Automatic robot positioning method based on vision assistance

Abstract

The invention relates to the technical field of robot positioning, in particular to an automatic robot positioning method based on vision assistance, which comprises a vision acquisition module (1), a multi-modal data fusion module (2), a deep learning feature extraction module (3), a self-adaptive positioning optimization module (4) and an execution control module (5). The visual collection module (1) collects multi-source data through the camera (11), the infrared sensor (12) and the laser radar (13), after the multi-mode data fusion module (2) integrates the multi-source data, key features are extracted through the deep learning feature extraction module (3), then the self-adaptive positioning optimization module (4) dynamically adjusts weights and corrects errors, and finally accurate positioning and navigation are achieved through the execution control module (5). According to the invention, through the adaptive weight adjustment system (24), the adaptive feature enhancement system (34) and the dynamic error compensation system (44), the positioning precision and robustness in a complex environment are significantly improved, and meanwhile, the intelligent obstacle avoidance system (53) ensures safe operation. According to the invention, the problems of insufficient complex environment adaptability, limited positioning precision and low algorithm efficiency in the prior art can be effectively solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot positioning, and in particular relates to an automatic robot positioning method based on vision assistance. Background Art

[0002] With the rapid development of robotics, vision-assisted automatic positioning methods are increasingly being used in industries such as industry, healthcare, and agriculture. However, existing vision-assisted positioning technologies still have limitations in terms of adaptability in complex environments, positioning accuracy, and algorithm efficiency, which hinders the automation level and work efficiency of robots.

[0003] After searching, a method and application of automatic meridian positioning by a robot was disclosed with the publication number CN111437185B, and the publication date is April 8, 2022. This patent obtains the human body's positive projection image through a visual recognition device, and uses color difference analysis to extract edge lines and positioning points, thereby realizing fully automatic meridian positioning. However, this technical solution mainly relies on a single color difference analysis method for positioning, which may lead to a decrease in positioning accuracy when the lighting conditions change or the background is complex. In addition, this method does not fully combine advanced algorithms such as deep learning, and has limited ability to extract image features, which may affect its applicability in complex scenes.

[0004] After searching, a visual positioning method for the automatic grasping of special-shaped catheters by a robot was disclosed with the publication number CN111546335B on May 14, 2021. This patent achieves precise grasping of special-shaped catheters by establishing the relationship between the manipulator coordinate system and the visual coordinate system and creating a visual positioning template. However, when dealing with targets with complex or variable shapes, the technical solution lacks the flexibility of template matching, which may lead to positioning failure or increased error. At the same time, this method has weak adaptability to changes in the target's posture, and the template needs to be adjusted multiple times to adapt to different postures, which increases the complexity and computational burden of the system.

[0005] The above issues demonstrate that existing vision-assisted automatic robot positioning methods still have limitations in terms of adaptability to complex environments, positioning accuracy optimization, and algorithm flexibility. Therefore, the present invention provides a vision-assisted automatic robot positioning method. This method aims to improve the robustness, accuracy, and efficiency of the positioning system by combining multimodal data fusion, deep learning feature extraction, and adaptive algorithm optimization, thereby meeting the demand for high-precision, intelligent robot positioning in various fields. Summary of the Invention

[0006] To address the challenges of existing technologies, this paper provides a vision-assisted automated robot positioning method that addresses the existing challenges of vision-assisted positioning technology, including insufficient adaptability to complex environments, limited positioning accuracy, and inefficient algorithms. Through multimodal data fusion, deep learning feature extraction, and adaptive algorithm optimization, this paper aims to construct a novel solution with high robustness, high precision, and high efficiency, addressing the demand for intelligent robot positioning in industries such as industry, healthcare, and agriculture.

[0007] The technical solution adopted by the present invention to solve the above technical problems is: a vision-assisted automatic robot positioning method, including a vision acquisition module, a multimodal data fusion module, a deep learning feature extraction module, an adaptive positioning optimization module and an execution control module, the vision acquisition module is electrically connected to the multimodal data fusion module, the multimodal data fusion module is electrically connected to the deep learning feature extraction module, the deep learning feature extraction module is electrically connected to the adaptive positioning optimization module, and the adaptive positioning optimization module is electrically connected to the execution control module; the vision acquisition module includes multiple cameras, infrared sensors and laser radars, the multimodal data fusion module includes a data preprocessing unit, a feature mapping unit and a data integration unit, the deep learning feature extraction module includes a convolutional neural network processor, a feature pyramid network and an attention mechanism unit, the adaptive positioning optimization module includes a dynamic weight allocation unit, an error compensation unit and a path planning unit, and the execution control module includes a motion controller and a drive unit.

