Multi-sensor fusion mechanical arm joint module force control system and debugging method

Abstract

The application relates to the field of mechanical arms and discloses a multi-sensor fusion mechanical arm joint module force control system and a debugging method.The multi-sensor fusion mechanical arm joint module force control system comprises the following: a sensing module: used for collecting state data of a mechanical arm joint module in real time through a multi-sensor; a processing module: used for pre-processing the state data by adopting a federal learning algorithm and fusing the pre-processed state data by utilizing an extended Kalman filtering algorithm to obtain processing data; and a decision module: used for constructing a multi-objective optimization function based on a mechanical arm dynamics model according to the processing data.Through multi-sensor collection of multi-dimensional data, pre-processing of the data by utilizing federal learning, combination of the extended Kalman filtering to fuse output accurate state estimation values, construction of a multi-objective optimization function based on the fused data, real-time identification of dynamics parameters by utilizing a recursive least square method, and dynamic adjustment of a weighting factor by utilizing a particle swarm optimization algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mechanical arm, in particular to a multi-sensor fusion mechanical arm joint module force control system and a debugging method. BACKGROUND

[0002] As the core component of intelligent manufacturing and intelligent household appliances, the force control technology of the mechanical arm is crucial in the fields of industrial assembly, service robots, etc. In the smart home scenario, the mechanical arm needs to complete fine operations such as food material grabbing, tableware placing, and cleaning work, which puts forward higher requirements for force control precision, environmental adaptability, and man-machine cooperation safety. The existing mechanical arm force control system mainly relies on a single sensor (such as a strain type force sensor) or a fixed parameter control algorithm, and it is difficult to cope with the challenges of object material diversity (such as glass, ceramic, and flexible food materials) in the home environment, dynamic changes of the load (such as grabbing different weight objects), and unstructured scenes (such as messy table work).

[0003] The traditional mechanical arm force control system usually adopts the "single sensor feedback + PID control" architecture. For example, an industrial mechanical arm collects the end contact force by installing a six-dimensional force sensor, and adjusts the joint torque in combination with a PID controller to realize force control closed loop. However, the parameters of the PID controller need to be preset manually, and when the mechanical arm grabs different weight objects, the change of the inertia matrix will cause the force control error to increase significantly; it cannot identify the mechanical arm dynamics parameters in real time, and it is difficult to cope with the model mismatch problems caused by load changes, joint wear, etc. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a multi-sensor fusion mechanical arm joint module force control system and a debugging method, which solves the problem that the traditional mechanical arm force control system cannot identify the mechanical arm dynamics parameters in real time and is difficult to cope with the model mismatch problems caused by load changes, joint wear, etc.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: a multi-sensor fusion mechanical arm joint module force control system, comprising:

[0006] A perception module for collecting state data of the mechanical arm joint module in real time through multiple sensors;

[0007] A processing module for pre-processing the state data using a federated learning algorithm, and fusing the pre-processed state data using an extended Kalman filter algorithm to obtain processing data;

[0008] A decision module for constructing a multi-objective optimization function based on a mechanical arm dynamics model according to the processing data, and performing parameter identification and optimization using a recursive least squares method and a particle swarm optimization algorithm in cooperation to generate a control instruction;

[0009] Execution module: Used to drive the joint modules of the robotic arm to move according to control commands.

[0010] By adopting the above technical solution, multi-dimensional data is collected through multiple sensors to provide rich input for the model. Federated learning is used to preprocess the data, and extended Kalman filtering is combined to fuse the output accurate state estimates. A multi-objective optimization function is constructed based on the fused data. Dynamic parameters are identified in real time using recursive least squares method, and the weighting factor is dynamically adjusted using particle swarm optimization algorithm to balance multiple objectives such as accuracy and energy consumption. The joint movement is driven according to the optimized instructions, and state monitoring and emergency braking are realized by human-computer interaction. This achieves real-time online identification of the robot arm's dynamic parameters, adaptive adjustment of multi-objective control strategies, and generation of control instructions. It solves the problem that traditional robot arm force control systems cannot identify robot arm dynamic parameters in real time and are difficult to deal with model mismatch caused by load changes and joint wear.

