Robot operation arm constraint safety system and method based on communication perception

By acquiring communication parameters and the angular velocity and control torque of the robot arm in real time, calculating the network score and switching to the local TDE prediction mode, and adjusting the control parameters in combination with fuzzy rules, the problem of control stability of the robot arm in the unstable industrial Wi-Fi environment is solved, and stable operation and high flexibility control are achieved under abnormal communication conditions.

CN120921404AActive Publication Date: 2025-11-11XIANGTAN UNIV

Patent Information

Application Number
CN202511451033.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-10-11
Publication Date
2025-11-11
Estimated Expiration
2045-10-11

AI Technical Summary

Technical Problem

In industrial Wi-Fi unstable environments, existing technologies make it difficult for the control system of a robot arm to maintain stable control performance after switching when communication is abnormal. It lacks the ability to dynamically judge based on communication quality, resulting in control lag or error accumulation.

Method used

By acquiring communication parameters and the angular velocity and control torque of the robot arm in real time, calculating the network score, and smoothly switching to the local TDE prediction mode when communication is abnormal, and adjusting the control parameters in combination with fuzzy rules, adaptive control mode switching and parameter adjustment are achieved.

Benefits of technology

In the event of communication failure, ensure the smooth operation of the robot arm, avoid sudden stops or dangerous actions, improve the flexibility and anti-interference capability of the control system, and balance control stability and system responsiveness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120921404A_ABST
    Figure CN120921404A_ABST
Patent Text Reader

Abstract

The invention discloses a robot operation arm constraint safety system based on communication perception and a method thereof. The method comprises the steps that communication parameters with an upper computer and the angular speed and control torque of a robot operation arm are obtained in real time; de-noising the communication parameters, and performing evaluation calculation to obtain a network score; judging whether the network score is greater than a preset first threshold value, if so, judging that communication is abnormal and accumulating, if the communication is abnormal in continuous N periods, smoothly switching from an RL online mode to a local TDE prediction mode, and calculating a disturbance estimation value of the current period after mode switching; calculating an impact trend index of the current period based on the current control torque, the angular velocity and the historical period data difference value, and outputting a degradation factor from the disturbance estimation value and the impact trend index through a fuzzy rule; and adjusting the control parameter through the reduction factor, and generating the control torque of the next period. The fuzzy rule is fused with multi-factor decision, and control stability and system responsiveness are both considered.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of robot manipulator control technology, specifically to a robot manipulator constraint safety system and method based on communication sensing. Background Technology

[0002] With the development of industrial automation, robot manipulators have been widely used to assist or even replace humans in various tasks such as assembly, machine operation, deburring, drilling, transportation, and manufacturing. Typical robot systems often adopt a host computer-slave computer architecture, where the host computer sends control commands to the robot body (slave computer) to achieve precise motion control and task execution.

[0003] In such systems, the communication quality between the host computer and the robot directly affects the stability and responsiveness of the control. Delays, packet loss, or interference in the communication link can lead to deviations in control command execution, abnormal manipulator movements, and even equipment damage or safety accidents. Therefore, real-time detection of communication anomalies and adaptive adjustment of control strategies are crucial for ensuring the safe and stable operation of the robot.

[0004] Existing methods for handling communication anomalies mainly involve repeating historical cycle instructions or replacing instructions based on model predictions. However, in industrial Wi-Fi environments where communication is unstable, repeating historical cycle instructions cannot fully consider the dynamic changes in the control process, which can easily lead to robot lag or error accumulation. Model predictions are mostly preset static mode switching and lack the ability to make dynamic judgments based on the actual communication status.

[0005] Therefore, existing technologies cannot guarantee stable control performance after switching in unstable industrial Wi-Fi environments, and cannot dynamically switch to a more robust predictive control mode based on communication quality. Summary of the Invention

[0006] To address the problem that existing systems cannot simultaneously ensure control stability and system responsiveness under abnormal communication conditions, this invention provides a communication-aware robot manipulator constraint safety system and method.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] In a first aspect, this application discloses a communication-aware robotic arm constraint safety method, comprising the following steps:

[0009] Real-time acquisition of communication parameters with the host computer, as well as the angular velocity and control torque of the robot arm;

[0010] After denoising the communication parameters, a network score is obtained through evaluation calculation;

[0011] If the network score is greater than a preset first threshold, it is determined that the communication is abnormal and accumulated. If the communication is abnormal for N consecutive periods, it is smoothly switched from the RL online mode to the local TDE prediction mode, and the disturbance estimate of the current period is calculated after the mode switch. The number of consecutive periods N is an integer and 3≤N≤10.

[0012] The impact trend index for the current cycle is calculated based on the difference between the current control torque, angular velocity and historical cycle data. The disturbance estimate and the impact trend index are then output as a degradation factor through fuzzy rules.

[0013] By adjusting the control parameters by reducing the factor, the control torque for the next cycle is generated based on the control law.

[0014] Secondly, this application discloses a communication-aware robot arm constraint safety system, including a data acquisition module, a network scoring calculation module, a mode switching module, a degradation factor calculation module, and a result output module.

[0015] The data acquisition module is used to acquire communication parameters with the host computer in real time, as well as the angular velocity and control torque of the robot arm.

[0016] The network score calculation module is used to obtain the network score through evaluation calculation after denoising the communication parameters.

[0017] The mode switching module is used to determine whether the network score is greater than the preset first threshold. If it is, it determines that the communication is abnormal and accumulates it. If the communication is abnormal for N consecutive periods, it smoothly switches from the RL online mode to the local TDE prediction mode and calculates the disturbance estimate of the current period after the mode switch. The number of consecutive periods N is an integer and 3≤N≤10.

[0018] The degradation factor calculation module is used to calculate the impact trend index of the current cycle based on the difference between the current control torque, angular velocity and historical cycle data, and outputs the degradation factor by means of the disturbance estimate and the impact trend index through fuzzy rules.

[0019] The results output module is used to adjust the control parameters by reducing the factor and generate the control torque for the next cycle based on the control law.

