A robot interaction force control method for intelligent polishing demonstration

Abstract

The application provides a robot interaction force control method for intelligent polishing teaching, and relates to the technical field of industrial robot control, and comprises the following steps: step S1: establishing a deterministic communication environment; step S2: acquiring and processing force sense signals; step S3: establishing a virtual wall and calculating a penetration metric; step S4: generating a one-way torque constraint; step S5: mapping to a joint space; step S6: underlying damping to suppress oscillation; and step S7: differentiated instruction distribution. The application effectively solves the problems of end self-excitation oscillation and teaching trajectory fluctuation caused by the contradiction between communication delay and high-gain feedback in high-frequency force control; at the same time, by establishing a one-way virtual constraint mechanism for spatial decoupling, the technical bottlenecks of simple transparent transmission prone to cutting and traditional virtual wall prone to sticking in traditional force feedback teleoperation are broken through, and the perfect fusion of omnidirectional transparent transmission of real polishing resistance and active protection of normal anti-cutting is realized.

Description

Technical Field

[0001] This invention relates to the field of industrial robot control technology, specifically to a robot interactive force control method for intelligent grinding teaching, applicable to intelligent manufacturing scenarios such as automotive parts repair and metal mold grinding for complex curved surface grinding teaching. Background Technology

[0002] In the fields of intelligent manufacturing and flexible processing, grinding, polishing, and deburring operations performed by industrial robots are core processes for improving product surface quality and production efficiency. Interactive teaching modes based on zero-force drag and virtual wall constraints have become a key means of rapidly generating machining trajectories for complex curved surfaces. By teaching the path through a remotely operated end effector, the teaching threshold can be significantly reduced and the flexible production needs of multiple varieties and small batches can be met.

[0003] Currently, interactive force control of industrial robots typically employs a closed-loop architecture consisting of sensors, a host computer, and actuators, using an industrial bus to exchange sensor data and control commands. However, the following technical shortcomings exist in practical applications:

[0004] First, control platforms based on non-real-time operating systems generally face challenges of communication jitter and phase lag. Due to the inherent communication delay in the damping compensation calculation of the host computer software layer, the feedback torque lags behind the actual physical motion phase of the motor on the time axis. This converts the damping that should consume energy into negative damping for driving oscillation, causing the end-effector joints with small inertia to easily generate high-frequency self-excited oscillations with equal amplitude or even divergent amplitude during zero-force interaction, which seriously affects the smoothness of teaching and operational safety.

[0005] Secondly, existing virtual wall technologies are mostly bidirectional rigid constraints, unable to achieve the unidirectional conduction characteristic of resistance when pushing inward and complete freedom when pulling outward. For example, the virtual wall establishment method disclosed in patent document CN118322210A only performs torque compensation at the software level, without considering the separation of spatial normal and tangential directions. This results in a noticeable stickiness and dragging sensation when the operator retracts, leading to low teaching efficiency. Furthermore, the teleoperation scheme based on variable stiffness virtual walls and damping force fields disclosed in patent document CN119238501A relies on a long master-slave data closed loop, resulting in high control latency. This makes it difficult to meet the high-frequency response requirements of grinding tasks, and it lacks precise separation of the task space direction vector, making it easy for the virtual wall resistance to cancel out the operator's teaching intention force.

[0006] Finally, in non-real-time environments such as Windows, communication anomalies frequently occur due to the lack of effective memory alignment and protocol optimization, resulting in discontinuous torque output and sudden changes in the operating feel. Existing solutions mostly rely on high-cost real-time operating systems or dedicated controllers, which are costly to deploy and not conducive to large-scale promotion.

[0007] In summary, existing technologies have significant shortcomings in areas such as communication determinism in non-real-time environments, the interactive feel of one-way virtual walls, and the suppression of high-frequency self-excited oscillations at the end point, and urgently need improvement. Summary of the Invention

[0008] This invention overcomes the shortcomings of existing technologies and provides a robot interactive force control method for intelligent grinding teaching. It effectively solves the problems of end-effector self-excited oscillation and teaching trajectory fluctuation caused by the contradiction between communication delay and high-gain feedback in high-frequency force control. At the same time, by establishing a spatially decoupled unidirectional virtual constraint mechanism, it breaks through the technical bottlenecks of easy overcutting in traditional force feedback teleoperation and easy sticking of traditional virtual walls, and achieves a perfect integration of omnidirectional transmission of real grinding resistance and normal anti-overcutting active protection.