[0011] Preferably, the sensing module includes a data acquisition unit and a data transmission unit. The data acquisition unit is used to collect the status data of the robotic arm joint module in real time through multiple sensors, including force sensors, motion sensors, environmental sensors, and vision sensors. The status data includes the motion status information, force information, environmental information, and target object information of the robotic arm joint module. The data transmission unit is used to transmit the various data through communication protocols, including SPI, CAN, and Ethernet.

[0012] Preferably, the processing module includes a preprocessing unit and a fusion unit. The preprocessing unit is used to preprocess the sensor data at the edge computing node using a federated learning algorithm. The preprocessing includes data filtering, noise reduction, feature extraction, and normalization. The fusion unit is used to fuse the preprocessed state data using an extended Kalman filter algorithm to obtain processed data. The extended Kalman filter algorithm includes a state prediction equation, a covariance prediction equation, and a measurement update equation.

[0013] Preferably, the state prediction equation is: in, Let f be the predicted state at time k based on information from time k-1, and let f be the nonlinear state transition function. U is the state estimate at time k-1. k-1 This is the control input at time k-1;

[0014] The covariance prediction equation is: Among them, P k|k-1 To predict covariance, F k-1 Let P be the Jacobian matrix of the state transition matrix. k-1|k-1 Q is the estimated covariance at time k-1.k-1 The process noise covariance matrix;

[0015] The measurement update equation includes and P k|k =(IK k H k )P k|k-1 , where K k For Kalman gain, H k To measure the Jacobian matrix of the matrix, R k To measure the noise covariance matrix, z k Let h be the measurement value at time k, h be the nonlinear measurement function, and I be the identity matrix.

[0016] Preferably, the decision-making module includes a predictive control unit, an adaptive control unit, a scheduling management unit, and an instruction generation unit. The predictive control unit is used to construct a multi-objective optimization function based on the processed data and the robotic arm dynamics model, including force error, energy consumption, and speed smoothness. The robotic arm dynamics model is a model of the relationship between joint torque and motion state based on the Lagrange equation. The adaptive control unit is used to identify and optimize the parameters of the robotic arm dynamics model using a combination of recursive least squares method and particle swarm optimization algorithm to obtain the identification result. The iterative variation of the recursive least squares method... The quantities include parameter estimates, gain matrix, and covariance matrix. The recursive least squares method is used to continuously update the parameters for parameter estimation. The particle swarm optimization algorithm is used to perform a global search optimization on the weighting factors λ1, λ2, and λ3 when the parameter estimation converges to a local optimum during the parameter identification process of the recursive least squares method. The scheduling management unit is used to allocate computing resources based on task priority and system resource occupancy, using a scheduling strategy that combines time-slice round-robin and priority preemption. The instruction generation unit is used to generate control instructions based on the optimized multi-objective function and parameter identification results.

[0017] Preferably, the multi-objective optimization function is: Where N is the prediction time domain length, λ1, λ2, and λ3 are weighting factors, and f d f(k) is the expected force vector at time k, f(k) is the predicted actual force vector at time k, and τ(k) is the joint torque vector at time k. Let be the vector of joint velocity changes at time k.

[0018] Preferably, the execution module includes a drive control unit and a human-machine interaction unit. The drive control unit is used to drive the robotic arm joint module to move according to control commands by employing a servo driver based on magnetic field orientation control. The human-machine interaction unit is used to perform human-machine interaction through a touch screen and a graphical interface, including displaying system status, receiving user commands, and performing emergency braking operations.

[0019] The debugging method for force control of multi-sensor fusion robotic arm joint modules, applied to the aforementioned multi-sensor fusion robotic arm joint module force control system, includes the following steps:

[0020] Perception: Real-time collection of status data of the robotic arm joint module through multiple sensors;

[0021] Processing: The federated learning algorithm is used to preprocess the state data, and the extended Kalman filter algorithm is used to fuse the preprocessed state data to obtain the processed data;

[0022] Decision-making: Construct a multi-objective optimization function based on the processed data, and use recursive least squares method and particle swarm optimization algorithm in combination to identify and optimize parameters and generate control commands;

[0023] Execution: Drive the robotic arm joint module to move according to the control instructions.