[0008] The visual acquisition module is installed at key positions on the top and around the robot. Multiple cameras are used to collect visible light images, infrared sensors are used to detect thermal radiation information, and lidar is used to obtain three-dimensional point cloud data. The data collected in real time by the above sensors is transmitted to the multimodal data fusion module through the data transmission interface, and the data preprocessing unit performs noise filtering and data standardization.

[0009] In the multimodal data fusion module, the data preprocessing unit denoises and unifies the format of the received multi-source data, the feature mapping unit maps the data of different modalities to the same feature space, and the data integration unit integrates the features of different modalities through a weighted fusion algorithm to generate a unified multimodal feature representation, which is then transmitted to the deep learning feature extraction module.

[0010] In the deep learning feature extraction module, the convolutional neural network processor performs preliminary feature extraction on the multimodal feature representation, the feature pyramid network performs multi-scale decomposition and enhancement on the extracted features, and the attention mechanism unit weights and strengthens the key features, ultimately generating a high-resolution and highly discriminative feature map, which is transmitted to the adaptive positioning optimization module. In the adaptive positioning optimization module, the dynamic weight allocation unit dynamically adjusts the feature weights of different modal data based on current environmental conditions and task requirements. The error compensation unit corrects the current positioning results using historical positioning error data. The path planning unit generates optimal path planning instructions based on the positioning results and transmits them to the execution control module.

[0011] In the execution control module, the motion controller receives the path planning instructions and generates specific motion control signals. The drive unit drives the motor and actuator of the robot according to the motion control signals to complete the positioning and navigation tasks.

[0012] Preferably, the multimodal data fusion module also includes an adaptive weight adjustment system, which includes a weight calculation unit, an environmental analysis unit, and a feedback control unit. The weight calculation unit calculates the weight value of each modal data based on the environmental parameters provided by the environmental analysis unit. The environmental analysis unit generates environmental parameters by analyzing the current lighting conditions, background complexity, and target characteristics. The feedback control unit adjusts the weight value in real time based on the positioning error feedback to ensure the optimal feature fusion effect in a complex environment. The maximum weight range of the weight calculation unit is 0-1, the minimum adjustment step is 0.01, and the response time is 100 milliseconds.

[0013] Preferably, the deep learning feature extraction module also includes an adaptive feature enhancement system, which includes a feature selection unit, a feature amplification unit, and a feature correction unit. The feature selection unit selects key features based on task requirements, the feature amplification unit performs nonlinear amplification on the selected features, and the feature correction unit corrects the current features by comparing them with historical feature data to improve the accuracy and robustness of feature extraction. The feature amplification unit has a maximum amplification factor of 5 times, a minimum adjustment step size of 0.1, and a correction accuracy of ±0.01.

[0014] Preferably, the adaptive positioning optimization module also includes a dynamic error compensation system, which includes an error prediction unit, an error correction unit, and an error feedback unit. The error prediction unit predicts the current positioning error by analyzing historical positioning error data. The error correction unit corrects the positioning result based on the prediction result. The error feedback unit feeds the corrected error data back to the error prediction unit, forming a closed-loop control system to improve positioning accuracy. The error prediction unit has a maximum prediction error range of ±5 cm, a correction accuracy of ±1 mm, and a feedback period of 200 milliseconds.

[0015] Preferably, the execution control module also includes an intelligent obstacle avoidance system, which includes an obstacle detection unit, an obstacle avoidance path planning unit, and an obstacle avoidance execution unit. The obstacle detection unit detects obstacles ahead by analyzing lidar and camera data. The obstacle avoidance path planning unit generates an obstacle avoidance path based on the position and shape of the obstacle. The obstacle avoidance execution unit executes obstacle avoidance actions through the motion controller and drive unit to ensure the safe operation of the robot in complex environments. The obstacle detection unit has a detection range of 0-10 meters, the obstacle avoidance path planning unit has a maximum path length of 5 meters, and the obstacle avoidance execution unit has a response time of 300 milliseconds.

[0016] The adaptive weight adjustment system is connected to the multimodal data fusion module via the weight calculation unit and the environmental analysis unit, and to the adaptive positioning optimization module via the feedback control unit, adjusting the weight values ​​of each modal data in real time. The adaptive feature enhancement system is connected to the deep learning feature extraction module via the feature selection unit and the feature amplification unit, and to the adaptive positioning optimization module via the feature correction unit, enhancing and correcting feature data in real time. The dynamic error compensation system is connected to the adaptive positioning optimization module via the error prediction unit and the error correction unit, and to the execution control module via the error feedback unit, correcting positioning errors in real time. The intelligent obstacle avoidance system is connected to the execution control module via the obstacle detection unit and the obstacle avoidance path planning unit, and to the motion controller and drive unit via the obstacle avoidance execution unit, detecting and avoiding obstacles in real time.

[0017] The structural composition, implementation method, and operating principle of the present invention are as follows: the adaptive weight adjustment system monitors the current environmental conditions in real time through the environmental analysis unit, the weight calculation unit calculates the weight value of each modal data based on the environmental parameters, and the feedback control unit adjusts the weight value in real time based on the positioning error feedback, ensuring the optimal feature fusion effect in complex environments. The adaptive feature enhancement system selects key features through the feature selection unit, the feature amplification unit performs nonlinear amplification on the selected features, and the feature correction unit corrects the current features by comparing historical feature data, improving the accuracy and robustness of feature extraction. The dynamic error compensation system analyzes historical positioning error data through the error prediction unit to predict the current positioning error, the error correction unit corrects the positioning result based on the prediction result, and the error feedback unit feeds the corrected error data back to the error prediction unit to form a closed-loop control system to improve positioning accuracy. The intelligent obstacle avoidance system detects obstacles ahead through the obstacle detection unit, the obstacle avoidance path planning unit generates the obstacle avoidance path, and the obstacle avoidance execution unit executes the obstacle avoidance action through the motion controller and drive unit, ensuring the safe operation of the robot in complex environments.

[0018] Beneficial effects of the present invention: The adaptive weight adjustment system significantly improves the feature fusion effect in complex environments by dynamically adjusting the weight values ​​of each modal data, solving the limitations of single modal data in situations with changing lighting or complex backgrounds. The adaptive feature enhancement system significantly improves the accuracy and robustness of feature extraction by nonlinearly amplifying and correcting key features, solving the problem of insufficient feature extraction capabilities of traditional methods in complex scenarios. The dynamic error compensation system significantly improves positioning accuracy by correcting positioning errors in real time through closed-loop control, solving the problem of positioning error accumulation in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of the connection of the overall system modules of the present invention;

[0020] Figure 2 It is a structural diagram of the multimodal data fusion module and the adaptive weight adjustment system of the present invention;

[0021] Figure 3 It is a structural diagram of the deep learning feature extraction module and the adaptive feature enhancement system of the present invention;

[0022] In the figure: 1. Visual acquisition module; 11. Camera; 12. Infrared sensor; 13. LiDAR; 14. Data transmission interface; 2. Multimodal data fusion module; 21. Data preprocessing unit; 22. Feature mapping unit; 23. Data integration unit; 24. Adaptive weight adjustment system; 241. Weight calculation unit; 242. Environment analysis unit; 243. Feedback control unit; 3. Deep learning feature extraction module; 31. Convolutional neural network processor; 32. Feature pyramid network; 33. Attention mechanism unit; 34. Adaptive feature enhancement system ;341. Feature selection unit;342. Feature amplification unit;343. Feature correction unit;4. Adaptive positioning optimization module;41. Dynamic weight allocation unit;42. Error compensation unit;43. Path planning unit;44. Dynamic error compensation system;441. Error prediction unit;442. Error correction unit;443. Error feedback unit;5. Execution control module;51. Motion controller;52. Drive unit;53. Intelligent obstacle avoidance system;531. Obstacle detection unit;532. Obstacle avoidance path planning unit;533. Obstacle avoidance execution unit. DETAILED DESCRIPTION

[0023] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0024] The present invention addresses the challenges of existing vision-assisted positioning technologies, such as insufficient adaptability to complex environments, limited positioning accuracy, and inefficient algorithms, by providing a vision-assisted automated robot positioning method. By integrating multimodal data, deep learning feature extraction, and adaptive algorithm optimization, the present invention creates a novel solution with high robustness, high precision, and high efficiency, meeting the demand for intelligent robot positioning in industries such as industry, healthcare, and agriculture.

[0025] See Figures 1 to 3 A vision-assisted automatic robot positioning method includes a vision acquisition module 1, a multimodal data fusion module 2, a deep learning feature extraction module 3, an adaptive positioning optimization module 4 and an execution control module 5, wherein the vision acquisition module 1 is electrically connected to the multimodal data fusion module 2, the multimodal data fusion module 2 is electrically connected to the deep learning feature extraction module 3, the deep learning feature extraction module 3 is electrically connected to the adaptive positioning optimization module 4, and the adaptive positioning optimization module 4 is electrically connected to the execution control module 5.

[0026] See Figure 1 The visual acquisition module 1 is installed at key locations on and around the robot. It includes multiple cameras 11, infrared sensors 12, and lidar 13. These sensors transmit real-time collected data to the multimodal data fusion module 2 via a data transmission interface 14. The multiple cameras 11 are used to capture visible light images, the infrared sensors 12 are used to detect thermal radiation information, and the lidar 13 is used to obtain three-dimensional point cloud data. The cameras 11, infrared sensors 12, and lidar 13 are distributed on and around the robot to ensure comprehensive coverage of the environment.

[0027] See Figure 2The multimodal data fusion module 2 includes a data preprocessing unit 21, a feature mapping unit 22, a data integration unit 23, and an adaptive weight adjustment system 24. The data preprocessing unit 21 performs denoising and formatting on the received multi-source data. The feature mapping unit 22 maps data from different modalities into the same feature space. The data integration unit 23 integrates the features of different modalities using a weighted fusion algorithm to generate a unified multimodal feature representation, which is then transmitted to the deep learning feature extraction module 3. The adaptive weight adjustment system 24 includes a weight calculation unit 241, an environment analysis unit 242, and a feedback control unit 243. The weight calculation unit 241 calculates the weight values ​​of each modal data based on the environmental parameters provided by the environment analysis unit 242. The environment analysis unit 242 generates environmental parameters by analyzing current lighting conditions, background complexity, and target characteristics. The feedback control unit 243 adjusts the weight values ​​in real time based on positioning error feedback to ensure optimal feature fusion in complex environments. The maximum weight range of the adaptive weight adjustment system 24 is 0-1, the minimum adjustment step is 0.01, and the response time is 100 milliseconds.

[0028] See Figure 3 The deep learning feature extraction module 3 includes a convolutional neural network processor 31, a feature pyramid network 32, an attention mechanism unit 33, and an adaptive feature enhancement system 34. The convolutional neural network processor 31 performs preliminary feature extraction on the multimodal feature representation, the feature pyramid network 32 performs multi-scale decomposition and enhancement on the extracted features, and the attention mechanism unit 33 performs weighted enhancement on key features, ultimately generating a high-resolution and highly discriminative feature map, which is transmitted to the adaptive positioning optimization module 4. The adaptive feature enhancement system 34 includes a feature selection unit 341, a feature amplification unit 342, and a feature correction unit 343. The feature selection unit 341 selects key features based on task requirements, the feature amplification unit 342 performs nonlinear amplification on the selected features, and the feature correction unit 343 corrects the current features by comparing them with historical feature data to improve the accuracy and robustness of feature extraction. The feature amplification unit 342 has a maximum amplification factor of 5 times, a minimum adjustment step size of 0.1, and a correction accuracy of ±0.01.

[0029] See Figure 1The adaptive positioning optimization module 4 includes a dynamic weight allocation unit 41, an error compensation unit 42, a path planning unit 43, and a dynamic error compensation system 44. The dynamic weight allocation unit 41 dynamically adjusts the feature weights of different modal data based on current environmental conditions and task requirements. The error compensation unit 42 corrects the current positioning result using historical positioning error data. The path planning unit 43 generates optimal path planning instructions based on the positioning result and transmits them to the execution control module 5. The dynamic error compensation system 44 includes an error prediction unit 441, an error correction unit 442, and an error feedback unit 443. The error prediction unit 441 predicts the current positioning error by analyzing historical positioning error data. The error correction unit 442 corrects the positioning result based on the prediction result. The error feedback unit 443 feeds the corrected error data back to the error prediction unit 441, forming a closed-loop control system to improve positioning accuracy. The maximum prediction error range of the error prediction unit 441 is ±5 cm, the correction accuracy is ±1 mm, and the feedback cycle is 200 milliseconds.

[0030] See Figure 1 The execution control module 5 includes a motion controller 51, a drive unit 52 and an intelligent obstacle avoidance system 53. The motion controller 51 receives path planning instructions and generates specific motion control signals. The drive unit 52 drives the robot's motor and actuator according to the motion control signals to complete positioning and navigation tasks. The intelligent obstacle avoidance system 53 includes an obstacle detection unit 531, an obstacle avoidance path planning unit 532 and an obstacle avoidance execution unit 533. The obstacle detection unit 531 detects obstacles in front by analyzing the data of the laser radar 13 and the camera 11. The obstacle avoidance path planning unit 532 generates an obstacle avoidance path according to the position and shape of the obstacle. The obstacle avoidance execution unit 533 executes obstacle avoidance actions through the motion controller 51 and the drive unit 52 to ensure the safe operation of the robot in complex environments. The detection range of the obstacle detection unit 531 is 0-10 meters, the maximum path length of the obstacle avoidance path planning unit 532 is 5 meters, and the response time of the obstacle avoidance execution unit 533 is 300 milliseconds.

[0031] During specific operation, when the robot operates in a complex environment, the camera 11, infrared sensor 12, and lidar 13 in the visual acquisition module 1 collect environmental data in real time and transmit it to the multimodal data fusion module 2 through the data transmission interface 14. The data preprocessing unit 21 in the multimodal data fusion module 2 performs denoising and format unification on the received multi-source data, the feature mapping unit 22 maps data of different modalities to the same feature space, and the data integration unit 23 generates a unified multimodal feature representation through a weighted fusion algorithm. The environmental analysis unit 242 in the adaptive weight adjustment system 24 monitors the current environmental conditions in real time, the weight calculation unit 241 calculates the weight value of each modal data based on the environmental parameters, and the feedback control unit 243 adjusts the weight value in real time based on the positioning error feedback to ensure the optimal feature fusion effect in a complex environment.

[0032] Subsequently, the convolutional neural network processor 31 in the deep learning feature extraction module 3 performs preliminary feature extraction on the multimodal feature representation. The feature pyramid network 32 performs multi-scale decomposition and enhancement on the extracted features. The attention mechanism unit 33 performs weighted enhancement on key features to generate a high-resolution and highly discriminative feature map. The feature selection unit 341 in the adaptive feature enhancement system 34 selects key features based on task requirements. The feature amplification unit 342 performs nonlinear amplification on the selected features. The feature correction unit 343 corrects the current features by comparing them with historical feature data, improving the accuracy and robustness of feature extraction.

[0033] Next, the dynamic weight allocation unit 41 in the adaptive positioning optimization module 4 dynamically adjusts the feature weights of different modal data based on current environmental conditions and task requirements. The error compensation unit 42 corrects the current positioning result using historical positioning error data. The path planning unit 43 generates optimal path planning instructions based on the positioning result. The error prediction unit 441 in the dynamic error compensation system 44 predicts the current positioning error by analyzing historical positioning error data. The error correction unit 442 corrects the positioning result based on the prediction result. The error feedback unit 443 feeds the corrected error data back to the error prediction unit 441, forming a closed-loop control system to improve positioning accuracy.

[0034] Finally, the motion controller 51 in the execution control module 5 receives the path planning instructions and generates specific motion control signals. The drive unit 52 drives the robot's motors and actuators based on the motion control signals to complete the positioning and navigation tasks. The obstacle detection unit 531 in the intelligent obstacle avoidance system 53 detects obstacles ahead by analyzing data from the lidar 13 and camera 11. The obstacle avoidance path planning unit 532 generates an obstacle avoidance path based on the obstacle's position and shape. The obstacle avoidance execution unit 533 executes the obstacle avoidance action through the motion controller 51 and drive unit 52, ensuring the robot's safe operation in complex environments.

[0035] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above-described embodiments. The above-described embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A vision-assisted automatic robot positioning method, characterized in that: The following steps are involved: ● Collect environmental data through multiple cameras (11), infrared sensors (12) and laser radar (13) in the visual acquisition module (1), and transmit the collected data to the multimodal data fusion module (2) through the data transmission interface (14); In the multimodal data fusion module (2), the data preprocessing unit (21) is used to perform denoising and formatting of the collected data, the feature mapping unit (22) is used to map the data of different modalities to the same feature space, and the data integration unit (23) is used to generate a unified multimodal feature representation; The multimodal feature representation is transferred to the deep learning feature extraction module (3), and preliminary feature extraction is performed using the convolutional neural network processor (31). The features are decomposed and enhanced at multiple scales using the feature pyramid network (32), and key features are weighted and enhanced using the attention mechanism unit (33) to generate a high-resolution feature map. The feature map is transmitted to the adaptive positioning optimization module (4), the feature weights of different modal data are adjusted by the dynamic weight allocation unit (41), the positioning result is corrected by the error compensation unit (42), and the path planning instruction is generated by the path planning unit (43); ●The path planning instructions are transmitted to the execution control module (5), the motion control signal is generated through the motion controller (51), and the driving unit (52) drives the motor and actuator of the robot to complete the positioning and navigation tasks.

2. The method according to claim 1, characterized in that The multimodal data fusion module (2) further includes an adaptive weight adjustment system (24), wherein the adaptive weight adjustment system (24) includes a weight calculation unit (241), an environment analysis unit (242) and a feedback control unit (243), wherein: An environment analysis unit (242) is used to analyze current lighting conditions, background complexity, and target characteristics to generate environment parameters; A weight calculation unit (241) calculates the weight value of each modal data according to the environmental parameters; The feedback control unit (243) adjusts the weight value in real time according to the positioning error feedback.

3. The method according to claim 1, characterized in that The deep learning feature extraction module (3) further includes an adaptive feature enhancement system (34), wherein the adaptive feature enhancement system (34) includes a feature selection unit (341), a feature amplification unit (342) and a feature correction unit (343), wherein: ● Feature selection unit (341) is used to select key features according to task requirements; A feature amplification unit (342) is used to perform nonlinear amplification on the selected feature; ●The feature correction unit (343) is used to correct the current feature by comparing the historical feature data.

4. The method according to claim 1, wherein The adaptive positioning optimization module (4) further includes a dynamic error compensation system (44), wherein the dynamic error compensation system (44) includes an error prediction unit (441), an error correction unit (442) and an error feedback unit (443), wherein: An error prediction unit (441) is used to predict the current positioning error by analyzing historical positioning error data; ● Error correction unit (442) is used to correct the positioning result according to the prediction result The error feedback unit (443) is used to feed back the corrected error data to the error prediction unit (441).

5. The method according to claim 1, wherein The execution control module (5) further includes an intelligent obstacle avoidance system (53), the intelligent obstacle avoidance system (53) including an obstacle detection unit (531), an obstacle avoidance path planning unit (532) and an obstacle avoidance execution unit (533), wherein: The obstacle detection unit (531) is used to detect obstacles ahead by analyzing the data from the laser radar (13) and the camera (11); The obstacle avoidance path planning unit (532) is used to generate an obstacle avoidance path according to the position and shape of the obstacle; The obstacle avoidance execution unit (533) is used to execute obstacle avoidance actions through the motion controller (51) and the drive unit (52).

6. The method according to claim 2, characterized in that The maximum weight range of the weight calculation unit (241) is 0-1, the minimum adjustment step is 0.01, and the response time is 100 milliseconds.

7. The method according to claim 3, characterized in that The maximum magnification of the feature amplification unit (342) is 5 times, the minimum adjustment step is 0.1, and the correction accuracy of the feature correction unit (343) is ±0.

01.

8. The method according to claim 4, characterized in that The maximum prediction error range of the error prediction unit (441) is ±5 cm, the correction accuracy of the error correction unit (442) is ±1 mm, and the feedback cycle of the error feedback unit (443) is 200 milliseconds.

9. The method according to claim 5, characterized in that The detection range of the obstacle detection unit (531) is 0-10 meters, the maximum path length of the obstacle avoidance path planning unit (532) is 5 meters, and the response time of the obstacle avoidance execution unit (533) is 300 milliseconds.

10. The method according to claim 1, characterized in that The multiple cameras (11), infrared sensors (12) and laser radars (13) in the visual acquisition module (1) are respectively distributed at key positions on the top and around the robot.

Images