[0009] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0010] A robot interactive force control method for intelligent polishing teaching includes the following steps:

[0011] Step S1: Establish a deterministic communication environment

[0012] The main control loop thread is bound to the isolated core of the industrial PC processor and set to real-time priority; at least two independent communication links are preset, wherein the first communication link is used for force feedback command issuance and the second communication link is used for pose synchronization command issuance, and the synchronization message of the second communication link is byte-aligned and encapsulated according to the memory alignment rules of the target execution end, while the first communication link is not byte-aligned.

[0013] Step S2: Acquire and process force signals

[0014] Acquire the force signal from the actuator and perform filtering processing;

[0015] Step S3: Build a virtual wall and calculate permeability.

[0016] Obtain the spatial pose of the execution end, establish a virtual wall and determine its normal vector, calculate the scalar projection of the penetration displacement of the execution end relative to the virtual wall onto the normal vector, and obtain the penetration distance;

[0017] Step S4: Generate unidirectional torque constraints

[0018] A virtual wall constraint force is generated based on the penetration distance, and a direction discrimination operator is introduced. The direction discrimination operator is updated in real time in each control cycle according to the sign of the penetration distance. The virtual wall constraint force is activated only when the execution end penetrates into the virtual wall, and a generalized interaction force is synthesized.

[0019] Step S5: Map to joint space

[0020] The generalized interactive force is mapped to the joint space of the teaching end to generate a force-controlled additional torque;

[0021] Step S6: Bottom-level damping suppresses oscillations

[0022] The control mode of the teaching terminal driver is set to the built-in speed damping closed-loop mode, and the damping coefficient is deployed as a fixed parameter inside the driver. The driver performs damping calculations at a local sampling frequency higher than the communication link frequency. The gravity compensation torque and the force control additional torque are combined into a comprehensive feedforward torque and sent to the driver.

[0023] Step S7: Differentiated Instruction Distribution

[0024] The integrated feedforward torque is sent to the teaching terminal via the first communication link, and the pose synchronization command is sent to the execution terminal via the second communication link.

[0025] Furthermore, in step S1, the byte alignment encapsulation refers to inserting an alignment placeholder after the single-byte enable flag of the synchronization message; the synchronization message is a binary encapsulation, and its message structure includes: frame synchronization header, sequence number, timestamp, joint pose data, activation flag, alignment placeholder, and check field.

[0026] Furthermore, in step S2, the filtering process employs a first-order low-pass filtering algorithm:

[0027] in These are the filter coefficients used to smooth the interactive feel; F raw It is the original six-dimensional force;

[0028] The filtered six-dimensional force.

[0029] Furthermore, step S3 specifically includes: obtaining the current spatial pose of the end effector. When the execution end contacts the workpiece and the normal contact force is greater than the preset threshold, the current pose is used as the virtual wall pose. The unit normal vector is Calculate the relative displacement vector = - The scalar penetration distance is obtained through dot product operation:

[0030] in, Let be the relative displacement vector of the end relative to the wall. For scalar penetration distance; if If the value is ≥0, the end is determined to be in the free zone outside the wall;

[0031] like <0, indicating the endpoint has entered the virtual wall constraint zone, with a penetration depth of [value missing]. .

[0032] Furthermore, step S4 specifically includes:

[0033] like When <0, the stiffness is preset according to the teaching terminal. Calculate the normal virtual restoring force

[0034]

[0035] Direction discrimination operator The update rule is: when hour =1, otherwise =0, synthesizing generalized interaction force:

[0036]

[0037] in, Preset stiffness for the teaching end. For the penetration distance, For normal virtual restoring force, It is the unit normal vector.

[0038] Furthermore, in step S5, the Jacobian transpose matrix is ​​used... Mapping generalized interactive forces to the joint space of the teaching end:

[0039] in, Let be the Jacobian transpose matrix. For generalized interaction force, Additional torque for force control;

[0040] In step S6, the integrated feedforward torque for:

[0041]

[0042] in, For gravity compensation torque, To control the additional torque, To integrate the feedforward torque.

