Robust control method and system for robot real-time obstacle avoidance in dynamic environment

Abstract

The invention relates to the technical field of robot control, in particular to a robust control method and system for real-time obstacle avoidance of a robot in a dynamic environment. Through innovative algorithm design, for example, under different obstacle avoidance planning strategies, different processing modes are adopted for environment fusion data of a moving obstacle, an obstacle running route is accurately analyzed under a high-precision obstacle avoidance strategy, and an obstacle cluster running range path is analyzed under a hierarchical planning strategy, so that the path planning process is optimized; the unnecessary calculation amount is reduced, so that the consumption of calculation resources is reduced, and the robot can run in real time in a complex dynamic environment; a strong light irradiation area is determined by acquiring light environment data, and countermeasures are taken for the strong light irradiation area, so that the influence of strong light on the sensor is reduced; and meanwhile, the environment sensing analysis unit generates a self-adaptive starting instruction and adjusts the sensor data ratio for data fusion, so that the obstacle detection capability of the robot in a strong light environment is improved, and the obstacle avoidance reliability is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of robot control technology, and in particular to a robust control method and system for real-time obstacle avoidance of a robot in a dynamic environment. Background Art

[0002] Robust control is a key branch of modern control theory. Its core goal is to design a controller that ensures system stability and predictable performance in the face of model uncertainty, external disturbances, or complex environments. The idea is to minimize the system's sensitivity to uncertainty by optimizing the controller design, ensuring the reliability and safety of obstacle avoidance behavior. Robust control enables robots to avoid obstacles in dynamic crowds in service environments such as hospitals and warehouses.

[0003] Existing robust control for real-time obstacle avoidance of robots consumes a lot of computing resources when controlling robots in dynamic environments, such as dense crowds, fast-moving objects, and other complex dynamic environments. When performing path calculations through high-precision algorithms, it is difficult to run in real time. In addition, in strong light environments, the overexposure of the robot camera and the sparse superposition of the lidar point cloud can easily lead to missed obstacle detections. Summary of the Invention

[0004] The present invention provides a robust control method and system for real-time obstacle avoidance of a robot in a dynamic environment, which are used to solve the above technical problems.

[0005] A first aspect of the present invention provides a robust control method for real-time obstacle avoidance of a robot in a dynamic environment, comprising:

[0006] Step 1: The robot's built-in central processing unit controls the multimodal sensor group to collect and execute the raw environmental data, and then executes step 2 on the raw environmental data.

[0007] The environmental raw data includes the raw image data collected by the camera component of the multimodal sensor group and the raw radar data collected by the lidar component of the multimodal sensor group; the raw image data includes environmental obstacle images, overexposed area information and color distortion information; the raw radar data includes environmental obstacle three-dimensional data, effective ranging and detection point cloud data.

[0008] Step 2: The original image data of the environment is transmitted to the environmental perception analysis unit of the central controller and output to obtain an adaptive start instruction, and step 3 is executed on the adaptive start instruction; the graphics processing unit of the central controller performs dynamic environmental prediction analysis on the original image data of the environment to obtain an obstacle avoidance planning instruction, and step 4 is executed on the obstacle avoidance planning instruction.

[0009] As a further improvement of the present invention, the environment perception analysis unit is specifically:

[0010] Based on the over-exposure area information and color distortion information corresponding to the original image of the environment, as well as the effective ranging and false alarm rate corresponding to the original radar image; the environment collection area corresponding to the robot is divided into multiple obstacle avoidance direction blocks;

[0011] Based on the over-exposed area information of each camera component corresponding to each obstacle avoidance direction block, the number of over-exposed pixels is obtained, and the number of over-exposed pixels corresponding to each camera component is summed up by calculation to obtain the total number of over-exposed pixels. The total number of image pixels corresponding to the camera component corresponding to the obstacle avoidance direction block is obtained, and the ratio of the total number of over-exposed pixels to the total number of image pixels is calculated to obtain the over-exposure impact ratio.

[0012] Based on the color distortion information corresponding to each obstacle avoidance direction block, the corresponding distorted image is obtained, the reference image stored in the database is obtained, and the distorted image is compared with the reference image to obtain the CIELAB color space (L * ,a * ,b * ), where L * 、a * 、b * Respectively expressed as the brightness, red and green components, and yellow and blue components of the CIELAB color space, L*∈[0,100], a*∈[-128,127], b*∈[-128,127]; the pixel color difference is calculated by the formula Calculate the pixel color difference value in, They are respectively represented as the brightness, red and green components, and yellow and blue components of the i-th pixel in the reference image; the global color distortion is calculated by summing the pixel color difference values corresponding to each pixel in the obstacle avoidance direction block.