[0024] This invention provides a force control system and debugging method for a robotic arm joint module based on multi-sensor fusion. It has the following beneficial effects:

[0025] 1. This invention collects multi-dimensional data through multiple sensors, preprocesses the data using federated learning, and fuses the output with extended Kalman filtering to produce accurate state estimates. Based on the fused data, a multi-objective optimization function is constructed. Dynamic parameters are identified in real time using recursive least squares method, and weighting factors are dynamically adjusted using particle swarm optimization algorithm. Joint movement is driven according to the optimized instructions. This invention achieves real-time online identification of robotic arm dynamic parameters, adaptive adjustment of multi-objective control strategies, and generation of control instructions. It solves the problem that traditional robotic arm force control systems cannot identify robotic arm dynamic parameters in real time and are difficult to cope with model mismatch caused by load changes, joint wear, etc.

[0026] 2. This invention updates dynamic parameters in real time through recursive least squares method and combines particle swarm optimization algorithm to perform global search on weighting factors of multi-objective functions, forming a closed-loop feedback of "parameter identification-objective optimization". This enables the robotic arm to automatically correct model parameters and balance multiple objectives such as force control accuracy, energy consumption, and motion smoothness when grasping objects of different weights or when joint performance deteriorates. This ensures that the control strategy evolves dynamically with environmental changes and enhances the robustness of the system in unstructured scenarios.

[0027] 3. This invention constructs a multi-objective optimization function based on the dynamic model of the robotic arm, and ensures the real-time operation of the algorithm through time slice rotation and priority preemption strategies. This enables the system to complete the entire process of calculation from solving the objective function to generating instructions. By dynamically adjusting the weighting factors, multi-objective collaborative optimization is achieved, which not only ensures the control of the end contact force, but also reduces motor energy consumption and suppresses sudden changes in joint speed. This allows the robotic arm to output efficient and stable control commands in different task scenarios, thereby improving overall work efficiency and equipment reliability. Attached Figure Description

[0028] Figure 1 This is a system architecture diagram of the force control system for the multi-sensor fusion robotic arm joint module proposed in this invention;

[0029] Figure 2 This is a flowchart of the debugging method for force control of a multi-sensor fusion robotic arm joint module proposed in this invention. Detailed Implementation

[0030] The technical solution of the present invention will now be clearly and completely described 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.

[0031] Please see the appendix Figure 1 This invention provides a force control system for a robotic arm joint module based on multi-sensor fusion, comprising:

[0032] The sensing module is used to collect real-time status data of the robotic arm joint module through multiple sensors. The sensing module includes a data acquisition unit and a data transmission unit. The data acquisition unit is used to collect real-time status data of the robotic arm joint module through multiple sensors, including force sensors, motion sensors, environmental sensors, and vision sensors. The status data includes motion status information, force information, environmental information, and target object information of the robotic arm joint module. The data transmission unit is used to transmit various data through communication protocols, including SPI, CAN, and Ethernet.

[0033] Specifically, a "sensory nervous system" for the robotic arm is constructed using multiple types of sensors. Force sensors capture details of contact forces, motion sensors monitor joint dynamics, vision sensors build environmental maps, and environmental sensors sense variables such as temperature and humidity. After being preprocessed by edge computing, the data is transmitted to the processing module with low latency and high reliability, providing multi-dimensional data support for subsequent control.

[0034] The main force sensor, ATINano17, adopts the strain gauge measurement principle. It converts the force on the joint into a voltage signal through a Wheatstone bridge, which is then converted into a digital signal by a 24-bit ADC with a sampling rate of 1kHz. It can directly output six-dimensional force / torque data. The redundant FBG torque sensor utilizes the wavelength drift characteristics of fiber optic gratings (sensitivity 1pm / N·m) to calculate the torque in real time through a demodulator (such as MicronOpticssm125), avoiding system failure caused by the failure of a single sensor.

[0035] The motion sensor's IMU module fuses gyroscope and accelerometer data through complementary filtering, outputting joint attitude quaternions (update rate 100Hz) with a drift rate <0.1° / h; the absolute encoder uses a binary encoder disk, supports power-off position memory, and has a repeatability accuracy ≤±0.001°.

[0036] The binocular camera of the vision sensor focuses on the target object through the ROI (Region of Interest) algorithm and calculates the disparity map by combining it with the stereoBM algorithm of OpenCV, with a 3D reconstruction error of ≤1mm; the structured light camera generates 10 frames of point cloud data per second, and performs surface fitting through the PCL library to construct a 3D model of the object.

