CN118112991A - Position force hybrid control method for tracking unknown curved surface by robot - Google Patents
Position force hybrid control method for tracking unknown curved surface by robot Download PDFInfo
- Publication number
- CN118112991A CN118112991A CN202410248081.0A CN202410248081A CN118112991A CN 118112991 A CN118112991 A CN 118112991A CN 202410248081 A CN202410248081 A CN 202410248081A CN 118112991 A CN118112991 A CN 118112991A
- Authority
- CN
- China
- Prior art keywords
- robot
- force
- curved surface
- hybrid control
- unknown
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000013528 artificial neural network Methods 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 12
- 239000012636 effector Substances 0.000 claims abstract description 7
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000005259 measurement Methods 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 230000003044 adaptive effect Effects 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000011217 control strategy Methods 0.000 claims description 4
- 230000001276 controlling effect Effects 0.000 claims 1
- 230000001105 regulatory effect Effects 0.000 claims 1
- 238000003062 neural network model Methods 0.000 abstract description 5
- 238000013461 design Methods 0.000 abstract description 4
- 238000006073 displacement reaction Methods 0.000 description 10
- 230000000007 visual effect Effects 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 238000004590 computer program Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000014509 gene expression Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Landscapes
- Manipulator (AREA)
Abstract
The invention discloses a position force hybrid control method for tracking an unknown curved surface by a robot, which comprises the following steps: initializing and setting robot parameters; measuring motion trail information of the robot; training data processing: processing the motion trail information of the robot to form a training data set; predicting a future track of the tail end of the robot according to the training data set by using a neural network; performing force-position hybrid control, applying the calculated force to the robot end effector while moving the robot along the unknown curved surface; the neural network is updated in real time, and loop control is performed. According to the method, the future track of the tail end of the robot is predicted through the neural network, hybrid control of the position and the force is introduced, and the accurate tracking of an unknown curved surface is realized by combining a neural network model updated in real time. Through hybrid control design and real-time updating, the stability and adaptability of the robot in an unknown environment are improved, and the limitation of the traditional method is overcome.
Description
Technical Field
The invention relates to the field of robot control, in particular to a method for accurately tracking an unknown curved surface by adopting a bit force hybrid control method under the condition that the tail end of an actuator of a robot contacts the unknown curved surface.
Background
In robotic operations, accurate tracking of unknown surfaces is a challenging task. The conventional control method requires first performing visual recognition of an unknown curved surface and then performing robot operation control.
Current visual recognition methods face some challenges in robotic operations, particularly in the task of accurately tracking unknown surfaces. These challenges are mainly related to the selection of visual sensors, cost and computational complexity. First, visual recognition methods typically require the use of visual sensors to acquire image information in the environment. These sensors include cameras, lidar, etc. devices for capturing visual data of the robot work area. However, high precision vision sensors are often expensive, which may increase the overall cost of the robotic system. For some applications, cost may be a limiting factor. Second, accurate tracking of unknown surfaces typically requires complex operations and algorithms. Visual recognition involves computationally intensive tasks such as image processing, feature extraction, object detection, and the like. These tasks may require powerful computing resources and algorithm support, especially when processing complex environments or large amounts of data. This may result in a robotic system requiring higher performance hardware, increasing the complexity of the overall system.
In addition, the robot operation control needs to satisfy both position and force control, and the existing control method is difficult to cope with complex shapes and changes of unknown curved surfaces. In existing robot control methods, model-based control and conventional feedback control methods are generally employed to achieve control of the position and force of the robot. However, these methods have some drawbacks in the context of facing unknown surfaces. Conventional model-based methods typically require the establishment of an accurate system dynamics model in advance, but the shape and properties of unknown surfaces are difficult to model accurately in advance, resulting in reduced controller performance. The feedback control method has relatively slow response to the change of an unknown environment, and cannot meet the unknown curved surface tracking task with high real-time requirements. Traditional control methods rely on accurate system models, but in the case of unknown surfaces, the models may be difficult to obtain accurately. According to the invention, a deep learning method is adopted, and the characteristics of an unknown curved surface are learned from actual measurement data by training a neural network model, so that dependence on an accurate model is avoided.
