition can provide under the current pressure configuration. Based on the airbag zonal stiffness calculation rule, a short-term prediction is made on the selected air pressure distribution sequence, and the predicted air pressure stiffness sequence is output.

4. The method for variable stiffness control of a mechanical wrist based on pneumatic feedback according to claim 3, characterized in that, By fusing the collected real-time operating indicators with the predicted air pressure stiffness sequence, an airbag contribution coefficient is constructed to achieve dynamic sorting of the variable stiffness operation of each airbag zone. The stiffness adjustment priority of each airbag partition is quantified and assigned using the airbag contribution coefficient, and all airbag partitions are arranged from high to low priority according to the airbag contribution coefficient to form an airbag partition scheduling sequence table.

5. The method for variable stiffness control of a mechanical wrist based on pneumatic feedback according to claim 4, characterized in that, Based on the current available air pressure capacity of the air pressure regulation system and the real-time torque load, air pressure resources are allocated or reserved for airbag zones with higher priority from top to bottom: If the system capacity is insufficient when allocating air pressure resources to a certain airbag zone, it is determined whether the pressurization operation of that zone needs to be delayed or a gradient pressurization strategy should be adopted. During the high torque load phase, the pressurization requests of low-priority airbag partitions are cached in an adaptive queue. When the real-time torque load or its predicted value is detected to be lower than a set threshold, the pressurization requests in the adaptive queue are released and a new round of air pressure resource allocation is executed.

6. The method for variable stiffness control of a mechanical wrist based on pneumatic feedback according to claim 5, characterized in that, Define a stiffness maintenance value function that measures the necessity of maintaining a high-pressure state in airbag zones. When the stiffness maintenance value function exceeds a preset value threshold, the corresponding airbag section is maintained in a high-pressure state; otherwise, it is converted to a low-pressure or depressurized state to release air pressure resources and make room for higher-priority airbag sections.

7. The method for variable stiffness control of a mechanical wrist based on pneumatic feedback according to claim 6, characterized in that, If it is determined that a certain airbag section is converted to a low-pressure or depressurized state, the depressurization logic of that airbag section is triggered and the released air pressure capacity is registered in the air pressure resource pool. Based on the real-time torque load and the high-priority airbag partition list indicated by the airbag partition scheduling sequence table, air pressure capacity is reserved in advance for airbag partitions that will face high torque load in subsequent periods. If the current available capacity of the air pressure resource pool is insufficient to meet the reservation requirements of all candidate airbag partitions, then the allocation will be re-sorted and distributed based on the airbag contribution coefficient.

8. The method for variable stiffness control of a mechanical wrist based on pneumatic feedback according to claim 7, characterized in that, Define the airbag state switching overhead and embed it into the air pressure adjustment delay function to dynamically adjust the high-pressure / low-pressure state switching strategy, and introduce a delayed confirmation step in the air pressure state switching queue. When multiple airbag zones share a portion of the base pressure pipeline or air source, only the differentiated pressure configuration of each zone is adjusted during the execution state switch. By comparing the current status of the air pressure resource pool with that of the shared air pressure pipeline, only the portion of air pressure capacity that differs from that of the shared pipeline is recovered or allocated.

9. The method for variable stiffness control of a mechanical wrist based on pneumatic feedback according to claim 8, characterized in that, The air pressure control parameters and associated piping structures of each airbag zone are divided into blocks to form several air pressure control units; Define a similarity function for air pressure parameters. When the calculated value of the similarity function for air pressure parameters exceeds a set value, the two air pressure control units are considered as potentially shareable units. Deduplication is performed on air pressure control units that meet the sharing conditions, and a unified index is established in the shared air pressure parameter library; When an airbag zone needs to be depressurized or pressurized, the system checks whether its air pressure control unit is registered in the shared air pressure parameter library to determine the specific differential air pressure parameters that need to be adjusted.

10. The method for variable stiffness control of a mechanical wrist based on pneumatic feedback according to claim 9, characterized in that, The air pressure configuration of each airbag zone is recorded as a set of parameter segments. Some parameter segments match the shared units in the shared air pressure parameter library and are directly referenced as shared parameters. The remaining differential parameter segments need to be adjusted independently. If a shared pressure parameter unit is not referenced by any high-pressure airbag partition for a long time, a cleanup operation for that shared unit is triggered. Based on the latest air pressure distribution requirements and sharing status, the mapping relationship between the air pressure parameter segments and the sharing units is periodically updated, and a dynamic air pressure control spectrum is output to the air pressure regulation system.

Images