[0037] The data transmission unit employs a layered communication protocol: a high-speed channel (SPI / CAN) for transmitting real-time data from force sensors and encoders, with the CAN bus baud rate set to 1Mbps to meet the control cycle (1ms) requirement; and a low-speed channel (Ethernet) for transmitting visual point cloud data using the TCP / IP protocol, with a reserved bandwidth of 100Mbps. Anti-interference design is incorporated, with sensor cables using twisted-pair shielded wire, the CAN bus equipped with a 120Ω terminating resistor, and the Ethernet interface integrating a magnetic isolation module to ensure data transmission in industrial environments.

[0038] The processing module is used to preprocess the state data using a federated learning algorithm and fuse the preprocessed state data using an extended Kalman filter algorithm to obtain the processed data. The processing module includes a preprocessing unit and a fusion unit. The preprocessing unit uses a federated learning algorithm to preprocess the data from each sensor at the edge computing node. The preprocessing includes data filtering, noise reduction, feature extraction, and normalization. The fusion unit uses an extended Kalman filter algorithm to fuse the preprocessed state data to obtain the processed data. The extended Kalman filter algorithm includes a state prediction equation, a covariance prediction equation, and a measurement update equation.

[0039] The state prediction equation is in, Let f be the predicted state at time k based on information from time k-1, and let f be the nonlinear state transition function. U is the state estimate at time k-1.k-1 This is the control input at time k-1;

[0040] The covariance prediction equation is Among them, P k|k-1 To predict covariance, F k-1 Let P be the Jacobian matrix of the state transition matrix. k-1|k-1 Q is the estimated covariance at time k-1. k-1 The process noise covariance matrix;

[0041] Measurement update equations include and P k|k =(IK k H k )P k|k-1 , where K k For Kalman gain, H k To measure the Jacobian matrix of the matrix, R k To measure the noise covariance matrix, z k Let h be the measurement value at time k, h be the nonlinear measurement function, and I be the identity matrix.

[0042] Specifically, the processing module uses federated learning to achieve distributed feature extraction, reducing data transmission pressure and privacy risks; it uses extended Kalman filtering (EKF) to fuse multi-source data and construct a unified state space that includes joint position, velocity, force, and temperature, providing the decision-making module with accurate real-time state estimation of the robotic arm.

[0043] The preprocessing unit builds a federated learning network based on edge computing nodes (such as NVIDIA Jetson Nano). Each sensor node (force sensor, vision camera, etc.) acts as an independent client, completing data cleaning and feature extraction locally, and only uploading the feature vectors to the central server, thus avoiding raw data leakage and transmission bandwidth pressure.

[0044] The fusion unit is the core component of the processing module. Based on the Extended Kalman Filter (EKF) algorithm, it achieves optimal estimation of multi-source sensor data, providing the decision-making module with real-time and accurate robotic arm state information. The EKF algorithm takes as input force sensor measurements, joint position and velocity data from motion sensors, visual measurements, and joint torque commands from control inputs. Its output is a state estimate: a 12-dimensional vector containing joint position, velocity, and end effector force, updated at a frequency of 1 kHz; and a covariance matrix reflecting the uncertainty of the state estimate, used for parameter adjustment by the decision-making module.

[0045] In the state prediction phase, the fusion unit uses the following formula: Input the state estimate from the previous time step. Control input Uk-1 The processing involves predicting, calculating, and predicting forces using the robotic arm's dynamics model. The predicted state of the robotic arm's joint module at time k is output based on information from time k-1. By combining the dynamics model with the predicted state from the control input (such as joint velocity and end force), a preliminary estimate of the robotic arm's motion state at the next moment is achieved. This provides a basic prediction value for subsequent precise control, enabling the robotic arm to plan its motion trend in advance based on its current state and control commands.

[0046] In the covariance prediction stage: the formula is... Input the covariance matrix P from the previous time step k-1|k-1 Process noise matrix Q k-1 The processing involves a state transition matrix, which is simplified to an identity matrix. The covariance is then calculated, and the predicted covariance of the robotic arm joint module is output. This quantifies the prediction uncertainty (such as the variance of speed / force), reflects the model confidence, assesses the reliability of the state prediction, clarifies the fluctuation range of the prediction results, provides a standard for measuring prediction errors in the subsequent measurement and update stage, and assists the decision-making module in determining whether and how to adjust the control strategy.