Disclosure of Invention
According to the invention, based on the motion trail of the robot, the future trail of the unknown curved surface is predicted by the robot through the neural network, meanwhile, the hybrid control of the bit force is introduced, and the real-time update of the neural network model is combined, so that the robot can continuously optimize the understanding of the unknown curved surface in the motion process, and the real-time performance and the adaptability of a control system are improved.
The invention provides a position force hybrid control method for tracking an unknown curved surface by a robot, which comprises the following steps:
Initializing and setting robot parameters;
measuring motion trail information of the robot;
training data processing: processing the motion trail information of the robot to form a training data set;
predicting a future track of the tail end of the robot according to the training data set by using a neural network;
performing force-position hybrid control, applying the calculated force to the robot end effector while moving the robot along the unknown curved surface;
The neural network is updated in real time, and loop control is performed.
Specifically, the force bit hybrid control introduces a PID regulator and an adaptive law.
Specifically, the control law of the force-bit hybrid control is:
Wherein, X d is the expected position of the robot tip, X c is the current position of the robot tip, F d is the expected contact force, F c is the current contact force, M, B, K p、Kd is the robot control parameter, T is the control period, and λ is the update rate.
Specifically, the control law of the force-bit hybrid control is:
wherein, X d is the expected position of the robot end, X c is the current position of the robot end, M, B, K is the robot control parameter, I is the identity matrix, and R is the diagonal matrix with matrix elements of 1 or 0.
Specifically, the force-bit hybrid control controls the contact force error through a PID regulator, using the adjusted force error as an adaptive law.
Specifically, the force-bit hybrid control dynamically adjusts the control strategy according to the complexity of the unknown curved surface.
Specifically, the neural network is updated in real time, specifically: the neural network is adjusted according to the new data by adopting an online learning or incremental learning method.
Specifically, the online learning method is a random gradient descent algorithm.
Specifically, the measurement of the movement locus information of the robot is completed by a manual teaching operation.
Specifically, the neural network optimizes network parameters according to the gradient of the loss function through a back propagation algorithm.
The invention also provides a computer readable storage medium having stored thereon a computer program for execution by a processor of the steps of the aforementioned method.
The beneficial effects are that: by processing the movement track data of the robot, the requirement on the vision sensor is effectively reduced, the data volume is reduced, and the calculated amount is reduced. Such optimization may make the robotic system lighter, more efficient, and more adaptable to cost-sensitive or resource-constrained environments.
The invention adopts a deep learning method, can predict the unknown curved surface with high precision under a neural network model updated in real time, and ensures that the robot maintains highly accurate position force mixed control in a complex environment. The invention can adapt to the change of an unknown curved surface by continuously updating the neural network model, and meanwhile, the hybrid control strategy ensures the stable contact between the tail end of the robot and the curved surface, thereby improving the stability and the adaptability of the robot in an unknown environment. The invention overcomes the limitation of the traditional model-based and feedback control method, and provides an innovative solution for the bit force hybrid control of the robot on the unknown curved surface by combining the deep learning model and the bit force hybrid control, so that the method is more suitable for complex and dynamic working environments.
Drawings
Fig. 1 is a schematic diagram of a robot operating on an unknown surface.
Fig. 2 is a flowchart of a bit force hybrid control method for tracking an unknown curved surface by a robot according to the present invention.
Detailed Description
Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the illustrations provided in the following embodiments merely illustrate the basic idea of the present invention by way of illustration, and the following embodiments and features in the embodiments may be combined with each other without conflict.
Wherein the drawings are for illustrative purposes only and are shown in schematic, non-physical, and not intended to limit the invention; for the purpose of better illustrating embodiments of the invention, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the size of the actual product; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The method of the present invention will be further described with reference to the accompanying drawings and examples. As shown in fig. 1, the robot tip operates on an unknown surface, which may be any surface of the workpiece that may have burrs or be uneven.
As shown in fig. 2, the invention provides a position force hybrid control method for tracking an unknown curved surface by a robot, which comprises the following steps:
1. Initializing and setting robot parameters: setting initial joint position and speed, and determining kinetic model parameters of the robot, including a mass matrix, a Ke's matrix, a gravity matrix and the like.