[0013] Obtain the effective distance measurement corresponding to each obstacle avoidance direction block at multiple preset unit time points, and obtain the standard distance measurement corresponding to each obstacle avoidance direction block. Calculate the difference between the effective distance measurement corresponding to each preset unit time point and the standard distance measurement to obtain the distance measurement mutation value. Compile and draw the distance measurement graph corresponding to each unit time point to obtain the distance measurement mutation curve d ratio , and obtain the pre-designed distance measurement mutation threshold d max , draw the ranging mutation threshold on the ranging mutation curve to obtain the mutation threshold segmentation line d max-ratio , the part of the curve that exceeds the ranging mutation threshold and the mutation threshold dividing line are constructed to obtain a closed figure M, the area of the closed figure is calculated and the value of the area is recorded as the ranging anomaly value.

[0014] The total number of detection point clouds is obtained based on the detection point cloud data corresponding to each obstacle avoidance direction block. The signal of each detection power source is identified to obtain the number of false obstacle point clouds detected and recorded as the number of false alarm point clouds. The number of false alarm point clouds is compared with the total number of detection point clouds to obtain the false alarm ratio.

[0015] The overexposure effect ratio, global chromatic aberration distortion, ranging anomaly value and false alarm ratio corresponding to each obstacle avoidance direction block are compared with the preset thresholds. When the overexposure effect ratio is greater than the corresponding preset threshold, an overexposure signal is generated; similarly, a distortion signal corresponding to the global chromatic aberration distortion, a ranging anomaly signal corresponding to the ranging anomaly value and a false alarm signal corresponding to the false alarm ratio are obtained; based on the signal monitoring unit built into the environmental perception analysis unit, when either a distortion signal or an overexposure signal is detected, a corresponding camera anomaly signal is generated; when either a ranging anomaly signal or a false alarm signal is detected, a corresponding radar anomaly signal is generated; when both a camera anomaly signal and a radar anomaly signal are detected, an adaptive start instruction is generated.

[0016] As a further improvement of the present invention, dynamic environment prediction analysis is performed on the original environment image data, and the specific analysis process is as follows:

[0017] The original image data of the environment is identified to obtain original image data and original radar data. The monitoring time is divided into multiple basic monitoring points based on a predetermined monitoring and identification interval. The original image data and original radar data corresponding to each basic monitoring point are obtained. The environmental obstacle image is obtained based on the original image data. Three-dimensional point cloud data is extracted based on the original radar data and a spatial clustering algorithm. The three-dimensional point cloud data is identified to obtain three-dimensional environmental obstacle data. The environmental obstacle image corresponding to each basic monitoring point is matched with the three-dimensional environmental obstacle data to obtain corresponding obstacle information. The obstacle information corresponding to multiple consecutive basic monitoring points is dynamically analyzed to obtain the movement state corresponding to the obstacle information. When the movement state corresponding to the obstacle information is moving, the corresponding obstacle information is marked as a moving obstacle. When the movement state corresponding to the obstacle information is stationary, the corresponding obstacle information is marked as a stationary obstacle.

[0018] The number of moving obstacles is obtained by counting the number of moving obstacles through a counter and recorded as N mad ; Based on the pre-set movement disorder segmentation value, and recorded as N split ;

[0019] When N mad <N split When , the obstacle avoidance planning instruction is generated as a high-precision obstacle avoidance strategy;

[0020] When N mad >N split When , the obstacle avoidance planning instructions are generated as a hierarchical planning strategy.

[0021] Step 3: The central controller performs adaptive fusion analysis on the adaptive start instruction to obtain the sensor data ratio, performs data fusion according to the sensor data ratio to obtain environmental fusion data, and executes step 4 on the environmental fusion data.

[0022] As a further improvement of the present invention, an adaptive fusion analysis is performed on the adaptive startup instruction, and the specific analysis method is as follows:

[0023] The adaptive start command is identified to obtain the abnormal camera signal and the abnormal radar signal, which are respectively recorded as I pix and f r ; when I pix and f r When both are equal to 1, it means that abnormal camera signals and radar signals are detected; when I pix and f r When both are equal to 0, it means that no abnormal camera signal or radar signal is detected;

[0024] When I pix =1, f r =0, indicating that an abnormal camera signal is detected but no abnormal radar signal is detected, and the corresponding sensor data ratio is generated to reduce the proportion of camera data and increase the proportion of radar data;

[0025] When I pix =0, f r =1, indicating that an abnormal radar signal is detected but no abnormal camera signal is detected, and the corresponding sensor data ratio is generated to reduce the proportion of radar data and increase the proportion of camera data;

[0026] When I pix =1, f r =1, obtain the overexposure effect ratio and global chromatic aberration distortion corresponding to the camera abnormal signal, and the ranging abnormal value and false alarm ratio corresponding to the radar abnormal signal; normalize the overexposure effect ratio, global chromatic aberration distortion, ranging abnormal value and false alarm ratio and take their values; perform weighted calculation on the overexposure effect ratio and global chromatic aberration distortion according to a pre-designed weight factor and sum them to obtain the total camera abnormality value corresponding to the camera abnormal signal; similarly, obtain the total radar abnormality value corresponding to the radar abnormal signal; perform difference calculation on the camera impact total value and the radar abnormality total value with the corresponding pre-set abnormal reference value respectively to obtain the camera impact difference value and the radar impact difference; perform ratio calculation on the camera impact difference and the radar impact difference to obtain the difference ratio between the camera data and the radar data; and obtain the corresponding sensor data ratio according to the difference ratio.

[0027] Step 4: The central controller executes the corresponding strategy for the obstacle avoidance planning instructions, performs path planning analysis based on the environmental fusion data, outputs the path planning instructions, and controls the execution according to the path planning instructions.

[0028] As a further improvement of the present invention, path planning analysis is performed based on the environmental fusion data, and the specific analysis method is as follows:

[0029] Obtain and identify obstacle avoidance planning instructions. When the obstacle avoidance planning instructions correspond to a high-precision obstacle avoidance strategy, obtain the environmental fusion data of each moving obstacle, perform precise path analysis on the environmental fusion data of each moving obstacle, and obtain the obstacle running route. Based on the obstacle running route corresponding to each moving obstacle, obtain the robot's obstacle avoidance path;

[0030] When the obstacle avoidance planning instruction corresponds to a hierarchical planning strategy, the obstacle category and running direction are obtained according to the environmental fusion data of each moving obstacle, and the cluster standard range given by the design corresponding to each obstacle category is obtained. The cluster standard ranges corresponding to the moving obstacles with the same obstacle category and running direction are overlapped to obtain the cluster running range path of the corresponding obstacle cluster. The cluster behavior range path is analyzed to obtain the corresponding obstacle avoidance path.

[0031] The second aspect of the present invention provides a robust control system for real-time obstacle avoidance of a robot in a dynamic environment, comprising: a multimodal perception module, an anti-glare perception module, a dynamic risk assessment module and a robust control decision module.

[0032] Multimodal perception module, which collects environmental data through the robot's integrated multimodal sensor group;

[0033] The anti-glare perception module performs anti-glare execution analysis on environmental data to obtain anti-glare execution instructions; specifically, it obtains light environment data based on the collected environmental data, and obtains the strong light irradiation area corresponding to the multimodal sensor group through the light environment data, generates anti-glare execution instructions for the strong light irradiation area, and executes according to the anti-glare execution instructions. The anti-glare execution includes but is not limited to reducing the camera aperture to reduce the amount of light entering, adding optical filters, and using radar to increase the number of point clouds in the strong light area.

[0034] The dynamic risk assessment module obtains the dynamic risk of obstacles and performs risk assessment to obtain the dynamic risk of obstacles, and generates obstacle avoidance decision instructions based on the dynamic risk of obstacles.

[0035] The robust control decision module performs obstacle avoidance analysis on the environmental data corresponding to obstacles with high dynamic risk based on the obstacle avoidance decision instruction to obtain the corresponding obstacle avoidance path, and performs behavior control according to the obstacle avoidance path.

[0036] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:

[0037] 1. This invention uses innovative algorithm design. For example, under different obstacle avoidance planning strategies, different processing methods are adopted for the environmental fusion data of moving obstacles. Under the high-precision obstacle avoidance strategy, the obstacle operation route is accurately analyzed. Under the hierarchical planning strategy, the obstacle cluster operation range path is analyzed. This optimizes the path planning process, reduces unnecessary calculations, and thus reduces the consumption of computing resources, enabling the robot to operate in real time in complex dynamic environments.

[0038] 2. The present invention's anti-glare sensing module determines areas of strong light exposure by acquiring light environment data. It then reduces the impact of strong light on the sensor by taking measures such as narrowing the camera aperture to reduce incoming light, adding optical filters, and increasing the number of radar point clouds in areas of strong light. Simultaneously, the environmental perception and analysis unit, based on multimodal sensor data, comprehensively analyzes information such as overexposed areas, color distortion, effective ranging, and false alarm rates. It then generates adaptive start commands in a timely manner, adjusts sensor data ratios, and performs data fusion. This improves the robot's ability to detect obstacles in strong light environments, reduces missed detections, and ensures reliable obstacle avoidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for the description of the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present application.

[0040] Figure 1 This is a flow chart of a robust control method for real-time obstacle avoidance of a robot in a dynamic environment according to the present invention;

[0041] Figure 2 Schematic diagram of the distance measurement mutation curve of the present invention;

[0042] Figure 3 A specific schematic diagram of the path planning analysis of the present invention;

[0043] Figure 4 This is a principle block diagram of a robust control system for real-time obstacle avoidance of a robot in a dynamic environment according to the present invention. DETAILED DESCRIPTION

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0045] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1-4In an embodiment of the present invention, a robust control method for real-time obstacle avoidance of a robot in a dynamic environment includes the following steps:

[0046] Step 1: Synchronously collect multimodal sensor data. The robot's built-in central processing unit controls the multimodal sensor group to collect and execute the raw environmental data, and then execute step 2 on the raw environmental data.

[0047] The raw environmental data includes the raw image data collected by the camera component of the multimodal sensor group and the raw radar data collected by the lidar component of the multimodal sensor group; the raw image data includes environmental obstacle images, overexposed area information and color distortion information; the raw radar data includes three-dimensional environmental obstacle data, effective ranging and detection point cloud data; it provides comprehensive environmental information for subsequent analysis and solves the problem of insufficient environmental perception of the robot.

[0048] Step 2: Dynamic environment prediction: The original image data of the environment is transmitted to the environmental perception analysis unit of the central controller and output to obtain an adaptive start instruction, and step 3 is executed on the adaptive start instruction; the graphics processing unit of the central controller performs dynamic environment prediction analysis on the original image data of the environment to obtain an obstacle avoidance planning instruction, and step 4 is executed on the obstacle avoidance planning instruction.

[0049] Furthermore, the environmental perception analysis unit is specifically:

[0050] Based on the over-exposure area information and color distortion information corresponding to the original image of the environment, as well as the effective ranging and false alarm rate corresponding to the original radar image; the environment collection area corresponding to the robot is divided into multiple obstacle avoidance direction blocks, and the obstacle avoidance direction blocks include but are not limited to the front obstacle avoidance block, the rear obstacle avoidance block, the left obstacle avoidance block, and the right obstacle avoidance block;

[0051] Based on the over-exposed area information of each camera component corresponding to each obstacle avoidance direction block, the number of over-exposed pixels is obtained, and the number of over-exposed pixels corresponding to each camera component is summed up by calculation to obtain the total number of over-exposed pixels. The total number of image pixels corresponding to the camera component corresponding to the obstacle avoidance direction block is obtained, and the ratio of the total number of over-exposed pixels to the total number of image pixels is calculated to obtain the over-exposure impact ratio.

[0052] Based on the color distortion information corresponding to each obstacle avoidance direction block, the corresponding distorted image is obtained, the reference image stored in the database is obtained, and the distorted image is compared with the reference image to obtain the CIELAB color space (L * ,a * ,b * ), where L * 、a * 、b *Respectively expressed as the brightness, red and green components, and yellow and blue components of the CIELAB color space, L*∈[0,100], a*∈[-128,127], b*∈[-128,127]; the pixel color difference is calculated by the formula Calculate the pixel color difference value in, They are respectively represented as the brightness, red and green components, and yellow and blue components of the i-th pixel in the reference image; the global color distortion is calculated by summing the pixel color difference values corresponding to each pixel in the obstacle avoidance direction block.

[0053] Obtain the effective distance measurement corresponding to each obstacle avoidance direction block at multiple preset unit time points, and obtain the standard distance measurement corresponding to each obstacle avoidance direction block. Calculate the difference between the effective distance measurement corresponding to each preset unit time point and the standard distance measurement to obtain the distance measurement mutation value. Compile and draw the distance measurement graph corresponding to each unit time point to obtain the distance measurement mutation curve d ratio , and obtain the pre-designed distance measurement mutation threshold d max , draw the ranging mutation threshold on the ranging mutation curve to obtain the mutation threshold segmentation line d max-ratio , the part of the curve that exceeds the ranging mutation threshold and the mutation threshold dividing line are constructed to obtain a closed figure M, the area of the closed figure is calculated and the value of the area is recorded as the ranging anomaly value.

[0054] The total number of detection point clouds is obtained based on the detection point cloud data corresponding to each obstacle avoidance direction block. The signal of each detection power source is identified to obtain the number of false obstacle point clouds detected and recorded as the number of false alarm point clouds. The number of false alarm point clouds is compared with the total number of detection point clouds to obtain the false alarm ratio.

[0055] The overexposure effect ratio, global color difference distortion, ranging anomaly and false alarm ratio corresponding to each obstacle avoidance direction block are compared with the preset thresholds. When the overexposure effect ratio is greater than the corresponding preset threshold, an overexposure signal is generated; similarly, a distortion signal corresponding to the global color difference distortion, a ranging anomaly signal corresponding to the ranging anomaly and a false alarm signal corresponding to the false alarm ratio are obtained; based on the signal monitoring unit built into the environmental perception analysis unit, when either a distortion signal or an overexposure signal is detected, a corresponding camera anomaly signal is generated; when either a ranging anomaly signal or a false alarm signal is detected, a corresponding radar anomaly signal is generated; when both a camera anomaly signal and a radar anomaly signal are detected, an adaptive start instruction is generated; based on the overexposure and color distortion information of the original image, the effective ranging and false alarm rate of the original radar, the obstacle avoidance direction blocks are divided, the overexposure effect ratio, the global color distortion, the ranging anomaly and the false alarm ratio are calculated, and the corresponding signals are generated according to the comparison results with the preset thresholds, thereby generating an adaptive start instruction.

[0056] At the warehouse door where strong light is directly shining, the images captured by the camera component appear overexposed. By calculating the ratio of the number of overexposed pixels to the total number of image pixels, the degree of overexposure on the image can be determined. The working status of the sensor can be accurately judged, and environmental interference factors can be discovered in a timely manner. This provides a basis for adjusting the sensor data ratio and solves the problem of inaccurate sensor data in complex lighting and interference environments.

[0057] Furthermore, dynamic environment prediction analysis is performed on the original environment image data, and the specific analysis process is as follows:

[0058] The original image data of the environment is identified to obtain original image data and original radar data, the monitoring time is divided into multiple basic monitoring points based on a predetermined monitoring and identification interval, the original image data and original radar data corresponding to each basic monitoring point are obtained, the environmental obstacle image is obtained based on the original image data, three-dimensional point cloud data is extracted based on the original radar data and a spatial clustering algorithm, and three-dimensional environmental obstacle data is obtained by identifying the three-dimensional point cloud data; the environmental obstacle image corresponding to each basic monitoring point is matched with the three-dimensional environmental obstacle data to obtain corresponding obstacle information, the obstacle information corresponding to multiple consecutive basic monitoring points is dynamically analyzed to obtain the movement state corresponding to the obstacle information, when the movement state corresponding to the obstacle information is moving, the corresponding obstacle information is marked as a moving obstacle; when the movement state corresponding to the obstacle information is stationary, the corresponding obstacle information is marked as a stationary obstacle; moving obstacles include but are not limited to pedestrians, vehicles and equipment; stationary obstacles include terrain obstacles such as buildings, fixed facilities, potholes or bumps, and temporarily placed items such as goods or tools;

[0059] The number of moving obstacles is obtained by counting the number of moving obstacles through a counter and recorded as N mad ; Based on the pre-set movement disorder segmentation value, and recorded as N split ;

[0060] When N mad <N split When the obstacle avoidance planning instruction is generated, it is a high-precision obstacle avoidance strategy; for example, in a narrow corridor, two pedestrians are detected and their environment fusion data position, speed, motion trajectory, etc. are analyzed to generate independent obstacle avoidance paths;

[0061] When N mad >N split When , the obstacle avoidance planning instructions are generated as a hierarchical planning strategy.

[0062] Identify original images and radar data, divide basic monitoring points, obtain environmental data at each point, match obstacle information, analyze and mark their movement status, and generate obstacle avoidance planning instructions based on the comparison results of the number of moving obstacles with the preset segmentation value; for example, in the airport lobby, by continuously monitoring the data of basic monitoring points, identify moving obstacles such as passengers and luggage carts, count the number and compare it with the preset value, and determine the obstacle avoidance strategy; according to the dynamic situation of obstacles in the environment, reasonably select the obstacle avoidance strategy, optimize path planning, reduce computing resource consumption, and enable the robot to operate in real time in complex dynamic environments.

[0063] Step 3: Multi-sensor weight adaptive fusion, the central controller performs adaptive fusion analysis on the adaptive start instruction to obtain the sensor data ratio, performs data fusion according to the sensor data ratio to obtain environmental fusion data, and executes step 4 on the environmental fusion data.

[0064] The adaptive fusion analysis is performed on the adaptive startup instruction. The specific analysis method is as follows:

[0065] The adaptive start command is identified to obtain the abnormal camera signal and the abnormal radar signal, which are respectively recorded as I pix and f r ; when I pix and f r When both are equal to 1, it means that abnormal camera signals and radar signals are detected; when I pix and f r When both are equal to 0, it means that no abnormal camera signal or radar signal is detected;

[0066] When I pix =1, f r =0, indicating that an abnormal camera signal is detected but no abnormal radar signal is detected, and the corresponding sensor data ratio is generated to reduce the proportion of camera data and increase the proportion of radar data;

[0067] When I pix =0, f r =1, indicating that an abnormal radar signal is detected but no abnormal camera signal is detected, and the corresponding sensor data ratio is generated to reduce the proportion of radar data and increase the proportion of camera data;

[0068] When I pix =1, f r=1, obtain the overexposure effect ratio and global chromatic aberration distortion corresponding to the camera abnormal signal, and the ranging abnormal value and false alarm ratio corresponding to the radar abnormal signal; normalize the overexposure effect ratio, global chromatic aberration distortion, ranging abnormal value, and false alarm ratio and take their values; perform weighted calculation and summation on the overexposure effect ratio and global chromatic aberration distortion according to a pre-designed weight factor to obtain the total camera abnormality value corresponding to the camera abnormal signal; perform weighted calculation and summation on the values of the ranging abnormal value and the false alarm ratio according to a pre-designed weight factor to obtain the total radar abnormality value corresponding to the radar abnormal signal; perform difference calculation on the camera impact total value and the radar abnormality total value with the corresponding pre-set abnormal reference value respectively to obtain the camera impact difference value and the radar impact difference; perform ratio calculation on the camera impact difference and the radar impact difference to obtain the difference ratio between the camera data and the radar data; and obtain the corresponding sensor data ratio according to the difference ratio.

[0069] Step 4: Hierarchical path planning and robust control. The central controller executes the corresponding strategy for the obstacle avoidance planning instructions, and performs path planning analysis based on the environmental fusion data to output the path planning instructions and perform control execution according to the path planning instructions.

[0070] Path planning analysis is performed based on environmental fusion data. The specific analysis method is as follows: Obtain obstacle avoidance planning instructions and identify them. When the obstacle avoidance planning instructions correspond to a high-precision obstacle avoidance strategy, obtain the environmental fusion data of each moving obstacle, perform precise path analysis on the environmental fusion data of each moving obstacle, and obtain the obstacle running route. Based on the obstacle running route corresponding to each moving obstacle, the robot's obstacle avoidance path is obtained;

[0071] When the obstacle avoidance planning instruction corresponds to a hierarchical planning strategy, the obstacle category and running direction are obtained according to the environmental fusion data of each moving obstacle, and the cluster standard range given by the design corresponding to each obstacle category is obtained. The cluster standard ranges corresponding to the moving obstacles with the same obstacle category and running direction are overlapped to obtain the cluster running range path of the corresponding obstacle cluster. The cluster behavior range path is analyzed to obtain the corresponding obstacle avoidance path.

[0072] When the obstacle avoidance planning instruction uses a hierarchical planning strategy, the category and direction of movement of the moving obstacle are determined based on the environmental fusion data of the moving obstacle, and the cluster operation range path is determined in combination with the cluster standard range, and then the obstacle avoidance path is planned. For example, in a logistics warehouse, there are a large number of cargo trucks moving in the same direction. The robot determines their cluster operation range based on the vehicle category and operation direction and plans its own obstacle avoidance path. In scenarios with a large number of similar moving obstacles, the path planning process is simplified, the computational complexity is reduced, and the robot's operating efficiency is improved.

[0073] The present invention also provides a robust control system for real-time obstacle avoidance of a robot in a dynamic environment, comprising:

[0074] The multimodal perception module collects environmental data through the multimodal sensor group integrated in the robot; the multimodal sensor group integrated in the robot collects environmental data, provides basic information for the system, realizes comprehensive perception of the environment, and provides data support for subsequent anti-glare analysis, risk assessment and obstacle avoidance decision-making.

[0075] The anti-glare perception module performs anti-glare execution analysis on environmental data to obtain anti-glare execution instructions; specifically, it obtains light environment data based on the collected environmental data, and obtains the strong light irradiation area corresponding to the multimodal sensor group through the light environment data, generates anti-glare execution instructions for the strong light irradiation area, and executes according to the anti-glare execution instructions. The anti-glare execution includes but is not limited to reducing the camera aperture to reduce the amount of light entering, adding optical filters, and increasing the number of point clouds in the strong light area by radar; reducing the impact of strong light on sensors, improving the robot's detection capability in strong light environments, reducing missed detections, and ensuring the reliability of obstacle avoidance.

[0076] The dynamic risk assessment module obtains the dynamic risk of obstacles and performs risk assessment to obtain the dynamic risk of obstacles, and generates obstacle avoidance decision instructions based on the dynamic risk of obstacles. For example, at traffic intersections, it evaluates factors such as the movement speed and direction of vehicles and pedestrians, determines the degree of risk to the robot, and generates corresponding obstacle avoidance decisions. It provides decision-making basis for the robust control decision module, enabling the robot to take appropriate obstacle avoidance measures based on the risk level of obstacles.

[0077] The robust control decision module performs obstacle avoidance analysis on the environmental data corresponding to obstacles with high dynamic risks based on the obstacle avoidance decision instructions to obtain the corresponding obstacle avoidance path, and performs behavior control according to the obstacle avoidance path.

[0078] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A robust control method for real-time obstacle avoidance of a robot in a dynamic environment, characterized by: The steps include: Step 1: The robot's built-in central processing unit controls the multimodal sensor group to collect and execute raw environmental data, and then executes step 2 on the raw environmental data; Step 2: Transmitting the raw image data of the environment to the environment perception analysis unit of the central controller and outputting an adaptive start instruction, and executing step 3 on the adaptive start instruction; performing dynamic environment prediction analysis on the raw image data by the graphics processing unit of the central controller to obtain an obstacle avoidance planning instruction, and executing step 4 on the obstacle avoidance planning instruction; Step 3: The central controller performs adaptive fusion analysis on the adaptive start instruction to obtain sensor data ratio, performs data fusion according to the sensor data ratio to obtain environmental fusion data, and executes step 4 on the environmental fusion data; Step 4: The central controller executes the corresponding strategy for the obstacle avoidance planning instructions, performs path planning analysis based on the environmental fusion data, outputs the path planning instructions, and controls the execution according to the path planning instructions.

2. The robust control method for real-time obstacle avoidance of a robot in a dynamic environment according to claim 1, characterized in that: The environmental raw data includes the raw image data collected by the camera component of the multimodal sensor group and the raw radar data collected by the lidar component of the multimodal sensor group; the raw image data includes environmental obstacle images, overexposed area information and color distortion information; the raw radar data includes environmental obstacle three-dimensional data, effective ranging and detection point cloud data.