[0020] Thirdly, this application discloses a computer terminal, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the steps of the aforementioned communication-aware robot arm constraint safety method.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] This application enables adaptive and smooth switching of control modes based on communication network conditions, avoiding sudden pauses or dangerous actions caused by communication interruptions, and ensuring that the robot can still operate smoothly in weak network conditions. In addition, it adopts fuzzy rule fusion for multi-factor decision-making and flexibly adjusts control parameters, avoiding both over-conservatism and system overshoot, improving the flexibility and anti-interference of the control system, and taking into account both control stability and system responsiveness. Attached Figure Description

[0023] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0024] Figure 1 This is a flowchart of the communication-aware robot manipulator constraint safety method described in this invention;

[0025] Figure 2 Based on Figure 1 The flowchart for network score calculation;

[0026] Figure 3 Based on Figure 1 A flowchart illustrating mode switching based on the changing trends of communication parameters;

[0027] Figure 4 Based on Figure 2 Rating-state mapping diagram;

[0028] Figure 5 Based on Figure 3 The communication parameter timing diagram, in which, Figure 5 (a) Timing diagram of round-trip time (RTT) measured for each cycle; Figure 5 (b) is the timing diagram for packet loss rate PLR. Figure 5 (c) is a time series plot of latency growth rate and packet loss jitter rate;

[0029] Figure 6 Based on Figure 1 Fuzzy Q-learning flowchart;

[0030] Figure 7 This is a schematic diagram of a rigid robotic arm with two linkages.

[0031] Figure 8 The simulation diagram of the angular velocity of joint 1 under the condition of no system uncertainty, external disturbance and friction;

[0032] Figure 9 The simulation diagram of the angular velocity of joint 2 under the condition of no system uncertainty, external disturbance and friction;

[0033] Figure 10The simulation diagram of the angular velocity of joint 1 under the conditions of system uncertainty, external disturbance and friction;

[0034] Figure 11 The simulation diagram of the angular velocity of joint 2 under the conditions of system uncertainty, external disturbance and friction;

[0035] Figure 12 The simulation diagram shows the angular position of joint 1 under the condition of no system uncertainty, external disturbance and friction;

[0036] Figure 13 The simulation diagram shows the angular position of joint 2 under the condition of no system uncertainty, external disturbance and friction;

[0037] Figure 14 The simulation diagram shows the angular position of joint 1 under the conditions of system uncertainty, external disturbance and friction;

[0038] Figure 15 The simulation diagram shows the angular position of joint 2 under the conditions of system uncertainty, external disturbance and friction;

[0039] Figure 16 This is a schematic diagram of the communication-aware robot arm constraint safety system described in Example 2. Detailed Implementation

[0040] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0041] Application Overview:

[0042] In existing technologies, control typically relies on real-time commands issued by a host computer to the robot. The quality of data transmission between the robot and the host computer directly affects the robot's actions. However, identifying safe robot operation largely depends on control algorithms. This "action-based, network-insensitive" approach is insufficient to guarantee remote intelligent control. Further research reveals that during the switch from online to local mode due to network anomalies, secondary risks such as unmodeled disturbances and instantaneous impacts arise. In this environment, control strategies lack adaptability and struggle to maintain a dynamic balance between performance and safety.

[0043] To address the aforementioned issues, considering the coupling relationship between communication status and system control performance, and the phased, sudden, and cumulative nature of communication quality changes, a joint perception of communication parameters and robot motion states is adopted as the foundation. In-depth research reveals that sudden changes in control torque caused by communication anomalies are transmitted through manipulator dynamics. Therefore, it is necessary to quantify the impact trend in advance and dynamically generate degradation factors using fuzzy rules to achieve smooth switching of control modes and adaptive parameter adjustment, minimizing mechanical impact while ensuring system stability under communication anomalies.

[0044] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] Example 1

[0046] like Figure 1 The diagram illustrates a communication-aware constraint safety method for robot manipulators, comprising the following steps:

[0047] S100. Real-time acquisition of communication parameters with the host computer, as well as the angular velocity and control torque of the robot arm;

[0048] S200. After denoising the communication parameters, the network score is obtained through evaluation calculation;

[0049] S300. Determine whether the network score is greater than the preset first threshold. If so, determine that the communication is abnormal and accumulate it. If the communication is abnormal for N consecutive periods, smoothly switch from the RL online mode to the local TDE prediction mode, and calculate the disturbance estimate for the current period after the mode switch. Wherein, N is an integer and 3≤N≤10.

[0050] S400. Based on the difference between the current control torque, angular velocity and historical period data, the current period impact trend index is calculated, and the disturbance estimate and impact trend index are output as a degradation factor through fuzzy rules.

[0051] S500. By reducing the factor, the control parameters are adjusted, and then the control torque for the next cycle is generated based on the control law.

[0052] Regarding step S100:

[0053] Communication parameters are collected in real time by a dedicated communication status monitoring module. For example, a communication status monitoring module can be integrated into the robot body controller to collect communication parameters (such as round-trip time, packet loss rate, and signal-to-noise ratio, SNR) in real time.

[0054] Obtain the current operating status of each joint from the controller, including:

[0055] angular velocity : Obtained through a motor encoder or speed estimation module;

[0056] Control torque Recorded by driver feedback or controller output;

[0057] It can also include the joint angle position. End-effector acceleration, etc. All data must be time-aligned, and the sampling frequency must be the same as or higher than the control cycle.

[0058] The angular velocity and control torque data from the most recent N* cycles are stored in a buffer to prepare historical time window data for subsequent steps (such as disturbance estimation and shock trend assessment). The communication parameters and control parameters are kept synchronized with the timestamps to avoid data mismatch.

[0059] Regarding step S200:

[0060] like Figure 2 As shown, the calculation process for network ratings is as follows:

[0061] S201. Obtain the round-trip time, packet loss rate, and signal-to-noise ratio as communication parameters when communicating with the host computer;

[0062] S202. After normalizing the round-trip delay, packet loss rate, and signal-to-noise ratio, the network score is obtained through weighted linear calculation.