Before the operation of the robot starts, it is necessary to determine the initial joint position and speed of the robot. This may be accomplished by moving the robot to a known location, or by sensors acquiring status information of the current joint. Accurate setting of the initial state is important for control and planning of subsequent movements.
The kinetic model of the robot describes the motion behavior of the robot system, including parameters such as a mass matrix, a coriolis matrix, a gravity matrix, and the like. The exact setting of these parameters is crucial for performing inverse kinematics and motion planning. Typically, these parameters may be obtained by making experimental measurements on the robot or from specifications provided by the manufacturer.
2. Measuring motion trail information of a robot: the motion trail information of the robot is measured, specifically, a force sensor arranged at the tail end of the robot can be used for measuring the contact force, and the position and displacement of the end effector under the world coordinate system are calculated according to the positive kinematics relation of the robot.
A force sensor mounted at the end of the robot is used to measure the contact force of the robot with the environment or work object. These force sensors can provide real-time contact force information that aids in the perception of the robot to the external environment or work object.
Through positive kinematics, the joint state of the robot can be mapped to the position and displacement of the end effector in the world coordinate system, which provides a basis for subsequent motion control and planning.
For complex surfaces or unknown surfaces, it is difficult for a robot to achieve the desired tracking effect. According to the complexity of the curved surface, manual teaching operation can be performed in the early stage or before the operation starts, so that measurement point data can be obtained.
3. Training data processing: and processing the motion trail information of the robot to form a training data set.
During the movement of the robot, the acquired movement track data may contain a large amount of redundant measurement points and redundant information. This results in a large computational burden on the one hand and in a non-smooth or less than expected fit or prediction surface on the other hand. In order to improve the learning efficiency of the neural network and reduce the calculation load, the trajectory data needs to be processed to form an effective training data set. Therefore, on the basis of ensuring the effectiveness of the data, the data of the measurement points are reduced as much as possible, so that the subsequent neural network training is facilitated.
Training data processing, specifically:
(1) Determining a starting point and an ending point of a motion trail to be processed, and adding the starting point and the ending point into a training data set;
(2) Sequentially calculating the distances from other measuring points on the track to the track where the starting point and the ending point are located;
(3) Calculating the maximum distance between the measuring point and the track, and if the maximum distance is smaller than the threshold value, removing all points except the starting point and the ending point; if the maximum distance is greater than the threshold value, segmenting the motion trail at a measuring point where the maximum distance is located;
(4) And (3) repeatedly executing the steps (1) - (3) on all the formed motion trajectories until the maximum distance is larger than a threshold value or the number of measurement points in the current training data set is larger than a preset number, and ending the iteration.
The threshold value is the roughness arithmetic average height of the unknown curved surface, the roughness arithmetic average height is the roughness parameter of the curved surface, and the average value of absolute values of height differences of each point relative to the average surface of the surface is represented. The value may be measured or calculated by a roughness measuring device.
4. Future trajectories of the robot tips are predicted using the back propagation neural network. At this stage, a back propagation neural network is used to learn and predict future trajectories of the robot tips.
The inputs to the back propagation neural network are the aforementioned training data set, and the outputs include a fitted expression of the unknown surface and future displacements of the robot tip. For surface fitting, the neural network may output parameters of the surface, such as polynomial coefficients or parameters of other fitting functions or implicit expressions. For the prediction of future displacement, the neural network outputs positional information of the robot tip at a future time step.
By supervised learning, the difference between the known displacement information and the displacement information predicted by the neural network is used as a loss function. The loss function may take a variety of forms, such as Mean Square Error (MSE), etc., for measuring the difference between the predicted and actual values. Through a back propagation algorithm, the neural network optimizes network parameters according to the gradient of the loss function to reduce the prediction error. This process is iterated until the network is able to obtain good fit and predictive performance on the training data.
Through this process, the neural network is able to learn the fitted expression of the unknown surface and the future trajectory of the robot tip. After training is completed, the network can be used for prediction of unknown surfaces and simulation of future trajectories of robots.
5. And performing force-bit hybrid control, and applying the calculated force to the end effector of the robot while moving the robot along the unknown curved surface.