[0043] Furthermore, in step S6, the synthesized torque of the driver's built-in speed damping closed-loop mode satisfies:

[0044]

[0045] in, For the synthesized torque inside the driver, For position gain, For the target location, This refers to the angular position of the motor. The damping coefficient is... For the target speed, The speed of the motor rotation. This is the torque feedforward term.

[0046] Furthermore, in step S1, the second communication link is an ADS protocol link or a custom binary protocol link based on UDP / TCP / IP; the main control cycle is 1kHz; and the local sampling frequency of the driver is not less than 10kHz.

[0047] Furthermore, in step S1, the industrial PC has a multi-core processor architecture, and the isolated core is at least one of its physical cores.

[0048] A robot interactive force control system for intelligent polishing teaching includes:

[0049] The host computer executes the instruction encapsulation steps in steps S1 to S5 and step S7 of the method.

[0050] The teaching end robotic arm is connected to the host computer via the first communication link, receives the comprehensive feedforward torque, and performs force feedback.

[0051] The execution end robotic arm is connected to the host computer via a second communication link, receives the pose synchronization command and performs pose synchronization. The end of the execution end robotic arm is equipped with a force sensor to collect force signals and upload them to the host computer.

[0052] The driver of the teaching end robotic arm has a built-in speed damping closed-loop mode, and the damping coefficient is deployed as a fixed parameter in the driver's internal register.

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] 1. Eliminating high-frequency self-excited oscillations at the end effector, ensuring system stability and mechanical lifespan; for the small inertia shaft at the end effector, this invention utilizes the MIT control mode built into its driver to directly reduce the velocity damping coefficient Kd to the actuator's underlying register, rather than calculating it at the upper-level software layer. Existing general solutions often calculate the damping torque in software and then send it down in non-real-time systems. Due to serial port delay, the feedback torque phase lag often induces periodic divergent jitter at the end effector. This invention improves the end effector oscillation convergence speed by more than 80% and reduces the mechanical fatigue wear of the reducer gears and motor magnets caused by high-frequency oscillations.

[0055] 2. Improve control determinism in non-real-time environments and reduce hardware deployment costs; this invention introduces a multi-threaded asynchronous ADS communication architecture and encapsulates control messages across platforms using 3-byte aligned placeholders. This architecture solves the time jitter problem of Windows systems under high loads and ensures isochronism of 1kHz command issuance. Through alignment mechanisms and thread isolation, this invention achieves industrial-grade control determinism without using expensive dedicated real-time controllers.

[0056] 3. Achieving unidirectional mechanical separation enhances the smoothness and efficiency of human-computer interaction. During grinding demonstrations, the operator can not only perceive the real physical grinding resistance and tangential friction force transmitted non-destructively by force sensors from all directions, but also additionally feel the high-stiffness virtual spring resistance that increases with penetration depth when pressing into the wall. Once the operator withdraws outward or moves tangentially, the virtual constraint torque instantly resets to zero, achieving a fusion of real road feel transmission and unidirectional flow prevention. This invention, through vector projection separation, improves the efficiency of teaching path generation by more than 30% and effectively reduces the physical and mental burden on operators. Attached Figure Description

[0057] The accompanying drawings are provided to further illustrate the invention and are used together with the embodiments of the invention to explain the invention. They do not constitute a limitation of the invention. In the drawings:

[0058] Figure 1 Overall flowchart of the intelligent polishing interactive force control program.

[0059] Figure 2 Geometric diagram of a virtual wall algorithm with unidirectional permeable projection characteristics.

[0060] Figure 3 This is the MIT mode control block diagram. Detailed Implementation

[0061] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0062] Example 1

[0063] like Figures 1 to 3 As shown, the robot interactive force control method for intelligent grinding teaching described in this invention achieves stable synchronization at a frequency of 1kHz through three stages: upper computer perception processing, core algorithm calculation, and heterogeneous link distribution. Specifically, it includes the following steps:

[0064] Step S1: Establish a deterministic communication environment based on a multi-threaded asynchronous architecture

[0065] To address the issue that control frequency is susceptible to interference from system scheduling and I / O blocking in Windows non-real-time system environments, this invention establishes an asynchronous control architecture based on the three-dimensional separation of computation, sensing, and communication.