[0047] During the measurement update phase: the formula is... Input measurement value z k This includes data acquired by force sensors, encoder data, data acquired by vision cameras, and measurement functions. The direct mapping of force and position predictions involves calculating residuals, Kalman gain, and state updates, outputting corrected state estimates of the robotic arm joint modules. This allows for the correction of predictions using sensor data (e.g., adjusting the end-effector force estimate based on actual force sensor measurements). This process integrates actual sensor measurements with model predictions, corrects prediction biases, and improves the accuracy and reliability of robotic arm state estimation. As a result, the robotic arm can adjust its motion state in a timely manner according to actual conditions, meeting the requirements for high-precision control.

[0048] Through three-stage iteration, the fusion unit achieves a closed loop from model prediction to data correction, ensuring force control accuracy in grasping tasks.

[0049] The decision-making module is used to construct a multi-objective optimization function based on the processed data and the robotic arm's dynamics model. It employs a combination of recursive least squares and particle swarm optimization algorithms to identify and optimize parameters, generating control commands. The module includes a predictive control unit, an adaptive control unit, a scheduling management unit, and a command generation unit. The predictive control unit constructs a multi-objective optimization function based on the processed data and the robotic arm's dynamics model, considering factors such as force error, energy consumption, and speed smoothness. The robotic arm's dynamics model is a model of the relationship between joint torques and motion states established based on the Lagrange equation. The adaptive control unit uses a combination of recursive least squares and particle swarm optimization algorithms to optimize the robotic arm's dynamics. The parameters of the mechanical model are identified and optimized to obtain the identification results. The iterative variables of the recursive least squares method include the parameter estimates, gain matrix, and covariance matrix. The recursive least squares method is used to continuously update the parameters for parameter estimation. The particle swarm optimization algorithm is used to perform a global search optimization on the weighting factors λ1, λ2, and λ3 when the parameter estimation converges to a local optimum during the parameter identification process of the recursive least squares method. The scheduling management unit is used to allocate computing resources based on task priority and system resource occupancy, using a scheduling strategy that combines time-slice round-robin and priority preemption. The instruction generation unit is used to generate control instructions based on the optimized multi-objective function and parameter identification results.

[0050] The multi-objective optimization function is Where N is the prediction time domain length, λ1, λ2, and λ3 are weighting factors, and f d f(k) is the expected force vector at time k, f(k) is the predicted actual force vector at time k, and τ(k) is the joint torque vector at time k. Let be the vector of joint velocity changes at time k.

[0051] Specifically, the decision-making module is based on the dynamic model of the robotic arm and constructs a multi-objective optimization function that integrates force error, energy consumption, and speed smoothness. Based on the multi-objective optimization function, it comprehensively considers multiple objectives such as the force control accuracy, motion stability, and energy consumption of the robotic arm, outputs joint torque commands, drives the robotic arm to perform tasks, identifies model parameters in real time through recursive least squares (RLS) method, and breaks through local optima by combining particle swarm optimization (PSO) to dynamically generate optimal control commands and realize the adaptive evolution of control strategies.

[0052] The predictive control unit, based on the robotic arm's dynamics model and fused state data, constructs a multi-objective optimization function and outputs an optimized sequence of control objectives. Its input processing module outputs fused state data including a 12-dimensional state vector with joint position, velocity, and end effector force; and task objective parameters such as desired force and trajectory points in the grasping task. Its core processing utilizes formulas... Wherein: Force error term: the force error term is calculated by f(k) fused from the force sensor and the expected force f. d The difference calculation of (k) ensures the accuracy of the end effector's contact force (e.g., error ≤ 0.5N when grasping an egg); Energy consumption term: Based on the sum of squares of the joint torque τ(k), minimize motor energy consumption (e.g., reduce energy consumption by 15% during continuous operation); Speed ​​smoothing term: Through the change in joint speed Suppress motion shocks (e.g., speed fluctuations ≤5% during plate swirl). Rolling optimization solution: Utilizing a Model Predictive Control (MPC) framework, the optimization function is solved in a rolling manner over the prediction time domain N (set as 20 steps, corresponding to 20ms), outputting a sequence of control variables for the future time period. Output: Optimized multi-objective function parameters (weighting factors λ1, λ2, λ3) and the control variable sequence, used for adaptive control unit parameter identification and command generation.

[0053] For example, in the "grab a glass" task, the input is the visually detected position of the glass and the desired gripping force f. d =1.5N, the optimization function automatically adjusts λ1=0.7 (force control priority), and outputs the joint torque sequence to make the end force rise smoothly to the target value, avoiding the glass from breaking due to excessive force.

[0054] The adaptive control unit identifies the dynamic model parameters of the robotic arm in real time, optimizes the weighting factors of the multi-objective function, and improves the robustness of the system. Its inputs include: the control quantity sequence τ(k) output by the predictive control unit; fused state data from the processing module and raw sensor measurements (such as encoder position and raw force sensor force values). Its core processing steps are: recursive least squares (RLS) parameter identification to estimate the dynamic model parameters (inertia matrix, Coriolis force matrix, gravity vector) in real time, and iterative processing; particle swarm optimization (PSO) weighting factor adjustment: the trigger condition is when the parameter update magnitude identified by RLS is <1% (i.e., trapped in a local optimum), the PSO algorithm is started; minimizing the multi-objective function value J, with particle positions corresponding to weighting factors λ1, λ2, and λ3, iteratively searching for the global optimum (such as reallocating weights during load changes), and outputting the identified dynamic model parameters and the optimized weighting factors λ1, λ2, and λ3.

[0055] For example, when the robotic arm grasps food of different weights (such as switching from grasping eggs to grasping canned food), the RLS updates the inertia matrix parameters in real time, and the PSO adjusts λ2 (energy consumption weight) synchronously from 0.3 to 0.5 to balance the grasping force and motor load.

[0056] The scheduling and management unit dynamically allocates computing resources based on task priority and system resources to ensure real-time performance. It dynamically allocates computing resources based on task priority and system resource status, defining three priority levels (level 1 for real-time control tasks and level 3 for non-real-time tasks) and employing a time-slice round-robin and priority preemption strategy to ensure the control cycle remains stable within 1ms. After inputting task requests and system resource status from each unit, tasks are first sorted by priority, then tasks of the same priority are allocated equal time slices. When a higher-priority task arrives, a lower-priority task is immediately interrupted. Finally, the task execution sequence and resource allocation table are output. In multi-task concurrent scenarios, priority is given to ensuring CPU resources for control command generation, avoiding control latency.

[0057] The instruction generation unit converts the optimized control parameters into executable physical instructions to drive the robotic arm. After converting the optimized control parameters into executable physical instructions, and inputting the optimized control sequence from the predictive control unit, the parameter identification results from the adaptive control unit, and the execution priority instructions from the scheduling management unit, the unit first converts the joint torque instruction into a 12-bit precision PWM signal. Then, based on the identified dynamic parameters, it calculates a feedforward compensation term to counteract inertial and Coriolis forces. Simultaneously, it performs physical limit verification on the instructions (e.g., joint torque ≤ 20 N·m). Finally, it outputs drive control instructions and safety status signals. For example, in the "stirring ingredients" task, it accurately generates PWM signals and verifies the torque safety threshold to ensure uniform stirring force and equipment safety.

[0058] Execution Module: Used to drive the robotic arm joint modules to move according to control commands. The execution module includes a drive control unit and a human-machine interface unit. The drive control unit uses a servo driver based on field-oriented control to drive the robotic arm joint modules to move according to control commands. The human-machine interface unit is used for human-machine interaction through a touch screen and graphical interface, including displaying system status, receiving user commands, and emergency braking operations.

[0059] Specifically, the execution module converts digital instructions into mechanical motion, the servo drive unit achieves precise output of joint torque, and the human-machine interaction unit constructs a safe and convenient operating interface through a touch screen and an emergency braking mechanism, ensuring the reliability and ease of use of the robotic arm in human-machine collaboration scenarios.

[0060] The drive control unit receives the torque command τ(k), realizes the torque closed loop through the current loop (bandwidth 1kHz), receives encoder feedback, performs trajectory tracking through the PID controller, and compensates for the centrifugal force and Coriolis force in real time during the high-speed movement of the robotic arm through the feedforward control algorithm.

[0061] The human-machine interface unit employs a safety mechanism: when any axial force exceeds 30N, emergency braking is triggered, with a braking distance ≤10mm (at a speed ≤0.1m / s); a ToF sensor (accuracy ±1cm) monitors the distance to the operator, automatically switching to low-speed mode (speed ≤0.05m / s) when the distance is <20cm. The interactive interface uses the Qt framework to develop a GUI, including a 3D simulation interface: real-time rendering of the robotic arm's motion trajectory, error visualization (force error is represented by color gradients, red >0.5N, green ≤0.2N); and a data dashboard: dynamically displaying parameters such as joint temperature, motor current, and multi-objective function values.

[0062] By coordinating the sensing, processing, decision-making, and execution modules, a multi-sensor fusion force control system for robotic arm joint modules is constructed, thereby enabling real-time identification of robotic arm dynamic parameters and dynamic optimization of the model.

[0063] The sensing module uses multiple devices, including force sensors, motion sensors, and vision sensors, to collect data in real time, such as joint position, force, ambient temperature, and target object coordinates. This data is then transmitted to the processing module via SPI / CAN / Ethernet communication protocols, providing multi-dimensional input data for the dynamic model. For example, when grasping ingredients of different weights, the force sensor provides real-time feedback on changes in contact force, while the vision camera identifies the object's shape and position in advance, providing a basis for model parameter identification.

[0064] The processing module utilizes a federated learning algorithm to filter, denoise, and extract features from sensor data at edge nodes. It then fuses the data using an extended Kalman filter (EKF) to output accurate state estimates, including joint states and end-effector forces. The EKF state prediction equation is constructed based on the robotic arm's dynamics model (Lagrange equations). By fusing multi-source data, the model's input accuracy is improved, laying the foundation for subsequent parameter identification.

[0065] The predictive control unit in the decision-making module constructs a multi-objective optimization function, including force error, energy consumption, and speed smoothness, based on the state estimates from the processing module, thus clarifying the control objectives and parameter optimization directions. The adaptive control unit uses recursive least squares (RLS) to identify dynamic parameters in real time, dynamically matching changes in the robot arm load by iteratively updating parameter estimates, gain matrix, and covariance matrix. Simultaneously, the particle swarm optimization (PSO) algorithm performs a global search of the weighting factors of the multi-objective function when RLS gets stuck in a local optimum, balancing objectives such as force control accuracy and energy consumption. The scheduling management unit ensures that the RLS / PSO algorithm completes parameter updates within a 1ms control cycle through time-slice round-robin and priority preemption strategies, avoiding computational delays that affect real-time performance. The instruction generation unit generates joint torque instructions based on the optimized model parameters and control objectives, which are then output to the execution module after feedforward compensation and safety verification.

[0066] The execution module drives joint movement through a servo driver based on magnetic field orientation control and monitors the status in real time using a human-machine interface unit. For example, when joint wear causes changes in the inertia matrix, the RLS corrects the model parameters in real time, the PSO adjusts the weighting factor to prioritize force control accuracy, and the servo driver accurately outputs torque according to the updated instructions to ensure stable gripping force. At the same time, an emergency braking mechanism (such as force over-limit trigger braking) ensures human-machine safety.

[0067] This enables dynamic optimization of the entire process from data acquisition, fusion, parameter identification to control command generation, improving the robotic arm's adaptability to scenarios such as load changes and component wear, and solving the model mismatch problem that traditional robotic arm force control systems cannot cope with load changes and joint wear.

[0068] Please see the appendix Figure 2 The debugging method for force control of multi-sensor fusion robotic arm joint modules, applied to the aforementioned multi-sensor fusion robotic arm joint module force control system, includes the following steps:

[0069] Perception: Real-time collection of status data of the robotic arm joint module through multiple sensors;

[0070] Processing: The federated learning algorithm is used to preprocess the state data, and the extended Kalman filter algorithm is used to fuse the preprocessed state data to obtain the processed data;

[0071] Decision-making: Construct a multi-objective optimization function based on the processed data, and use recursive least squares method and particle swarm optimization algorithm in combination to identify and optimize parameters and generate control commands;

[0072] Execution: Drive the robotic arm joint module to move according to the control instructions.

[0073] 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 force control system for a robotic arm joint module based on multi-sensor fusion, characterized in that, include: Sensing module: Used to collect real-time status data of the robotic arm joint module through multiple sensors; Processing module: Used to preprocess the state data using a federated learning algorithm, and then use an extended Kalman filter algorithm to fuse the preprocessed state data to obtain the processed data; Decision module: Based on the processed data, it constructs a multi-objective optimization function based on the robotic arm dynamics model, and uses recursive least squares method and particle swarm optimization algorithm in combination to identify and optimize parameters and generate control commands; Execution module: Used to drive the joint modules of the robotic arm to move according to control commands; The decision-making module includes a predictive control unit, an adaptive control unit, a scheduling management unit, and an instruction generation unit. The predictive control unit is used to construct a multi-objective optimization function based on the processed data and the robotic arm dynamics model, including force error, energy consumption, and speed smoothness. The robotic arm dynamics model is a model of the relationship between joint torque and motion state based on the Lagrange equation. The adaptive control unit is used to identify and optimize the parameters of the robotic arm dynamics model by using recursive least squares method and particle swarm optimization algorithm in combination, and obtain the identification result. The iterative variables of the recursive least squares method include parameter estimates, gain matrix, and covariance matrix. The recursive least squares method is used to continuously update the parameters for parameter estimation. The particle swarm optimization algorithm is used to perform global search optimization on the weighting factors λ1, λ2, and λ3 when the parameter estimation converges to a local optimum during the parameter identification process of recursive least squares method. The scheduling management unit is used to allocate computing resources based on task priority and system resource occupancy, using a scheduling strategy that combines time slice round-robin and priority preemption. The instruction generation unit is used to generate control instructions based on the optimized multi-objective function and parameter identification results. The multi-objective optimization function is: Where N is the prediction time domain length, λ1, λ2, and λ3 are weighting factors, and f d f(k) is the expected force vector at time k, f(k) is the predicted actual force vector at time k, and τ(k) is the joint torque vector at time k. Let be the vector of joint velocity changes at time k.

2. The force control system for a multi-sensor fusion robotic arm joint module according to claim 1, characterized in that: The sensing module includes a data acquisition unit and a data transmission unit. The data acquisition unit is used to collect the status data of the robotic arm joint module in real time through multiple sensors, including force sensors, motion sensors, environmental sensors, and vision sensors. The status data includes the motion status information, force information, environmental information, and target object information of the robotic arm joint module. The data transmission unit is used to transmit the various data through communication protocols, including SPI, CAN, and Ethernet.

3. The force control system for a multi-sensor fusion robotic arm joint module according to claim 1, characterized in that: The processing module includes a preprocessing unit and a fusion unit. The preprocessing unit is used to preprocess the sensor data at the edge computing node using a federated learning algorithm. The preprocessing includes data filtering, noise reduction, feature extraction, and normalization. The fusion unit is used to fuse the preprocessed state data using an extended Kalman filter algorithm to obtain processed data. The extended Kalman filter algorithm includes a state prediction equation, a covariance prediction equation, and a measurement update equation.

4. The force control system for a multi-sensor fusion robotic arm joint module according to claim 3, characterized in that: The state prediction equation is: ,in, Let f be the predicted state at time k based on information from time k-1, and let f be the nonlinear state transition function. U is the state estimate at time k-1. k-1 This is the control input at time k-1; The covariance prediction equation is: , where P k∣k-1 To predict covariance, F k-1 Let P be the Jacobian matrix of the state transition matrix. k-1∣k-1 Q is the estimated covariance at time k-1. k-1 The process noise covariance matrix; The measurement update equation includes , and , where K k For Kalman gain, H k To measure the Jacobian matrix of the matrix, R k To measure the noise covariance matrix, Z k Let h be the measurement value at time k, h be the nonlinear measurement function, and I be the identity matrix.

5. The force control system for a multi-sensor fusion robotic arm joint module according to claim 1, characterized in that: The execution module includes a drive control unit and a human-machine interaction unit. The drive control unit is used to drive the robotic arm joint module to move according to control commands by employing a servo driver based on magnetic field orientation control. The human-machine interaction unit is used to perform human-machine interaction through a touch screen and a graphical interface, including displaying system status, receiving user commands, and performing emergency braking operations.

6. A debugging method for force control of a multi-sensor fusion robotic arm joint module, characterized in that: The force control system for the multi-sensor fusion robotic arm joint module according to any one of claims 1-5 includes the following steps: Perception: Real-time collection of status data of the robotic arm joint module through multiple sensors; Processing: The federated learning algorithm is used to preprocess the state data, and the extended Kalman filter algorithm is used to fuse the preprocessed state data to obtain the processed data; Decision-making: Construct a multi-objective optimization function based on the processed data, and use recursive least squares method and particle swarm optimization algorithm in combination to identify and optimize parameters and generate control commands; Execution: Drive the robotic arm joint module to move according to the control instructions.

Images