[0063] Use sliding window mean filtering to eliminate short-term abnormal fluctuations (denoising):

[0064] Average round-trip time: ;

[0065] Average packet loss rate: ;

[0066] Average signal-to-noise ratio: ;

[0067] in, This refers to the sliding window size, ranging from 3 to 5. This represents the round-trip delay recorded in the i-th control cycle preceding the current time t; similarly, and This represents the packet loss rate and signal-to-noise ratio recorded in the i-th control cycle prior to the current time t; all parameters are updated once within one control cycle.

[0068] Then, a maximum-minimum range is set to normalize each parameter to the [0,1] interval, so as to facilitate the use of uniform units in the weighted calculation.

[0069] Maximum-minimum range of round-trip delay : 0ms~100ms;

[0070] Maximum-minimum range of packet loss rate 0%~30%;

[0071] Maximum-minimum signal-to-noise ratio range 5dB~30dB;

[0072] Average communication delay based on maximum-minimum range Average packet loss rate and average signal-to-noise ratio Normalized calculation:

[0073] Normalized round-trip delay ;

[0074] Normalized packet loss rate ;

[0075] Inverse normalized signal-to-noise ratio ;

[0076] in, A higher value indicates worse latency. A higher value indicates more severe packet loss. A higher value indicates a lower signal-to-noise ratio.

[0077] Online rating calculation:

[0078] ;

[0079] Among them, weight ,satisfy In this embodiment, , , Packet loss and latency are equally important; signal-to-noise ratio (SNR) is used to fine-tune communication quality assessment. Therefore, network scoring... The larger the value, the worse the communication quality.

[0080] Based on network ratings, communication quality is categorized into three levels: "Normal," "Suboptimal," and "Abnormal." See the table below for details.

[0081] Table 1: Network Rating Classification Table

[0082]

[0083] like Figure 4 The graph shown is a rating-state mapping, with the horizontal axis representing network rating and the vertical axis representing state membership degree. The blue curve represents Normal (ratio ≤ 0.4), indicating stable communication and a gradually decreasing membership degree of 1. The yellow curve represents Suboptimal (ratio between 0.4 and 0.7), with a mid-peak membership degree. The green curve represents Abnormal (ratio > 0.7), indicating poor communication and a gradually increasing membership degree.

[0084] This embodiment provides a specific example of how to divide interval values. In practical applications, it can be adapted to the standards of the communication environment.

[0085] Regarding step S300:

[0086] In this embodiment, the first threshold A score of 0.7 is preferred, indicating severe degradation in at least two metrics. This is when the network score is calculated. If the value is greater than 0.7, a communication anomaly is determined, and the communication anomaly cycle is accumulated. If the accumulated number reaches N, the control mode is switched from RL online mode to local TDE prediction mode. The control torque is synthesized by smooth interpolation to avoid control torque jumps.

[0087] Smooth interpolation control torque: ;

[0088] in, The controller weighting coefficient represents the degree of mixing between RL control and TDE control at the current moment. , It decreases linearly with the switching cycle, such as from 1 to 0; This is a flexibility range parameter for mode switching, used to smoothly adjust the decay range of control weights. , These are the control outputs for RL online mode and local TDE prediction mode, respectively. The second threshold is preferably 0.4.

[0089] The perturbation estimator is activated immediately in the first cycle after switching to local TDE prediction mode:

[0090] ;

[0091] in, Let be the angular acceleration vector, during the delay period The acceleration at time t, where L is the number of delay periods. Let the inverse matrix of the estimated mass-inertia coupling term be represented. The joint angle position vector, This is the input vector for the control torque.

[0092] The calculated disturbance estimate Cache, used for subsequent fuzzy rule input.

[0093] Regarding step S400:

[0094] Calculate the instantaneous angular velocity increment: ; and ω and ω are the angular velocities of the i-th joint at time t and time t-1, respectively.

[0095] Calculate the rate of change of control torque: ; and These are the control torques of the i-th joint at time t and time t-1, respectively.

[0096] If both exceed the set threshold, a "sudden shock" risk is considered to have occurred, and the shock trend is obvious. If only one exceeds the set threshold, a "significant shock" risk is considered to have occurred, and the shock trend is moderate. If neither exceeds the set threshold, a "minor shock" risk is considered to have occurred, and the shock trend is negligible.

[0097] For setting a threshold for the instantaneous angular velocity increment, offline sampling statistical analysis is used. The distribution of angular velocity changes is recorded under normal operating conditions, and the 95th percentile value under normal conditions is used as the threshold. In this embodiment, statistical analysis is performed, for example, by setting a threshold. The threshold for the rate of change of control torque can be set in the same way. .

[0098] The input to the fuzzy controller consists of two quantities: the disturbance estimate. and shock trend indicators For information on setting fuzzy rules, please refer to the table below:

[0099] Table 2: Degradation Factor Fuzzy Inference Rule Setting Table

[0100]

[0101] The output of the fuzzy controller is the degradation factor γ (0–0.5).

[0102] The specific steps for defuzzification are as follows:

[0103] (1) Define the output fuzzy membership function.

[0104] The fuzzy set of the output variable γ is defined as shown in the table below:

[0105] Table 3: Fuzzy Set Partition Table of Degradation Factors

[0106]

[0107] (2) Inference rules activate fuzzy sets:

[0108] Activate fuzzy output using a rule base:

[0109] High×Severe→Strong (Activation level 0.8);

[0110] Mid×Moderate→Medium (Activation level 0.6);

[0111] The fuzzy output membership function is clipped using the minimum value method.

[0112] (3) Aggregate the output fuzzy set:

[0113] The fuzzy set of all activated outputs becomes a single overall output function. :

[0114] .

[0115] Where i is the number of the fuzzy rule, Let be the membership function of the output fuzzy set corresponding to the i-th rule. The maximum value is taken for all rule results at each degradation factor γ position; min is the intersection of the rule output fuzzy set and its activation intensity; Let be the activation strength of the i-th rule.

[0116] For example, if there are two rules with output membership degrees at γ=0.25, respectively:

[0117] Rule 1 outputs "Medium": (0.25) = 0.6, activation strength =0.7;

[0118] Rule 2 outputs "Strong": (0.25) = 0.4, activation strength =0.5;

[0119] The aggregation result is:

[0120] .