[0066] Specifically, a multi-core industrial PC is used, with Core 6 and Core 7 configured as isolated cores at the system level; preferably, in this embodiment, an 8-core industrial PC is used. The main control loop thread is bound to the aforementioned isolated cores, and its process priority is set to real-time (Real, time) to minimize millisecond-level scheduling jitter caused by operating system task switching. The main control loop period is set to 1kHz.

[0067] The system pre-configures two completely independent communication links and allocates corresponding instruction buffers in memory:

[0068] (1) Local feedback link, also known as the first communication link: responsible for sending torque commands to the teaching end. It uses native byte stream encapsulation and does not perform any byte alignment processing to maintain the extremely sensitive interactive feel of the teaching end.

[0069] (2) Synchronous distribution link, also known as the second communication link, preferably using the ADS protocol, Thread_ADS): responsible for synchronizing pose to the execution end, and needs to match the memory access rules of the execution end PLC.

[0070] For synchronization messages sent to the PLC, a strict 120-byte binary encapsulation is enforced. Specifically, a 3-byte alignment placeholder is manually inserted after the 1-byte enable flag to ensure data integrity and eliminate communication anomalies such as "parameter size not correct" caused by inconsistent memory alignment rules. The message structure of the synchronization message is shown in the table below:

[0071] 0 Header Frame synchronization header (0x12345678) 4 4 Sequence Message sequence number (auto-incrementing count) 4 8 Timestamp microsecond-level local timestamp 8 16-111 Payload Joint position and velocity data (12*Double) 96 112 Enabled Algorithm activation flag 1 113-115 Padding Address alignment placeholder (critical alignment step) 3 116 Checksum Accumulated sum and check bits 4

[0072] It should be noted that the first communication link does not perform byte alignment processing, while the second communication link must strictly follow the above 120-byte binary encapsulation format to assemble messages, in order to ensure data consistency when communicating between different hardware architectures, such as x86 and ARM.

[0073] In alternative embodiments, the second communication link can also adopt a custom binary protocol based on UDP or TCP / IP. As long as similar address alignment placeholders are introduced in the message to adapt to the memory alignment rules of different hardware, the communication anomaly caused by message size mismatch can be resolved.

[0074] Step S2: Acquisition and filtering of six-dimensional force signals

[0075] The raw six-dimensional force vector (Fraw) is obtained by continuously polling the six-dimensional force sensor of the robotic arm at a 1M baud rate via a serial interface. This thread runs independently of the main loop to ensure the real-time acquisition of force data.

[0076] The acquired raw force vectors were subjected to CRC co-check to verify the integrity of the data transmission. Subsequently, a first-order low-pass filter algorithm was used to suppress high-frequency noise; the filter formula is as follows:

[0077]

[0078] in These are the filter coefficients used to smooth the interactive feel;

[0079] It is the original six-dimensional force;

[0080] The filtered six-dimensional force.

[0081] This filtering process effectively suppresses high-frequency noise in the force signal while keeping the phase lag within an acceptable range. In an alternative embodiment, the filtering coefficients can be adjusted according to the specific sensor noise characteristics and system bandwidth requirements.

[0082] Step S3: Establish the task space coordinate system and extract the normal vector.

[0083] When the robotic arm contacts the workpiece being ground, and the filtered end-effector normal contact force exceeds a pre-set contact threshold, the system obtains the current pose of the end-effector through forward kinematics calculation. The Z-axis vertical plane of the current pose is set as the virtual wall. As the robotic arm moves under continuous force, the virtual wall is updated in real time. Let the unit normal vector of the wall be... The direction pointing outward from the wall is defined as the positive normal direction.

[0084] Real-time acquisition of the current spatial pose of the end effector Calculate the relative displacement vector of the end relative to the wall:

[0085] = -

[0086] Step S4: Calculate the scalar projection of the penetration distance

[0087] By using the dot product operation, the relative displacement vector is projected onto the normal vector. The scalar penetration distance is calculated as follows:

[0088]

[0089] in, Let be the relative displacement vector of the end relative to the wall. For scalar penetration distance; if If the value is ≥0, the end is determined to be in the free zone outside the wall;

[0090] like <0, indicating the endpoint has entered the virtual wall constraint zone, with a penetration depth of [value missing]. .

[0091] Step S5: Generate unidirectional torque constraints

[0092] like When <0, the stiffness is preset according to the teaching terminal. Calculate the normal virtual restoring force

[0093]

[0094] in, The teaching terminal has a preset stiffness, which is set by the operator based on the workpiece material and grinding process requirements. The principle for selecting the stiffness value is: it should not generate a significant additional force under normal grinding conditions, but only provide a sense of resistance that increases with depth during overcutting. The specific value can be determined through on-site debugging.

[0095] Direction discrimination operator The update rule is: when hour =1, otherwise =0, based on the scalar permeation distance in each control cycle. Real-time updates The value of . Furthermore, when the end force sensor detects that the normal contact force Fz is unloaded and falls below a set threshold, Similarly, it is instantly reset to zero to further ensure zero-delay response of the pullback action.

[0096] Synthetic generalized interaction force:

[0097]

[0098] in, Preset stiffness for the teaching end. For the penetration distance, For normal virtual restoring force, It is the unit normal vector.

[0099] This synthesis method ensures that:

[0100] On the one hand, operators can use the filtered six-dimensional force It can perceive the real grinding contact resistance and tangential friction in real time and in all directions, preserving the real environment transmission of traditional teaching.

[0101] On the other hand, the system only adds a normal virtual wall restoring torque when the operator presses into the wall and geometric penetration occurs. This mechanism effectively locks the maximum grinding depth and prevents over-cutting by using unidirectional high-rigidity constraints without compromising the feel of tangential free grinding.

[0102] The reason why traditional virtual walls cause stickiness, that is, the feeling of suction, dragging or softness at the boundary when the operator retracts, is that traditional algorithms are usually based on pure geometric boundaries or admittance models. At the moment of retraction transition, due to the time domain delay caused by the filtering algorithm, the phase lag of the control loop, and the incomplete separation of tangential and normal torques, the restoring torque at the boundary cannot be absolutely synchronized with the operator's retraction action in time. As a result, there is still a residual reverse pull in the early stage of retraction.

[0103] This invention addresses this pain point through the following mechanism:

[0104] (1) The algorithm operates within a high-frequency control cycle. Once the operator intends to withdraw, the spatial geometric penetration distance is... If the value instantly becomes positive, or if the end force sensor detects that the normal contact force Fz is unloaded and falls below the set threshold, the direction discrimination operator σ will instantly change from 1 to 0 within a millisecond period, thus achieving zero-delay hard cut-off of the normal virtual wall constraint torque at the software level.

[0105] (2) This invention uses the normal vector The dot product projection makes the virtual wall moment Strictly constrained in the normal dimension perpendicular to the grinding surface. When the operator slides tangentially, since the real environmental contact force has been omnidirectionally mapped to the joint space as an independent feedforward term, the original physical frictional resistance is completely preserved in the tangential dimension, and the virtual wall does not generate any parasitic torque or damping pull in the non-constrained direction.

[0106] This asymmetric dynamic control mechanism, which allows torque to follow the action of the wall and to be activated immediately upon leaving the wall, ensures that the operator leaves the wall-touching state without any hesitation. While ensuring that the workpiece is not over-cut by grinding, it greatly improves the smoothness of human-machine interaction and the accuracy of industrial teaching.

[0107] Step S6: Map to the joint space of the teaching end and generate force-controlled additional torque.

[0108] Synthesized generalized interaction forces Through the Jacobian transpose matrix Mapped to the joint space at the teaching end, the force-controlled additional torque is obtained:

[0109]

[0110] in, Let be the Jacobian transpose matrix. For generalized interaction force, Additional torque for force control;

[0111] Step S7: Hardware-level bottom-layer Kd damping to suppress oscillations

[0112] The core method of this invention to solve the problem of high-frequency self-excited oscillation of the end shaft during zero-force interactive teaching lies in the parameter separation and sinking of the MIT control architecture inside the motor actuator.