In the process of tracking an unknown curved surface, the problems of overshoot, large tracking error and the like may exist, and the overshoot of contact force or displacement may damage a robot or a workpiece to be processed or easily generate vibration. The specific control law is as follows:
Wherein, X d is the expected position of the robot tip, X c is the current position of the robot tip, F d is the expected contact force, F c is the current contact force, M, B, K p、Kd is the robot control parameter, T is the control period, and λ is the update rate.
Alternatively, the control law may be:
wherein, X d is the expected position of the robot end, X c is the current position of the robot end, F d is the expected contact force, F c is the current contact force, M, B, K is the robot control parameter, I is the identity matrix, and R is the diagonal matrix with matrix elements of 1 or 0.
When the robot tracks an unknown curved surface, a hybrid control design is introduced, force-based control and position control are combined, and meanwhile, an adaptive law is adopted to reduce overshoot and improve tracking performance. By hybrid control, the robot tip is moved along an unknown surface while maintaining contact and applying the desired normal force to the surface. The hybrid control design realizes control of contact force errors by introducing PID control and self-adaptive law, reduces overshoot, and combines force-based control and position control to improve the tracking performance of the robot on an unknown curved surface. The goal of the hybrid control is to maintain the required contact force while achieving tip position control and to achieve smooth, stable motion during the tracking of an unknown surface.
6. The neural network is updated in real time, and loop control is performed. Specifically, updating the neural network in real time includes: in the motion process of the robot, continuously collecting new sensor data and actual displacement information; updating parameters of the neural network in real time by using new data; and applying the neural network updated in real time to a control system of the robot. The parameters of the neural network can be updated by adopting an online learning or incremental learning method, and the neural network is adjusted according to new data so as to continuously improve the accuracy of the curved surface model. In real-time updating, online learning strategies such as random gradient descent (SGD) or other incremental learning algorithms may be employed to adjust the network weights and offsets stepwise to better adapt the network to the characteristics of the unknown surface.
Specifically, performing loop control includes: predicting by using the updated neural network to obtain a future track or displacement of the tail end of the robot; according to the hybrid control design, the predicted displacement information and the actual sensor measurement data are combined, a control law is updated, and a new control force is calculated; the calculated force is applied to the robotic end effector to effect real-time movement of the robot along the unknown surface while maintaining contact and applying the desired normal force to the surface.
Through the process of updating the neural network and executing the cyclic control in real time, the robot can dynamically adjust the control strategy according to the latest information and the actual motion situation so as to adapt to the change of an unknown curved surface, continuously improve the accuracy of a curved surface prediction model, better adapt to the change of the curved surface and improve the tracking performance and the control precision. This strategy makes the robotic system more adaptive and real-time.
The various methods described above, in some embodiments, may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and/or installed onto the device via the ROM and/or the communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more actions or steps of the method described above may be performed. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for performing aspects of the present disclosure. The computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
The foregoing is a further detailed description of the application in connection with specific/preferred embodiments, and it is not intended that the application be limited to such description. It will be apparent to those skilled in the art that several alternatives or modifications can be made to the described embodiments without departing from the spirit of the application, and these alternatives or modifications should be considered to be within the scope of the application.
Claims (9)
1. A position force hybrid control method for tracking an unknown curved surface by a robot comprises the following steps:
Initializing and setting robot parameters;
measuring motion trail information of the robot;
training data processing: processing the motion trail information of the robot to form a training data set;
predicting a future track of the tail end of the robot according to the training data set by using a neural network;
performing force-position hybrid control, applying the calculated force to the robot end effector while moving the robot along the unknown curved surface;
Updating the neural network in real time and executing circulation control;
the control law of the force-bit mixed control is as follows:
Wherein, X d is the expected position of the robot tip, X c is the current position of the robot tip, F d is the expected contact force, F c is the current contact force, M, B, K p、Kd is the robot control parameter, T is the control period, and λ is the update rate.
2. The bit force hybrid control method for tracking an unknown curved surface by a robot according to claim 1, wherein the control law of the force-bit hybrid control is replaced with:
Wherein, X d is the expected position of the robot end, X c is the current position of the robot end, F d is the expected contact force, F c is the current contact force, M, B, K is the robot control parameter, I is the identity matrix, and R is the diagonal matrix with matrix elements of 1 or 0.