[0121] (4) Centroid method for resolving fuzzy issues:

[0122] Numerical output calculated using the centroid method. :

[0123]

[0124] This is achieved through numerical integration, yielding a real-valued degradation factor. This is the deblurring output value (degradation factor); For the overall output function; It is approximately equal to the distance from the walking distance.

[0125] Regarding step S500:

[0126] The original controller parameters are dynamically adjusted, such as the damping coefficient and compliance coefficient, to obtain the defuzzified output value of the S400, i.e., the degradation factor. :

[0127] Damping coefficient after degradation: ;

[0128] The reduced compliance coefficient: ;

[0129] in, Represents the original compliance coefficient. This represents the original damping coefficient.

[0130] Since we are in local TDE prediction mode, the control torque for the next cycle is:

[0131] ;

[0132] in, , indicating position error, The joint's desired angular position vector; Indicates speed error; This is the estimated value of the disturbance; Let this be the matrix representing the estimated mass-inertia coupling term; Let be the desired joint angular acceleration vector.

[0133] The currently used control torque ,state( , ), control parameters, and other caches are used for judging the impact trend, state estimation, and TDE prediction in the next cycle.

[0134] If it is in RL online mode, the control torque for the next cycle is:

[0135]

[0136] in, For state (such as angular position, velocity, error, etc.); The dynamically adjusted parameter weights in the RL strategy, i.e. , This refers to the parameter set of RL strategies (such as DDPG, PPO, SAC, etc.). For safe mode parameters; Deblurring output values ​​(degradation factor).

[0137] It should be noted that the control frequency of the RL online mode is not synchronized with that of the local TDE prediction mode. It is preferable that the control frequency of the RL online mode is 1 / 20 of that of the local TDE prediction mode.

[0138] That is, when the local TDE controller cycles During operation, the RL controller cycles Operation. The output control torque remains constant for 20 TDE control cycles within the RL control cycle, during which high-frequency disturbance compensation is performed by the local TDE controller. The TDE module collects state data such as joint position, speed, and disturbance estimate of the actuator in each cycle, and uses this data to update the reinforcement learning strategy when the RL control cycle is triggered, realizing the coordinated operation of low-frequency strategy decision-making and high-frequency compensation control.

[0139] This application performs weighted evaluation of multiple communication parameters after denoising, which can accurately reflect changes in network quality. It combines reinforcement learning control and time delay estimation predictive control, and introduces a state-driven automatic switching mechanism to enhance the system's fault tolerance to communication anomalies through mode switching, thereby improving stability and task continuity. Breaking away from traditional rigid logic, it adopts fuzzy control to achieve multi-factor fusion judgment, and flexibly adjusts control parameters when the system disturbance risk is high, ensuring a balance between system responsiveness and stability.

[0140] The overall scheme of this application has been described above. The following section details the mode switching based on the changing trends of communication parameters after calculating the network score. Figure 3 As shown, the specific steps are as follows:

[0141] S601. Calculate the delay growth rate and jitter rate by taking the difference between the round-trip delay and packet loss rate of the current period and the previous period.

[0142] S602. If the delay growth rate is greater than the preset third threshold or the jitter rate is greater than the preset fourth threshold, it is determined that the communication is about to deteriorate, and the RL online mode is smoothly switched to the local TDE prediction mode.

[0143] The formula for calculating the delay growth rate is: ; , Let represent the round-trip delays at times t and t-1, respectively.

[0144] The formula for calculating packet loss jitter rate is: ; , Let represent the packet loss rates at times t and t-1, respectively.

[0145] The calculated latency growth rate and packet loss jitter rate can be adjusted using short-time moving averages or exponential smoothing to filter out instantaneous jitter.

[0146] If the delay growth rate (Third threshold) or packet loss jitter rate (Fourth threshold), that is, judging that "communication is about to deteriorate", switch mode in advance, switch status flag switch_flag=1, trigger the control mode to switch from RL to TDE, and record the switching operation in the log.

[0147] like Figure 5 The figure shows the time series diagrams for Round Trip Time (RTT), Packet Loss Rate (PLR), Delay Rate Increase, and Packet Loss Jitter Rate. Among them, Figure 5 (a) represents the network round-trip time measured in each cycle. In the early stage, the RTT is stable at around 50ms, indicating that the network is unobstructed. In the middle stage, the RTT gradually rises to 80ms, indicating that the network latency is deteriorating. In the later stage, the RTT begins to decrease, and the network gradually recovers. Figure 5 (b) indicates the real-time change in packet loss rate. Figure 5 In (c), the blue solid line represents the latency growth rate, which is the relative change rate of RTT compared to the previous period's RTT. The green solid line represents the packet loss jitter rate, which is the change in PLR of the current period relative to the PLR ​​of the previous period. The blue and green dashed lines represent the preset thresholds for the latency growth rate (third threshold) and packet loss jitter rate (fourth threshold), respectively. When the latency growth rate and packet loss jitter rate consistently exceed their respective thresholds, it indicates a rapid deterioration of the network condition, triggering a switch from RL mode to TDE mode. As can be seen from the figure, the network is in a deterioration phase during the mid-term.

[0148] Mode switching based on changing trends takes precedence over mode switching based on N consecutive cycles of communication anomalies. The settings for the third and fourth thresholds need to be fine-tuned based on on-site statistics. Initially, the third threshold... The fourth threshold is 0.10–0.30 (meaning a 10%–30% increase in round-trip time (RTT) triggers an early warning). : 0.05-0.20 (that is, an increase of 5% to 20% in packet loss rate (PLR) will trigger an early warning).

[0149] In practical applications, the smooth switching from RL online mode to local TDE prediction mode can be done in the same way as the smooth switching method in step S300.

[0150] This solution analyzes the gradient change characteristics of communication parameters in real time, enabling the system to smoothly transition between control modes before a substantial decline in communication quality occurs. This effectively avoids control command execution deviations caused by mode switching lag. This technical solution is particularly suitable for industrial environments with intermittent communication interference, such as scenarios where wireless signals are affected by equipment movement or electromagnetic interference, significantly improving the operational stability of the robot arm during periods of fluctuating communication quality.

[0151] The above describes the mode switching based on the changing trends of communication parameters. The following section details the adaptive adjustment based on the magnitude of the degradation factor after outputting the disturbance estimate and impact trend index using fuzzy rules. The specific steps are as follows:

[0152] S401. Determine whether the degradation factor exceeds the preset threshold;

[0153] S402. If so, the degradation factor is kept unchanged and used continuously for the control parameter update for at least M subsequent cycles, where the number of consecutive cycles M is an integer and 3≤M≤5.

[0154] Set the maximum degradation limit (preset threshold). Degradation factor .

[0155] Once the degradation factor output by a certain fuzzy logic judgment reaches or exceeds the preset threshold, the degradation state will be maintained for M consecutive control cycles, and this degradation intensity will be maintained regardless of whether the subsequent disturbance is weakened.

[0156] The fuzzy outputs from the most recent L periods are averaged to further smooth the output trend of the degradation factor. A maximum of 3 consecutive periods is preferred to stabilize the parameter output curve and avoid short-term fluctuations. If there are fewer than L periods of historical data, the existing number of periods is used.

[0157] Damping coefficient after degradation: ;

[0158] The reduced compliance coefficient: ;

[0159] in, To meet the current effective degradation factor (from the moving average output) that meets the period requirements; Represents the original compliance coefficient. This represents the original damping coefficient.

[0160] This solution analyzes the gradient change characteristics of communication parameters in real time, enabling the system to smoothly transition to the control mode before a substantial decline in communication quality occurs. This avoids control torque fluctuations caused by frequent adjustments to the degradation factor during communication anomaly recovery. By forcibly maintaining the stability of the degradation factor value, the robot arm can still maintain a smooth motion transition during communication state fluctuations, thereby improving the anti-interference capability and operational reliability of the control system.

[0161] After adjusting the control parameters by reducing the factor and going through M cycles, if the disturbance estimate, shock trend index, and network score of communication parameters are all within the normal range for N' consecutive cycles, the control parameters are gradually restored to their default values ​​in a linear or exponential manner; where the number of consecutive cycles N' is an integer and 8≤N'≤15.

[0162] Once the counter reaches 0, the system enters the observation period. During the observation period, continuously monitor whether the following three items are all within the normal range (per cycle):

[0163] absolute value of disturbance estimate That is, the disturbance estimate is less than the normal disturbance threshold;

[0164] Impact Trend Indicator That is, the impact trend is less than the normal impact threshold;

[0165] Online ratings ≤0.7.

[0166] If all three conditions are met within N′ consecutive cycles, the recovery phase begins; if any condition is not met in any cycle, the recovery counter is reset to zero and needs to be counted again for N′ cycles.

[0167] For the normal threshold of the disturbance The settings can be adjusted empirically. For a typical industrial robotic arm, the normal disturbance threshold can be set to 5%–10% of the rated joint control torque. Alternatively, the upper limit can be calculated based on the robotic arm's dynamic model and the maximum permissible disturbance control torque. Or, it can be determined through experimental statistics, under stable network conditions and without external interference, by collecting data for a period of time. Maximum and mean. For example, in a steady state. The mean value is 0.3 N·m, and the maximum value is 0.8 N·m, so we can set: ∈[0.8,1.0]N·m.

[0168] For impact normal threshold The threshold set in step S400 can be used.

[0169] Once the recovery startup conditions are met (N′ consecutive normal cycles), the control parameters are gradually restored to the baseline values ​​using one of the following two optional strategies (to prevent sudden changes):

[0170] (1) Linear recovery:

[0171] In the following Incremental recovery is performed within N periods (which can be equal to N′):

[0172] Damping parameters after incremental recovery: ;

[0173] Incremental recovery compliance coefficient: .

[0174] in, , The damping parameters and compliance coefficients before incremental recovery; Represents the original compliance coefficient. This represents the original damping coefficient.

[0175] (2) Index recovery:

[0176] Use attenuation factor ∈(0,1) (e.g.) =0.2):

[0177] .

[0178] The same applies to the flexibility coefficient. If any abnormality occurs during the recovery process (disturbance / impact / network score exceeds the threshold again), the recovery should be immediately stopped and the system should re-enter the hold state.

[0179] Specifically, when communication quality is restored, the system first verifies that the degraded control has been running stably for at least M cycles, ensuring that the mechanical system is under control. Subsequently, it continuously monitors whether the disturbance estimate is below the preset upper limit, whether the impact trend indicator remains stable, and whether the network score meets the passing standard within N' cycles. The simultaneous satisfaction of these three conditions indicates that both the communication link quality and the mechanical dynamic characteristics have returned to normal. The parameter recovery process employs a linear increment or exponential curve adjustment method; for example, the parameter value is increased by a fixed step size each control cycle, or it slowly approaches the default value using an exponential function. The lower limit of N' ensures that it at least covers the inertial response cycle of the mechanical system, while the upper limit avoids recovery delays due to excessive conservatism. This mechanism combines phased verification with gradual adjustment, maintaining degraded control parameters in the initial stage of communication recovery, and initiating the recovery process only after the system dynamic characteristics have fully stabilized, thereby eliminating step changes in control torque caused by parameter mutations.

[0180] In addition to gradual recovery, the system can also smoothly switch from the RL online mode to the local TDE prediction mode. If the network score of the communication parameters for N* consecutive periods is less than the preset second threshold, the system can smoothly switch from the local TDE prediction mode to the RL online mode. During the continuous judgment process, if the disturbance estimate and shock trend indicator for M+1 consecutive periods are within the normal range, the N* periods will be reduced by one-third. The number of consecutive periods N* is an integer and 10≤N*≤20.

[0181] After switching to local TDE prediction mode, the network continuous normal counter and disturbance / impact continuous normal counter Set to 0.

[0182] Execution is performed every control cycle t (TDE high-frequency cycle):

[0183] If online ratings ≤0.4, then ← +1, otherwise it will not be accumulated.

[0184] If the absolute value of the disturbance estimate And impact trend indicators ,but ← +1, otherwise the disturbance / impact continuous normal counter Setting it to 0 will not affect the number of consecutive cycles N*.

[0185] like ≥M+1 and current If the value is >0 (meaning the network is already in the process of continuous normal counting), then reduce the number of N* periods by one-third. The reduced number of consecutive periods should also satisfy the condition that it is not lower than the minimum number of consecutive periods (preferably set to 5).

[0186] When the network continuously normal counter Upon reaching the required number of cycles, if the RL is reachable (if the host computer is available), request and receive the latest policy output from the RL. If the RL is unavailable (communication link but network score is low but still unreachable), use historical RL output or set a safety reference to synthesize control torque through smooth interpolation.

[0187] If the network deteriorates again during the smooth handover process (e.g., network score exceeds the threshold or communication is interrupted), the handover should be immediately stopped and the system should revert to TDE mode. The original value of N can be increased to a larger value to avoid repeated attempts (e.g., multiplied by 1.2).

[0188] Specifically, once communication quality is restored, communication stability is verified by continuously monitoring the network score for N cycles to ensure it remains below a second threshold, thus avoiding misjudgments due to momentary fluctuations. Simultaneously, disturbance estimates and impact trend indicators are introduced as internal system state monitoring parameters. When the system remains within the normal range for M+1 consecutive cycles, it indicates that external interference and mechanical shocks have been effectively suppressed, at which point the N judgment cycle is shortened by one-third. For example, initially N* is set to 15 cycles, but this is shortened to 10 cycles after the internal state conditions are met, thereby accelerating mode switching response while ensuring communication reliability. This dual-condition judgment mechanism achieves dynamic optimization and adjustment of the switching cycle by combining communication quality and system dynamic state.

[0189] When smoothly switching from local TDE prediction mode to RL online mode, the specific control steps of RL online mode are as follows:

[0190] S701. Obtain the angular velocity of the robot arm. and tracking error ;

[0191] S702. The acquired data is mapped to fuzzy linguistic variables through triangular membership functions, and the fuzzy linguistic variables infer action candidate values ​​through predefined fuzzy rules;

[0192] Using triangle membership functions and The mapping is performed using fuzzy linguistic variables, such as "small," "medium," and "large." A set of predefined fuzzy rules is used to constrain the angular velocity. Provide the inference output, for example:

[0193] if large and Small, then It should be big;

[0194] if Small and Large, then It should be small.

[0195] Discretized The set of candidate values ​​serves as candidate values ​​for actions.

[0196] S703. Based on the Q value of the action candidate value, a greedy mechanism is used to select one action candidate value as the input of the control parameter, and the control torque is updated to drive the robot arm;

[0197] Among them, the greed mechanism: ;

[0198] It is a set of action candidate values The action is randomly selected from the choices. , is the probability of selecting that action; This is the optimal greedy action.

[0199] The value is calculated using the discharge rate and the selected action:

[0200] ;

[0201] It is the fuzzy input set at time t. For the nth rule unit in activation intensity, To sum the output action values ​​weighted by activation intensity, The sum of all activation intensities is used for normalization, making the result a weighted average. This is the output action value corresponding to the nth rule; for The overall output result at time t; The number of fuzzy rules.

[0202] The Q value at time t can be designed as follows:

[0203] ; State-action value function (Q value); It is the fuzzy input set at time t; For the nth rule unit in The activation intensity; Action value; The number of fuzzy rules; Indicates the contribution of the rule output; It is a weighted average of the activation degree, action selection and output contribution of each rule.

[0204] S704. Calculate the reward value and update the Q value after receiving the control effect of the driven robot arm, and use it as the control parameter selection for the next cycle.

[0205] state The target value is calculated as follows:

[0206] ; For the nth rule unit in The activation intensity; The number of fuzzy rules; Action value; Contribute to the rule output (optimal action).

[0207] Based on the current state Driven to the next state The temporal difference (TD) is calculated based on the reward obtained at time t:

[0208] ;

[0209] It is a time-series difference; The target value for the next state; State-action value function (Q value); This is a discount factor that reflects the contribution of future rewards. Reward Value for:

[0210] ;

[0211] in, For error-related terms; , which is a positive constant.

[0212] Update Q value:

[0213] .

[0214] This is the Q value at time t+1, i.e., the updated Q value; Let Q be the value at time t; To update the stride, please refer to the fuzzy Q-learning flowchart. Figure 6 The control strategy is adaptively adjusted based on the real-time communication status to reduce control command lag caused by network latency or packet loss, suppress error accumulation effects, and improve the motion accuracy and system robustness of the robot arm in industrial Wi-Fi unstable environments.

[0215] To verify feasibility, the following methods were adopted: Figure 7 The two-link rigid robotic arm shown was subjected to a simulation experiment with the following parameter settings:

[0216] Table 4: Robotic Arm Parameter Table

[0217]

[0218] The initial angular positions and angular velocities of the two joints are set as follows: =0°, =0°, =0° / s and =0° / s. The target angular positions of the two joints are set as follows: =30° and =45°.

[0219] The upper limits of the angular position and angular velocity of the i-th joint of the robotic arm (respectively...) and (This option is selected by the user based on specific task requirements.)

[0220] To successfully implement the proposed controller, any controller that satisfies... and Values ​​greater than 0. for or Therefore, in the simulation of this application, angular position constraints are selected. =50°, =60°, angular velocity constraint =10° / s and =12° / s.

[0221] Figure 8 and Figure 9The simulation results for joints 1 and 2 are shown in the absence of system uncertainties, external disturbances and friction. It can be seen that the angular velocities (blue solid lines) of joints 1 and 2 are both within the constraints (black dashed lines). Figure 10 and Figure 11 The simulation results for joints 1 and 2 under the conditions of system uncertainty, external disturbance and friction show that the angular velocities of joints 1 and 2 (blue solid lines) are within the constraints (black dashed lines).

[0222] Figure 12 and Figure 13 The simulation results for joint 1 and joint 2 are shown in the absence of system uncertainties, external disturbances and friction. The angular positions (blue solid lines) of joint 1 and joint 2 are both within the constraints (black dashed lines). Figure 14 and Figure 15 The simulation results for joints 1 and 2 under the conditions of system uncertainty, external disturbance and friction show that the angular positions of joints 1 and 2 (blue solid lines) are within the constraints (black dashed lines).

[0223] To facilitate understanding of the above embodiments, a specific application scenario of the above embodiments will be used as an example for illustration below:

[0224] Taking a smart manufacturing workshop as an example, a 6-DOF collaborative robot (rated joint control torque 20 N·m, rated joint angular velocity 2 rad / s) is responsible for picking up precision parts from the conveyor belt and assembling them onto the target assembly (tolerance ±0.05 mm). The workshop's network environment relies on industrial Wi-Fi (covering a 20 m² work area). Due to interference from multiple large welding robots and automated transport vehicles, communication is subject to periodic latency fluctuations and data packet loss.

[0225] The robot arm's local controller continuously acquires the round-trip time (RTT), packet loss rate (PLR), and signal-to-noise ratio (SNR) with the host computer, while simultaneously collecting the angular velocities of each joint. and the currently executing control torque .

[0226] In the initial state (RL online mode): RTT=52ms; PLR=0.8%; SNR=30dB. The calculated network score is 0.28 < 0.4, which is considered normal. Robot end effector position error: ±0.02mm (normal range).

[0227] If starting from the 120th cycle, RTT increases from 52ms to 78ms, and PLR increases from 0.8% to 2.5%.

[0228] Delay growth rate: It exceeds the third threshold of 0.15;

[0229] Packet loss jitter rate: It exceeds the fourth threshold of 0.1;

[0230] If the threshold is exceeded for three consecutive cycles (M=3), it is determined that the communication is about to deteriorate.

[0231] In cycle 123, the control torque output by the RL mode is 8.6 N·m. The transition to the TDE output occurs within a 5-cycle smooth window, with the initial compensation disturbance estimate being... =1.2N·m. After the switch is completed, the local TDE takes over the joint control. The end position error after the switch is controlled within ±0.04mm (still meeting the assembly tolerance).

[0232] After mode switching, the following was detected: joint angular velocity increment = 0.25 rad / s (greater than the threshold of 0.2 rad / s), control torque change rate = 1.5 N·m / cycle (greater than the threshold of 1.2 N·m / cycle). Fuzzy inference outputs a degradation factor. =0.35 (>threshold 0.3).

[0233] Maintaining this degradation strength for 4 cycles (M=4), the damping coefficient will be... The compliance coefficient decreased from 6.0 to 3.9. It decreased from 0.8 to 0.52.

[0234] Over 12 consecutive periods (N'=12), the disturbance estimate, shock trend indicator, and network score were all within the normal range, and the control parameters gradually recovered to their default values ​​in a linear or exponential manner.

[0235] At the 170th cycle, the network stability condition of N*=15 cycles is met. During the continuous M+1=5 cycles, the impact and disturbance are normal. N* is shortened to 10 cycles, and the network smoothly switches back to RL mode. The RL output takes over the low-frequency decision-making again, and the high-frequency compensation continues to be provided by TDE.

[0236] In summary, by acquiring communication parameters and manipulator status data in real time, dynamically switching control modes based on network scoring, and adaptively adjusting control parameters based on disturbance estimation and impact trend indicators, the problem of existing technologies being unable to perceive changes in communication quality in real time and struggling to dynamically adjust control strategies is solved, thereby improving the control stability and safety of the manipulator when communication is abnormal.

[0237] Example 2

[0238] like Figure 16 As shown in the figure, this embodiment introduces a communication-aware robot manipulator constraint safety system, including a data acquisition module 801, a network scoring calculation module 802, a mode switching module 803, a degradation factor calculation module 804, and a result output module 805.

[0239] The data acquisition module 801 is used to acquire communication parameters with the host computer in real time, as well as the angular velocity and control torque of the robot arm.

[0240] The network score calculation module 802 is used to obtain the network score through evaluation calculation after denoising the communication parameters.

[0241] The mode switching module 803 is used to determine whether the network score is greater than the preset first threshold. If it is, it determines that the communication is abnormal and accumulates it. If the communication is abnormal for N consecutive periods, it smoothly switches from the RL online mode to the local TDE prediction mode and calculates the disturbance estimate of the current period after the mode switch. The number of consecutive periods N is an integer and 3≤N≤10.

[0242] The degradation factor calculation module 804 is used to calculate the impact trend index of the current cycle based on the difference between the current control torque, angular velocity and historical cycle data, and output the degradation factor by means of the disturbance estimate and the impact trend index through fuzzy rules.

[0243] The result output module 805 is used to adjust the control parameters by reducing the factor and generate the control torque for the next cycle based on the control law.

[0244] Specifically, the data acquisition module 801 continuously collects communication parameters and the angular velocity and control torque data of the operating arm via the industrial bus and sensors, and transmits the data to the network score calculation module 802 for noise reduction. The network score calculation module 802 performs a weighted summation of the normalized round-trip delay, packet loss rate, and signal-to-noise ratio to generate a network score reflecting communication quality. The mode switching module 803 determines abnormal communication states by comparing the network score with a preset threshold. When an abnormal state occurs continuously for a set number of cycles, it triggers a switch to the local prediction mode and initiates disturbance estimation calculation. The degradation factor calculation module 804 calculates the impact trend index based on the difference between the current control torque and historical data, and infers the degradation factor through a fuzzy rule base using the disturbance estimation value. The result output module 805 applies the degradation factor to the control parameters to generate the control torque for the next cycle, thereby maintaining control stability through a dual mechanism of parameter degradation and mode switching when communication deteriorates.

[0245] This embodiment has the same beneficial effects as Embodiment 1.

[0246] Example 3

[0247] This embodiment introduces a computer terminal, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The feature is that when the processor executes the program, it implements the steps of the aforementioned communication-aware robot arm constraint safety method.

[0248] The computer terminal refers to a hardware device with data processing capabilities, which can be implemented using an industrial control computer or an embedded system. It is used to independently execute control algorithms and is independent of a host computer. The memory stores program code containing logic for acquiring communication parameters, calculating network scores, and switching modes. It can be implemented using flash memory or a solid-state drive, providing a storage medium for executable instructions for the control strategy. The processor parses program instructions and performs real-time calculations. It can be implemented using a multi-core CPU or an FPGA chip, achieving dynamic decision-making by executing control logic cycle by cycle.

[0249] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A robot manipulator constraint safety method based on communication sensing, characterized in that, It includes the following steps: Real-time acquisition of communication parameters with the host computer, as well as the angular velocity and control torque of the robot arm; After denoising the communication parameters, a network score is obtained through evaluation calculation; If the network score is greater than a preset first threshold, a communication anomaly is determined and accumulated. If there are N consecutive periods of communication anomalies, the RL online mode is smoothly switched to the local TDE prediction mode, and the disturbance estimate for the current period is calculated after the mode switch. The number of consecutive periods N is an integer and 3≤N≤10. The impact trend index for the current period is calculated based on the difference between the current control torque, angular velocity and historical period data. The disturbance estimate and the impact trend index are then output as a degradation factor through fuzzy rules. By adjusting the control parameters by reducing the factor, the control torque for the next cycle is generated based on the control law.

2. The robot manipulator constraint safety method based on communication sensing according to claim 1, characterized in that, The calculation process for online ratings is as follows: The round-trip time, packet loss rate, and signal-to-noise ratio during communication with the host computer are obtained as communication parameters. After normalizing the round-trip time, packet loss rate, and signal-to-noise ratio, the network score is obtained through weighted linear calculation.

3. The robot manipulator constraint safety method based on communication sensing according to claim 2, characterized in that, After calculating the network score, mode switching is also performed based on the changing trends of communication parameters. The specific steps are as follows: The delay growth rate and jitter rate are calculated by taking the difference between the round-trip delay and packet loss rate of the current period and the previous period. If the latency growth rate is greater than a preset third threshold or the jitter rate is greater than a preset fourth threshold, it is determined that the communication is about to deteriorate, and the system smoothly switches from RL online mode to local TDE prediction mode.

4. The robot manipulator constraint safety method based on communication sensing according to claim 1, characterized in that, After outputting the disturbance estimate and shock trend index as a degradation factor through fuzzy rules, the process also includes adaptive adjustment based on the magnitude of the degradation factor reduction. The specific steps are as follows: Determine if the degradation factor exceeds a preset threshold; If so, the degradation factor is kept unchanged and used for control parameter updates for at least M subsequent cycles, where the number of consecutive cycles M is an integer and 3≤M≤5.

5. The robot manipulator constraint safety method based on communication sensing according to claim 4, characterized in that, After adjusting the control parameters by reducing the factor and going through M cycles, if the disturbance estimate, shock trend index, and network score of communication parameters are all within the normal range for N' consecutive cycles, the control parameters are gradually restored to their default values ​​in a linear or exponential manner; where the number of consecutive cycles N' is an integer and 8≤N'≤15.

6. The robot manipulator constraint safety method based on communication sensing according to claim 5, characterized in that, After smoothly switching from RL online mode to local TDE prediction mode, if the network score of communication parameters for N* consecutive periods is less than the preset second threshold, then smoothly switch from local TDE prediction mode to RL online mode. In the continuous judgment process, if the disturbance estimate and shock trend indicator for M+1 consecutive periods are within the normal range, then the N* periods will be reduced by one-third; where the number of consecutive periods N* is an integer and 10≤N*≤20.

7. The robot manipulator constraint safety method based on communication sensing according to claim 1, characterized in that, The control frequency of the RL online mode is not synchronized with that of the local TDE prediction mode; the control frequency of the RL online mode is 1 / 20 of that of the local TDE prediction mode.

8. The robot manipulator constraint safety method based on communication sensing according to claim 7, characterized in that, If the current mode is RL online, the control steps for RL online mode are as follows: Obtain the angular velocity and tracking error of the robot arm; The acquired data is mapped to fuzzy linguistic variables through triangular membership functions, and the fuzzy linguistic variables infer action candidate values ​​through predefined fuzzy rules. Based on the Q value of the action candidate value, a greedy mechanism is used to select one action candidate value as the input of the control parameter, and the control torque is updated to drive the robot arm. After receiving the control effect of the driven robot arm, calculate the reward value and update the Q value, which is used for the selection of control parameters in the next cycle.

9. A robot manipulator constraint safety system based on communication sensing, characterized in that, It includes: The data acquisition module is used to acquire communication parameters with the host computer in real time, as well as the angular velocity and control torque of the robot arm. A network score calculation module is used to obtain a network score by evaluating and calculating the communication parameters after denoising them. The mode switching module is used to determine whether the network score is greater than a preset first threshold. If it is, it determines that the communication is abnormal and accumulates it. If the communication is abnormal for N consecutive cycles, it smoothly switches from the RL online mode to the local TDE prediction mode and calculates the disturbance estimate of the current cycle after the mode switch. The number of consecutive cycles N is an integer and 3≤N≤10. The degradation factor calculation module is used to calculate the impact trend index of the current cycle based on the difference between the current control torque, angular velocity and historical cycle data, and output the degradation factor by means of the disturbance estimate and the impact trend index through fuzzy rules. The results output module is used to adjust the control parameters by reducing the factor and generate the control torque for the next cycle based on the control law.

10. A computer terminal comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the communication-aware robot manipulator constraint safety method as described in any one of claims 1-8.

Citation Information

Patent Citations

  • Robot control method and device, electronic equipment and storage medium

    CN118493431A

  • Incomplete multi-mode sensing data recovery method and system based on conditional diffusion

    CN119337074A

  • Mechanical arm control system and method for emergency rescue

    CN120023836A

  • Adaptive multi-mode switching communication method and device

    CN120321724A

  • Industrial equipment fault prediction method based on multi-modal data

    CN120654024A

Cited By

  • Robot remote control method and system and storage medium

    CN121608146A

  • Bridge intelligent dehumidification system operation method based on environment parameter perception

    CN122172633A