[0113] according to Figure 3 The diagram shown is a MIT mode control block diagram of the motor, where the synthesized torque inside the driver is... The mathematical expression is:

[0114]

[0115] in, For the synthesized torque inside the driver, For position gain, For the target location, This refers to the angular position of the motor. The damping coefficient is... For the target speed, The speed of the motor rotation. This is the torque feedforward term.

[0116] To achieve a transparent interactive feel and suppress oscillations caused by communication latency, this embodiment assigns the following specific parameters to the formula:

[0117] Set target location Target speed The position gain Kp = 0. The formula then simplifies to:

[0118] .

[0119] in, The damping coefficient is set to a constant in this embodiment and is directly written into the corresponding bit of the MIT control message, serving as a fixed parameter resident in the driver's internal register. This is the torque feedforward term.

[0120] The host computer control program calculates the gravity compensation torque in real time using the dynamics library. The force-controlled additional torque obtained from step S5 above. The combined feedforward torque is then generated.

[0121]

[0122] Will As a torque feedforward term, it is sent to the driver via serial port messages.

[0123] In zero-force drag mode, if the system only provides gravity-compensated feedforward, the end joint is prone to excessive slippage and velocity divergence when subjected to external disturbances due to the lack of energy dissipation mechanisms. Therefore, the system needs to introduce damping torque to absorb redundant kinetic energy and improve the smoothness and dynamic stability during dragging.

[0124] In existing technologies, if the damping torque is calculated at the host computer software level, due to Windows system scheduling jitter and the physical bandwidth limitation of serial communication, the feedback torque will lag behind the actual physical motion phase of the motor on the time axis. This phase lag will convert the damping that should consume energy into negative damping that drives oscillation, causing the end shaft to generate high-frequency oscillations of equal amplitude or even divergent amplitude under contact or disturbance.

[0125] In this invention, The gain is a fixed parameter residing in the driver's internal register. The driver's internal MCU obtains the motor speed using its local extremely high sampling frequency. And execute directly within microseconds. Calculation.

[0126] This invention moves the damping suppression logic from the host computer software layer to the driver's underlying MCU, compressing the closed-loop cycle from milliseconds to microseconds, completely eliminating the phase lag caused by traditional communication and system scheduling, and fundamentally suppressing the high-frequency self-excited oscillation of the small inertia end shaft in zero-force interaction.

[0127] In an alternative embodiment, using a high-performance host computer with a real-time operating system to directly issue high-frequency damping commands via a very short-cycle real-time bus can also partially achieve the purpose of suppressing oscillations. However, the solution proposed in this invention has greater cost and deployment advantages in a typical Windows environment.

[0128] Step S8: Serialization and distribution of control instructions

[0129] In the interactive control loop described in this invention, the motion synchronization of the robotic arm is set as the final execution stage of the current control cycle. Only after the main loop completes all force signal processing, unidirectional virtual wall logic separation, and joint space torque mapping does the system execute the following heterogeneous distribution process to ensure a high degree of consistency in spatial pose between the execution end and the teaching end:

[0130] 1. Real-time force feedback at the teaching terminal, for the first communication link / serial port link:

[0131] The main loop will calculate the gravity compensation torque. The force-controlled additional torque calculated in step S6 Combined into a comprehensive feedforward torque :

[0132]

[0133] via serial port protocol The constant term damping gain Kd is directly sent to the motor driver at the teach pendant. This process, as the first output branch of the main loop, does not undergo any byte alignment processing to maintain the extremely sensitive interactive feel of the teach pendant.

[0134] 2. Pose command synchronization at the execution end, for the second communication link / ADS link:

[0135] After confirming that all main loop logic operations have been completed, extract the joint target position and target velocity data generated in the current frame as execution instructions.

[0136] The control instructions are strictly encapsulated into 120-byte binary format, consistent with the message structure described in step S1. A 3-byte alignment placeholder is manually inserted after the 1-byte enable flag to ensure the PLC correctly parses the message. After encapsulation, the instructions are sent to the execution PLC via a synchronous distribution link to control the execution robotic arm to follow the movement of the teach pendant.

[0137] Through the above-mentioned serial and ADS dual-link synchronization mechanism, motion synchronization and real-time mapping between the execution end robotic arm and the teaching end robotic arm under heterogeneous communication protocols are achieved.

[0138] Example 2

[0139] The present invention also provides a robot interactive force control system for intelligent grinding teaching, comprising:

[0140] The host computer is used to execute the instruction encapsulation steps in steps S1 to S7 of the method described in Embodiment 1. The host computer is preferably a multi-core industrial PC running a Windows operating system. Through system-level configuration, a designated core is set as an isolated core, and the main control loop thread is bound to this isolated core and set to real-time priority.

[0141] The teaching arm connects to the host computer via the first communication link to receive the comprehensive feedforward torque. It also performs force feedback. The driver of the robotic arm at the teaching end is configured in MIT control mode, with the damping coefficient Kd deployed as a fixed parameter in the driver's internal register. The driver performs damping calculations at a sampling frequency of not less than 10kHz.

[0142] The actuator-end robotic arm connects to the host computer and TwinCAT3 via a second communication link, configured to receive 120-byte binary synchronization messages and perform pose synchronization. A six-dimensional force sensor is installed at the end effector of the actuator-end robotic arm to acquire raw six-dimensional force vectors. The data is then uploaded to the host computer. The execution-end robotic arm is responsible for performing the grinding operation at the actual grinding station, and its movement trajectory is synchronized with the operation of the teaching-end robotic arm.

[0143] This invention improves system stability and user experience during the teaching process by introducing multi-threaded reconstruction and underlying hardware feedback mechanisms into the teaching robot control system. Specific benefits are as follows:

[0144] 1. Eliminating high-frequency self-excited oscillations at the end effector, ensuring system stability and mechanical lifespan; for the small inertia shaft at the end effector, this invention utilizes the MIT control mode built into its driver to directly reduce the velocity damping coefficient Kd to the actuator's underlying register, rather than calculating it at the upper-level software layer. Existing general solutions often calculate the damping torque in software and then send it down in non-real-time systems. Due to serial port delay, the feedback torque phase lag often induces periodic divergent jitter at the end effector. This invention improves the end effector oscillation convergence speed by more than 80% and reduces the mechanical fatigue wear of the reducer gears and motor magnets caused by high-frequency oscillations.

[0145] 2. Improve control determinism in non-real-time environments and reduce hardware deployment costs; this invention introduces a multi-threaded asynchronous ADS communication architecture and encapsulates control messages across platforms using 3-byte aligned placeholders. This architecture solves the time jitter problem of Windows systems under high loads and ensures isochronism of 1kHz command issuance. Through alignment mechanisms and thread isolation, this invention achieves industrial-grade control determinism without using expensive dedicated real-time controllers.

[0146] 3. Achieving unidirectional mechanical separation enhances the smoothness and efficiency of human-computer interaction. During grinding demonstrations, the operator can not only perceive the real physical grinding resistance and tangential friction force transmitted non-destructively by force sensors from all directions, but also additionally feel the high-stiffness virtual spring resistance that increases with penetration depth when pressing into the wall. Once the operator withdraws outward or moves tangentially, the virtual constraint torque instantly resets to zero, achieving a fusion of real road feel transmission and unidirectional flow prevention. This invention, through vector projection separation, improves the efficiency of teaching path generation by more than 30% and effectively reduces the physical and mental burden on operators.

[0147] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. However, any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A robot interactive force control method for intelligent grinding teaching, characterized in that, Includes the following steps: Step S1: Establish a deterministic communication environment The main control loop thread is bound to the isolated core of the industrial PC processor and set to real-time priority; at least two independent communication links are preset, wherein the first communication link is used for force feedback command issuance and the second communication link is used for pose synchronization command issuance, and the synchronization message of the second communication link is byte-aligned and encapsulated according to the memory alignment rules of the target execution end, while the first communication link is not byte-aligned. Step S2: Acquire and process force signals Acquire the force signal from the actuator and perform filtering processing; Step S3: Build a virtual wall and calculate permeability. Obtain the spatial pose of the execution end, establish a virtual wall and determine its normal vector, calculate the scalar projection of the penetration displacement of the execution end relative to the virtual wall onto the normal vector, and obtain the penetration distance; Step S4: Generate unidirectional torque constraints A virtual wall constraint force is generated based on the penetration distance, and a direction discrimination operator is introduced. The direction discrimination operator is updated in real time in each control cycle according to the sign of the penetration distance. The virtual wall constraint force is activated only when the execution end penetrates into the virtual wall, and a generalized interaction force is synthesized. Step S5: Map to joint space The generalized interactive force is mapped to the joint space of the teaching end to generate a force-controlled additional torque; Step S6: Bottom-level damping suppresses oscillations Set the control mode of the teach pendant driver to the built-in speed damping closed-loop mode, deploy the damping coefficient as a fixed parameter inside the driver, and have the driver perform damping calculations at a local sampling frequency higher than the communication link frequency. The gravity compensation torque and the force control additional torque are combined into a comprehensive feedforward torque and sent to the driver. Step S7: Differentiated Instruction Distribution The integrated feedforward torque is sent to the teaching terminal via the first communication link, and the pose synchronization command is sent to the execution terminal via the second communication link.

2. The method according to claim 1, characterized in that, In step S1, the byte alignment encapsulation refers to inserting an alignment placeholder after the single-byte enable flag of the synchronization message; the synchronization message is a binary encapsulation, and its message structure includes: frame synchronization header, sequence number, timestamp, joint pose data, activation flag, alignment placeholder and check field.

3. The method according to claim 1, characterized in that, In step S2, the filtering process employs a first-order low-pass filtering algorithm: in These are the filter coefficients used to smooth the interactive feel; It is the original six-dimensional force; The filtered six-dimensional force.

4. The method according to claim 1, characterized in that, Step S3 specifically includes: obtaining the current spatial pose of the end effector. When the execution end contacts the workpiece and the normal contact force is greater than the preset threshold, the current pose is used as the virtual wall pose. The unit normal vector is Calculate the relative displacement vector = - The scalar penetration distance is obtained through dot product operation: in, Let be the relative displacement vector of the end relative to the wall. For scalar penetration distance; if If the value is ≥0, the end is determined to be in the free zone outside the wall; like <0, indicating the endpoint has entered the virtual wall constraint zone, with a penetration depth of [value missing]. .

5. The method according to claim 4, characterized in that, Step S4 specifically includes: like When <0, the stiffness is preset according to the teaching terminal. Calculate the normal virtual restoring force Direction discrimination operator The update rule is: when hour =1, otherwise =0, synthesizing generalized interaction force: in, Preset stiffness for the teaching end. For the penetration distance, For normal virtual restoring force, It is the unit normal vector.

6. The method according to claim 1, characterized in that, In step S5, the Jacobian transpose matrix is ​​used. Mapping generalized interactive forces to the joint space of the teaching end: in, Let be the Jacobian transpose matrix. For generalized interaction force, Additional torque for force control; In step S6, the integrated feedforward torque for: in, For gravity compensation torque, To control the additional torque, To integrate the feedforward torque.

7. The method according to claim 1, characterized in that, In step S6, the synthesized torque of the driver's built-in velocity damping closed-loop mode satisfies: in, For the synthesized torque inside the driver, For position gain, For the target location, This refers to the angular position of the motor. The damping coefficient is... For the target speed, The speed of the motor rotation. This is the torque feedforward term.

8. The method according to claim 1, characterized in that, In step S1, the second communication link is an ADS protocol link or a custom binary protocol link based on UDP / TCP / IP; the main control cycle is 1kHz; and the local sampling frequency of the driver is not less than 10kHz.

9. The method according to claim 1, characterized in that, In step S1, the industrial PC has a multi-core processor architecture, and the isolated core is at least one of the physical cores.

10. A robot interactive force control system for intelligent grinding teaching, characterized in that, include: The host computer executes the instruction encapsulation steps in steps S1 to S5 and step S7 of the method described in any one of claims 1 to 9. The teaching end robotic arm is connected to the host computer via the first communication link, receives the comprehensive feedforward torque, and performs force feedback. The execution end robotic arm is connected to the host computer via a second communication link, receives the pose synchronization command and performs pose synchronization. The end of the execution end robotic arm is equipped with a force sensor to collect force signals and upload them to the host computer. The driver of the teaching end robotic arm has a built-in speed damping closed-loop mode, and the damping coefficient is deployed as a fixed parameter in the driver's internal register.

Images