3. The bit force hybrid control method for tracking an unknown surface by a robot according to claim 1, wherein the force bit hybrid control incorporates a PID regulator and an adaptive law.
4. A bit force hybrid control method for tracking an unknown curved surface by a robot according to claim 1 or 3, wherein the force bit hybrid control controls the contact force error by a PID regulator, using the regulated force error as an adaptive law.
5. The bit force hybrid control method for tracking an unknown curved surface by a robot according to claim 1, wherein the force bit hybrid control dynamically adjusts a control strategy according to the complexity of the unknown curved surface.
6. The method for controlling the hybrid position force of the robot tracking the unknown curved surface according to any one of claims 1 to 5, wherein the neural network is updated in real time, specifically: the neural network is adjusted according to the new data by adopting an online learning or incremental learning method.
7. The bit force hybrid control method for tracking an unknown curved surface by a robot according to claim 6, wherein the online learning method is a random gradient descent algorithm.
8. The bit force hybrid control method for tracking an unknown curved surface by a robot according to claim 1, wherein the measurement of the motion trajectory information of the robot is accomplished by a manual teaching operation.
9. The bit force hybrid control method for tracking an unknown curved surface by a robot according to claim 1, wherein the neural network optimizes network parameters according to a gradient of a loss function by a back propagation algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410248081.0A CN118112991A (en) | 2024-03-05 | 2024-03-05 | Position force hybrid control method for tracking unknown curved surface by robot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410248081.0A CN118112991A (en) | 2024-03-05 | 2024-03-05 | Position force hybrid control method for tracking unknown curved surface by robot |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118112991A true CN118112991A (en) | 2024-05-31 |
Family
ID=91213484
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410248081.0A Pending CN118112991A (en) | 2024-03-05 | 2024-03-05 | Position force hybrid control method for tracking unknown curved surface by robot |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118112991A (en) |
-
2024
- 2024-03-05 CN CN202410248081.0A patent/CN118112991A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110462531B (en) | Model predictive control system and method for controlling machine operation | |
| JP2022145915A (en) | Inference method, inference program, inference device, learning method, learning program, learning device, and model generation method | |
| WO2020207219A1 (en) | Non-model robot control method for multi-shaft-hole assembly optimized by environmental prediction | |
| US11440183B2 (en) | Hybrid machine learning-based systems and methods for training an object picking robot with real and simulated performance data | |
| US20100114807A1 (en) | Reinforcement learning system | |
| CN111673733B (en) | Intelligent self-adaptive compliance control method of robot in unknown environment | |
| US20190018392A1 (en) | Control apparatus and learning device | |
| US11185980B2 (en) | Machine learning-based systems and methods for controlling robotic object picking and placement in training and run time environments | |
| CN111752151A (en) | Adaptive tracking and compensating method and system for grinding and polishing industrial blade | |
| JP2022543926A (en) | System and Design of Derivative-Free Model Learning for Robotic Systems | |
| Zhang et al. | Robotic curved surface tracking with a neural network for angle identification and constant force control based on reinforcement learning | |
| CN116533249A (en) | Mechanical arm control method based on deep reinforcement learning | |
| Navarro-Alarcon et al. | Visually servoed deformation control by robot manipulators | |
| JP2023523656A (en) | Apparatus and method for planning contact interaction trajectories | |
| JP2019185742A (en) | Controller and control method | |
| CN114571444A (en) | Q-learning-based impedance control method for slagging-off robot | |
| Gao et al. | Time-optimal trajectory planning of industrial robots based on particle swarm optimization | |
| Jandaghi et al. | Motion dynamics modeling and fault detection of a soft trunk robot | |
| Tong et al. | Neural network based visual servo control under the condition of heavy loading | |
| CN117086877A (en) | Industrial robot shaft hole assembly method, device and equipment based on deep reinforcement learning | |
| CN118112991A (en) | Position force hybrid control method for tracking unknown curved surface by robot | |
| CN116834014A (en) | Intelligent cooperative control method and system for capturing non-cooperative targets by space dobby robot | |
| CN114080304B (en) | Control device, control method, and control program | |
| CN117817674A (en) | Self-adaptive impedance control method for robot | |
| JP2018140471A (en) | Control device and machine learning device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |