Autonomous navigation and positioning method for semiconductor mechanical arm based on AI vision

Abstract

The invention discloses a semiconductor mechanical arm autonomous navigation and positioning method based on AI vision, and relates to the technical field of mechanical arms, and the method comprises the following specific steps: visual perception and multi-modal information fusion: collecting multi-modal data through configured industrial, depth, light field and polarized light cameras, pre-processing the multi-modal data, and carrying out multi-modal information fusion; realizing information fusion by using a deep learning network containing a feature extraction layer and a fusion layer and an attention mechanism, and outputting a unified visual description; according to the method, a deep learning target detection model is utilized, an attention mechanism is introduced to enhance the attention and extraction capability of small target features, the recognition result is optimized in combination with surface feature information recognized by polarized light imaging, the category and position of a target object can be accurately determined, meanwhile, a mechanical arm motion error model is established, and the recognition accuracy is improved. And the fused visual positioning information is fused with sensor data such as a mechanical arm joint encoder and a gyroscope by utilizing a sensor fusion technology, and a positioning result is optimized and compensated.

Description

Technical Field

[0001] The present invention relates to the field of robotic arm technology, and specifically to an autonomous navigation and positioning method for a semiconductor robotic arm based on AI vision. Background Art

[0002] In the field of semiconductor manufacturing, robotic arms play a vital role, undertaking core tasks such as wafer handling, chip picking and placement. As semiconductor manufacturing processes continue to move towards high precision and high integration, stringent requirements are placed on the navigation and positioning accuracy of robotic arms. High-precision navigation and positioning are key factors in ensuring accurate operation of wafers and chips and improving production efficiency and product quality. In this context, how to improve the autonomous navigation and positioning capabilities of robotic arms in complex production environments has become a key technical issue that needs to be urgently addressed in the semiconductor manufacturing industry.

[0003] The application of traditional robotic arms in semiconductor manufacturing mainly relies on preset trajectories and fixed positioning methods. This model has great limitations when it comes to the complex and changeable semiconductor production environment. For example, when the equipment layout is adjusted, traditional robotic arms cannot quickly adapt to the new environment due to their lack of autonomous adaptability, and need to re-perform tedious trajectory setting and positioning adjustments. This not only consumes a lot of time and manpower, but also leads to a significant reduction in operational efficiency. When there is a deviation in the wafer position or when tiny chips need to be precisely grasped, the positioning accuracy of traditional robotic arms is far from meeting the requirements, and grasping errors are prone to occur, seriously affecting production progress and product quality. In addition, traditional visual perception technology has obvious deficiencies in acquiring scene information. It is difficult to obtain the three-dimensional structure and light direction information of the scene. The ability to recognize subtle surface features of semiconductor components is also extremely limited, and it cannot provide the robotic arm with sufficiently rich and accurate visual input, making the operation of the robotic arm in complex environments face many difficulties. Summary of the Invention

[0004] The purpose of the present invention is to make up for the shortcomings of the existing technology and provide an autonomous navigation and positioning method for a semiconductor robotic arm based on AI vision. It can give the robotic arm powerful visual perception capabilities by introducing light field imaging and polarized light imaging technology. In the visual perception link, it is equipped with a variety of professional cameras to obtain multimodal image data, and uses deep learning algorithms for comprehensive processing to form an accurate visual description. In terms of target recognition and positioning, the improved deep learning target detection model is trained using labeled multimodal image data, combined with the surface feature information of polarized light imaging recognition, to accurately determine the category and position of the target object, and calculate its precise coordinates in three-dimensional space. At the same time, a real-time environmental map is constructed based on the fused visual data, and a path planning algorithm is used to plan the optimal path for the robotic arm, and the path is adjusted in real time during the navigation process to cope with dynamic obstacles. In addition, a robotic arm motion error model is established, and the positioning results are optimized and compensated through multi-source data fusion technology, thereby comprehensively improving the autonomous navigation and positioning capabilities of the robotic arm in a complex semiconductor production environment.

[0005] To solve the above technical problems, the present invention provides the following technical solution: an AI vision-based autonomous navigation and positioning method for a semiconductor robot arm, the method comprising the following specific steps: Visual perception and multimodal information fusion: Multimodal data is collected through configured industrial, depth, light field, and polarization cameras. After preprocessing, a deep learning network with feature extraction and fusion layers and an attention mechanism are used to achieve information fusion and output a unified visual description. Target recognition and positioning model training: Build and partition a multimodal dataset covering various targets. Build and train a target detection model with an attention mechanism based on a deep learning framework. Calculate the target's three-dimensional coordinates by combining coordinate mapping with multi-source information. Environmental modeling and map construction: After the robotic arm starts, it collects data to extract feature points and build an initial map. Feature points and annotation information are updated during movement. The map is managed using an octree and regularly optimized. Path planning and navigation control: Plans and smoothes the path based on the environment map and target location, locally adjusts the path when dynamic obstacles are detected, and adjusts the robot's motion parameters through visual feedback for precise navigation; Positioning accuracy optimization and compensation: After testing, a robotic arm motion error model is established, multi-source data is integrated and the compensation amount is calculated, motion instructions are adjusted with real-time feedback, and the model is calibrated regularly to ensure positioning accuracy.

[0006] Furthermore, in the visual perception and multimodal information fusion step, the industrial camera is used to obtain two-dimensional image information of the environment; the depth camera captures the depth information of the scene in real time; the light field camera uses light field imaging technology to obtain the three-dimensional structure and light direction information of the scene; and the polarized light camera uses the sensitive detection capability of polarized light imaging to detect the surface characteristics of different materials to identify subtle features on the surface of semiconductor components.

[0007] Furthermore, in the step of visual perception and multimodal information fusion, after preprocessing the data, a deep learning network with a feature extraction layer and a fusion layer is constructed. The extraction layer obtains the deep features of the light field and polarized light through convolution and pooling, and the fusion layer adopts the attention mechanism according to the formula Dynamic weighted fusion, where is the fused feature vector, It is The basic weights of modal features, represents the light field imaging characteristics, Represents the polarized light imaging characteristics, For the The deep feature vector of the modality, is the sensitivity coefficient, For the The saliency values ​​of the modal features are combined and unified visual description information is finally output.

[0008] Furthermore, in the target recognition and positioning model training step, the target three-dimensional coordinates are calculated by combining coordinate mapping with multi-source information. Specifically, the target positioning calculation combines the image position information output by the model with the camera calibration parameters to establish a coordinate mapping, fuse the depth and light field three-dimensional information, and according to the formula Get the three-dimensional coordinates of the target in the robotic arm coordinate system, where is the target three-dimensional coordinate, is the coordinate transformation coefficient, is the camera intrinsic parameter matrix, is the target image coordinate, is the depth information fusion coefficient, is the target depth value, is the three-dimensional structure correction factor.

[0009] Furthermore, in the environmental modeling and map construction step, after the robot arm starts at the initial position, the visual perception module collects the surrounding multimodal fusion data, extracts the edge feature points of the equipment and matches them to determine the initial environmental features, and builds the initial map frame that records the three-dimensional coordinates based on the initial feature points. The robot arm collects new data in real time during movement to extract feature points and matches them with the existing feature points on the map. According to the formula Update coordinate information, where for The set of map feature points at each moment, is the historical feature weight coefficient, for The set of map feature points at each moment, for The new feature point set is collected at every moment. To increase the credibility factor of feature matching, the location range of obstacles and device boundaries is marked at the same time. An octree structure is used to divide the three-dimensional space into cubic units and mark the existence of objects to manage the map in layers. The back-end optimization algorithm is used regularly to globally optimize the map to eliminate accumulated errors to ensure consistency with the actual environment.

[0010] Furthermore, in the path planning and navigation control steps, based on the constructed environmental map, the starting and target positions are determined according to the target identification and positioning results, and a feasible path is searched for based on the comprehensive path length, energy consumption and safety, and the sharp corners are removed by smoothing to make it conform to the kinematic constraints and ensure smoothness. In the dynamic path adjustment, the environment is monitored in real time, and when a dynamic obstacle is detected, its position, speed and direction are obtained. A real-time adjustment algorithm is used in combination with the state of the robotic arm to re-plan the path locally, and the navigation control obtains the current position and posture of the robotic arm based on the fusion of visual feedback, compares it with the planned path to calculate the deviation, adjusts the speed and acceleration of each joint, and reduces the speed when approaching the target to improve the positioning accuracy.

[0011] Furthermore, in the path planning and navigation control steps, a real-time adjustment algorithm is used in combination with the robot arm state to replan the path locally. The algorithm formula is: ,in, is the path correction amount, is the path adjustment coefficient, is the moving velocity vector of the dynamic obstacle, is the real-time distance between the robotic arm and the obstacle, is the current movement speed of the robotic arm, is the gradient of the path cost function.

[0012] Furthermore, in the positioning accuracy optimization and compensation step, the robot arm is subjected to multiple groups of motion tests at different positions and postures, and the theoretical data calculated by the control instructions and the actual data measured by the sensor at each test point are collected. The relationship between the deviation and the joint angle and the motion speed is analyzed, and the least squares method is used to fit the motion error model that describes the error distribution of different working conditions. The fused visual positioning information, joint encoder data and gyroscope attitude data are collected, and the data are weightedly fused by establishing the state equation and the observation equation. According to the error model and the fused positioning result, according to the formula Calculate the error compensation of the current position and attitude, where is the positioning error compensation, is the error model weight coefficient, is the theoretical error value, is the number of sensors, For the sensor error weights, For the Each sensor measures the error value and feeds it back to the control system in real time to adjust the motion command to correct the error. The calibration error model is also updated regularly to ensure long-term compensation effect.

[0013] Compared with existing technologies, this AI vision-based semiconductor robotic arm autonomous navigation and positioning method has the following beneficial effects: 1. This invention introduces light field imaging and polarized light imaging technologies and integrates multimodal visual information, enabling the robotic arm to obtain rich and accurate scene information in real time. The three-dimensional structure and light direction information obtained by the light field camera, and the ability of the polarized light camera to identify subtle surface features of semiconductor components, enable the robotic arm to fully perceive the surrounding environment. The real-time map constructed based on this information can accurately mark and update obstacles and equipment boundaries. When faced with environmental changes, the robotic arm can quickly plan a reasonable path, avoid obstacles, and accurately reach the target location, greatly improving production efficiency and ensuring the stable operation of the semiconductor manufacturing process.

[0014] 2. The present invention utilizes a deep learning target detection model and introduces an attention mechanism to enhance the ability to focus on and extract small target features, and optimizes the recognition results in combination with the surface feature information of polarized light imaging recognition, which can accurately determine the category and position of the target object. At the same time, a robotic arm motion error model is established, and sensor fusion technology is used to fuse the fused visual positioning information with the robotic arm joint encoder, gyroscope and other sensor data to optimize and compensate the positioning results, effectively reducing the error rate in operations such as picking up tiny chips, reducing production costs, and improving the quality and output of semiconductor products.

[0015] Other advantages, objects and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be learned from the practice of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0017] Figure 1This is a flowchart of the autonomous navigation and positioning method for semiconductor robotic arms based on AI vision; Figure 2 A flowchart of the environmental modeling and map building steps for the AI ​​vision-based semiconductor robotic arm autonomous navigation and positioning method; Figure 3 This is a flow chart of the path planning and navigation control of the autonomous navigation and positioning method for semiconductor robotic arms based on AI vision. DETAILED DESCRIPTION

[0018] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0019] Example 1 In the semiconductor packaging and testing workshop, tiny chips with a size of 1mm×1mm need to be picked up from the wafer and placed on the lead frame. Before starting the picking work, technicians will pre-process the wafer and fix it on a dedicated wafer table to ensure that the wafer will not move during the picking process. At the same time, the lead frame will be inspected to confirm that its position is accurate and the surface is clean, in preparation for subsequent chip placement.

[0020] like Figure 1 As shown, first, multimodal data is collected by configuring industrial, depth, light field and polarization cameras to ensure that each camera can clearly capture the tiny chips on the wafer. Accurate coordinate transformation relationships are established through calibration. When working, the industrial camera obtains the two-dimensional image of the wafer, the depth camera captures the depth information of the chip, the light field camera obtains the three-dimensional structure information of the chip, and the polarization camera identifies subtle defects on the chip surface, such as cracks and stains. After preprocessing the multimodal data, according to the formula (in is the fused feature vector, It is The basic weights of modal features, For the The deep feature vector of the modality, is the sensitivity coefficient, For the The deep features of light field and polarized light imaging data are weightedly fused using the saliency value of the modal features to obtain a fused feature vector which is then input into the target recognition and positioning model.

[0021] The target recognition and positioning model has been optimized for tiny chips during training. The introduced attention mechanism enhances the focus on small target features and can accurately identify the location of each chip on the wafer and whether there are defects. (in is the target three-dimensional coordinate, is the coordinate transformation coefficient, is the camera intrinsic parameter matrix, is the target image coordinate, is the depth information fusion coefficient, is the target depth value, is the three-dimensional structure correction factor) and the three-dimensional coordinates of the chip in the robot coordinate system are calculated, as shown in Figure 2 As shown, the robot arm starts and builds a map of the workshop environment during the initial movement. According to the formula (in for The set of map feature points at each moment, is the historical feature weight coefficient, for The set of map feature points at each moment, for The new feature point set is collected at every moment. The map is marked with the location information of the wafer stage, lead frame, peripheral equipment and obstacles (for feature matching confidence factor), and the map is regularly updated and optimized to ensure its accuracy.

[0022] When receiving the chip picking task instruction, Figure 3 As shown in the figure, the robot arm plans a global path based on the position of the chip and the target position of the lead frame. During the movement, the path is adjusted in real time through visual feedback. If dynamic obstacles such as temporary movement of staff and short-distance movement of other equipment are detected, the robot arm will adjust the path according to the formula (in is the path correction amount, is the path adjustment coefficient, is the moving velocity vector of the dynamic obstacle, is the real-time distance between the robotic arm and the obstacle, is the current movement speed of the robotic arm, The path is adjusted based on the gradient of the path cost function. After reaching the wafer table, the robot arm pauses for a moment and reconfirms the position and status of the target chip through the visual perception module. Then, the position and posture of the end effector are precisely adjusted based on the positioning information. Utilizing high-precision positioning and gripping mechanisms, vacuum adsorption is used to pick up qualified chips (avoiding defective chips). During the adsorption process, the adsorption force is monitored in real time to ensure that the chip can be firmly adsorbed without causing damage.

[0023] After picking up the chip, the robot arm moves it to the lead frame. During this period, the light field camera monitors the chip's posture in real time to ensure that the chip does not tilt or fall during the transportation process. If a slight deviation in the posture is found, it will be corrected in time by adjusting the robot arm joint parameters. At the same time, the positioning accuracy is continuously optimized and compensated. According to the formula (in is the positioning error compensation, is the error model weight coefficient, is the theoretical error value, is the number of sensors, For the sensor error weights, For the The robot adjusts the motion parameters based on the multi-source data fusion results (the measurement error value of each sensor) to ensure that the chip can be accurately placed on the designated pad of the lead frame. During placement, the robot arm will slowly lower the end effector. When the chip approaches the pad, the suction force is gradually reduced until the chip is stably placed on the pad.

[0024] After placement is completed, the robotic arm will not leave immediately, but will check the placement position and status of the chip through the visual perception module. After confirmation, it will return to the wafer table to continue picking up the next chip and repeat the above process until all qualified chips are picked up and placed. If unqualified placement is found, an alarm will be issued in time to notify the staff to handle it.

[0025] Example 2 In a semiconductor wafer manufacturing workshop, wafers with a diameter of 300mm need to be transported from the lithography equipment to the etching equipment. First, in accordance with the requirements for visual perception module installation and multimodal fusion debugging, a high-resolution industrial camera, a depth camera, a light field camera, and a polarization camera are installed near the end effector of the robotic arm. The camera parameters are adjusted to adapt to the lighting conditions in the workshop. The internal and external parameters of each camera are calibrated using the checkerboard calibration method, and the conversion relationship between the image coordinates and the actual physical coordinates is established. Then, the robotic arm is started, and the industrial camera obtains a two-dimensional image of the wafer in the lithography equipment, the depth camera captures the depth information of the wafer, the light field camera obtains the three-dimensional structure and light direction information of the wafer and the surrounding environment, and the polarization camera identifies whether there are subtle scratches and other features on the wafer surface. After preprocessing these multimodal data, a feature fusion network based on deep learning is used to weight the deep features of the light field and polarization imaging data through the attention mechanism to obtain a fused feature vector. The trained target recognition and positioning model is used to process the fused visual information, accurately identify the position of the wafer, and combine the image and physical coordinate mapping relationship as well as the depth and three-dimensional structure information to calculate the three-dimensional coordinates of the wafer in the robotic arm coordinate system. At the same time, during the initial movement of the robotic arm, it collects fused multimodal data and gradually constructs a real-time map of the workshop environment including lithography equipment, etching equipment and surrounding obstacles based on simultaneous positioning and map construction technology, and uses the octree data structure to manage the map in layers. When the robotic arm receives the task instruction to move the wafer, it plans a global path in the constructed environmental map based on the current position of the wafer and the target position of the etching equipment. During the movement of the robotic arm, it monitors environmental changes in real time through fused visual feedback. If dynamic obstacles such as moving material carts in the workshop are detected, it immediately switches to a real-time path adjustment algorithm to make local adjustments to the path. At the same time, the movement speed and posture of the robotic arm are continuously adjusted according to the visual feedback information to ensure that the robotic arm reaches the lithography equipment smoothly and accurately. After arriving at the lithography equipment, the robotic arm adjusts the posture of the end effector according to the positioning information and grabs the wafer. During the grabbing process, the wafer surface is monitored in real time through a polarized light camera to ensure that no new scratches are caused by the grabbing operation. Then, the robotic arm transports the wafer to the etching equipment according to the planned path, and continuously optimizes and compensates for the positioning accuracy during this period. According to the established robotic arm motion error model and multi-source data fusion results, the motion parameters are adjusted in real time, and the wafer is finally placed accurately at the designated position of the etching equipment.

[0026] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. The semiconductor robot arm autonomous navigation and positioning method based on AI vision is characterized by: The method comprises the following specific steps: Visual perception and multimodal information fusion: Multimodal data is collected through configured industrial, depth, light field, and polarization cameras. After preprocessing, a deep learning network with feature extraction and fusion layers and an attention mechanism are used to achieve information fusion and output a unified visual description. Target recognition and positioning model training: Build and partition a multimodal dataset covering various targets. Build and train a target detection model with an attention mechanism based on a deep learning framework. Calculate the target's three-dimensional coordinates by combining coordinate mapping with multi-source information. Environmental modeling and map construction: After the robotic arm starts, it collects data to extract feature points and build an initial map. Feature points and annotation information are updated during movement. The map is managed using an octree and regularly optimized. Path planning and navigation control: Plans and smoothes the path based on the environment map and target location, locally adjusts the path when dynamic obstacles are detected, and adjusts the robot's motion parameters through visual feedback for precise navigation; Positioning accuracy optimization and compensation: After testing, a robotic arm motion error model is established, multi-source data is integrated and the compensation amount is calculated, motion instructions are adjusted with real-time feedback, and the model is calibrated regularly to ensure positioning accuracy.

2. The semiconductor robot autonomous navigation and positioning method based on AI vision according to claim 1 is characterized in that: In the visual perception and multimodal information fusion step, the industrial camera is used to obtain two-dimensional image information of the environment; the depth camera captures the depth information of the scene in real time; the light field camera uses light field imaging technology to obtain the three-dimensional structure and light direction information of the scene; and the polarized light camera uses the sensitive detection ability of polarized light imaging to detect the surface characteristics of different materials to identify subtle features on the surface of semiconductor components.

3. The semiconductor robot autonomous navigation and positioning method based on AI vision according to claim 1, characterized in that: In the visual perception and multimodal information fusion step, after preprocessing the data, a deep learning network with a feature extraction layer and a fusion layer is constructed. The extraction layer obtains the deep features of the light field and polarized light through convolution and pooling. The fusion layer adopts the attention mechanism. According to the formula Dynamic weighted fusion, where is the fused feature vector, It is The basic weights of modal features, represents the light field imaging characteristics, Represents the polarized light imaging characteristics, For the The deep feature vector of the modality, is the sensitivity coefficient, For the The saliency values ​​of the modal features are combined and unified visual description information is finally output.

4. The semiconductor robot autonomous navigation and positioning method based on AI vision according to claim 1, characterized in that: In the target recognition and positioning model training step, the target three-dimensional coordinates are calculated by combining coordinate mapping with multi-source information. Specifically, the target positioning calculation combines the image position information output by the model with the camera calibration parameters to establish a coordinate mapping, fuse the depth and light field three-dimensional information, and according to the formula Get the three-dimensional coordinates of the target in the robotic arm coordinate system, where is the target three-dimensional coordinate, is the coordinate transformation coefficient, is the camera intrinsic parameter matrix, is the target image coordinate, is the depth information fusion coefficient, is the target depth value, is the three-dimensional structure correction factor.

5. The semiconductor robot autonomous navigation and positioning method based on AI vision according to claim 1, characterized in that: In the environmental modeling and map construction step, after the robot arm starts at the initial position, the visual perception module collects the surrounding multimodal fusion data, extracts the edge feature points of the equipment and matches them to determine the initial environmental features, and builds the initial map frame that records the three-dimensional coordinates based on the initial feature points. The robot arm collects new data in real time during movement to extract feature points and matches them with the existing feature points on the map. According to the formula Update coordinate information, where for The set of map feature points at each moment, is the historical feature weight coefficient, for The set of map feature points at each moment, for The new feature point set is collected at every moment. To increase the credibility factor of feature matching, the location range of obstacles and device boundaries is marked at the same time. An octree structure is used to divide the three-dimensional space into cubic units and mark the existence of objects to manage the map in layers. The back-end optimization algorithm is used regularly to globally optimize the map to eliminate accumulated errors to ensure consistency with the actual environment.

6. The semiconductor robot autonomous navigation and positioning method based on AI vision according to claim 1, characterized in that: In the path planning and navigation control steps, based on the constructed environment map, the starting and target positions are determined according to the target identification and positioning results. A feasible path is searched for based on the comprehensive consideration of path length, energy consumption and safety, and the path is smoothed to remove sharp corners so that it complies with kinematic constraints and ensures stability. In the dynamic path adjustment, the environment is monitored in real time. When a dynamic obstacle is detected, its position, speed and direction are obtained. A real-time adjustment algorithm is used in combination with the state of the manipulator to locally replan the path. The navigation control obtains the current position and posture of the manipulator based on the fusion of visual feedback, compares it with the planned path to calculate the deviation, adjusts the speed and acceleration of each joint, and reduces the speed when approaching the target to improve the positioning accuracy.

7. The AI ​​vision-based semiconductor robotic arm autonomous navigation and positioning method according to claim 6, characterized in that: In the path planning and navigation control steps, a real-time adjustment algorithm is used to replan the path locally in combination with the state of the manipulator. The algorithm formula is: ,in, is the path correction amount, is the path adjustment coefficient, is the moving velocity vector of the dynamic obstacle, is the real-time distance between the robotic arm and the obstacle, is the current movement speed of the robotic arm, is the gradient of the path cost function.

8. The semiconductor robot autonomous navigation and positioning method based on AI vision according to claim 1, characterized in that: In the positioning accuracy optimization and compensation step, the robot arm is subjected to multiple groups of motion tests at different positions and postures, and the theoretical data calculated by the control instructions and the actual data measured by the sensor at each test point are collected. The relationship between the deviation and the joint angle and the motion speed is analyzed, and the least squares method is used to fit the motion error model that describes the error distribution of different working conditions. The fused visual positioning information, joint encoder data and gyroscope attitude data are collected, and the data are weightedly fused by establishing the state equation and the observation equation. According to the error model and the fused positioning result, according to the formula Calculate the error compensation of the current position and attitude, where is the positioning error compensation, is the error model weight coefficient, is the theoretical error value, is the number of sensors, For the sensor error weights, For the Each sensor measures the error value and feeds it back to the control system in real time to adjust the motion command to correct the error. The calibration error model is also updated regularly to ensure long-term compensation effect.

Images