3. The robust control method for real-time obstacle avoidance of a robot in a dynamic environment according to claim 1, characterized in that: The specific analysis of the environmental perception analysis unit is as follows: Based on the over-exposure area information and color distortion information corresponding to the original image of the environment, as well as the effective ranging and false alarm rate corresponding to the original radar image; the environment collection area corresponding to the robot is divided into multiple obstacle avoidance direction blocks; Based on the overexposed area information of each camera component corresponding to each obstacle avoidance direction block, the number of overexposed pixels is obtained, and the number of overexposed pixels corresponding to each camera component is calculated and summed to obtain the total number of overexposed pixels. The total number of image pixels corresponding to the camera component corresponding to the obstacle avoidance direction block is obtained, and the ratio of the total number of overexposed pixels to the total number of image pixels is calculated to obtain the overexposure impact ratio; Based on the color distortion information corresponding to each obstacle avoidance direction block, the corresponding distorted image is obtained, the reference image stored in the database is obtained, and the distorted image is compared with the reference image to obtain the CIELAB color space (L * ,a * ,b * ), where L * 、a * 、b * They are respectively expressed as brightness, red and green components, and yellow and blue components of the CIELAB color space; the pixel color difference value is calculated using the pixel color difference calculation formula; the pixel color difference values corresponding to each pixel in the obstacle avoidance direction block are summed to obtain the global color distortion; Obtain the effective distance measurement corresponding to each obstacle avoidance direction block at multiple preset unit time points, and obtain the standard distance measurement corresponding to each obstacle avoidance direction block. Calculate the difference between the effective distance measurement corresponding to each preset unit time point and the standard distance measurement to obtain the distance measurement mutation value. Compile and draw the distance measurement graph corresponding to each unit time point to obtain the distance measurement mutation curve d ratio , and obtain the pre-designed distance measurement mutation threshold d max , draw the ranging mutation threshold on the ranging mutation curve to obtain the mutation threshold dividing line d max-ratio , the part of the curve that exceeds the ranging mutation threshold and the mutation threshold dividing line are constructed to obtain a closed figure M, the area of the closed figure is calculated and the value of the area is recorded as the ranging anomaly value; The total number of detection point clouds is obtained based on the detection point cloud data corresponding to each obstacle avoidance direction block, and the signal of each detection power source is identified to obtain the number of false obstacle point clouds detected and recorded as the number of false alarm point clouds. The number of false alarm point clouds is compared with the total number of detection point clouds to obtain the false alarm ratio; The overexposure effect ratio, global chromatic aberration distortion, ranging anomaly value and false alarm ratio corresponding to each obstacle avoidance direction block are compared with the preset thresholds. When the overexposure effect ratio is greater than the corresponding preset threshold, an overexposure signal is generated; similarly, a distortion signal corresponding to the global chromatic aberration distortion, a ranging anomaly signal corresponding to the ranging anomaly value and a false alarm signal corresponding to the false alarm ratio are obtained; based on the signal monitoring unit built into the environmental perception analysis unit, when either a distortion signal or an overexposure signal is detected, a corresponding camera anomaly signal is generated; when either a ranging anomaly signal or a false alarm signal is detected, a corresponding radar anomaly signal is generated; when both a camera anomaly signal and a radar anomaly signal are detected, an adaptive start instruction is generated.

4. The robust control method for real-time obstacle avoidance of a robot in a dynamic environment according to claim 3, characterized in that: The specific analysis process of performing dynamic environment prediction analysis on the environmental original image data is as follows: Recognize the original environmental image data to obtain original image data and original radar data, divide the monitoring time into multiple basic monitoring points based on the predetermined monitoring and recognition interval, obtain the original image data and original radar data corresponding to each basic monitoring point, obtain the environmental obstacle image based on the original image data, extract three-dimensional point cloud data based on the original radar data and a spatial clustering algorithm, and obtain three-dimensional environmental obstacle data by recognizing the three-dimensional point cloud data; The environmental obstacle image corresponding to each basic monitoring point is matched with the environmental obstacle three-dimensional data to obtain the corresponding obstacle information. The obstacle information corresponding to multiple consecutive basic monitoring points is dynamically analyzed to obtain the movement state corresponding to the obstacle information. When the movement state corresponding to the obstacle information is moving, the corresponding obstacle information is marked as a moving obstacle; when the movement state corresponding to the obstacle information is stationary, the corresponding obstacle information is marked as a stationary obstacle. The number of moving obstacles is obtained by counting the number of moving obstacles through a counter and recorded as N mad ; Based on the pre-set movement disorder segmentation value, and recorded as N split ; When N mad <N split When , the obstacle avoidance planning instruction is generated as a high-precision obstacle avoidance strategy; When N mad >N split When , the obstacle avoidance planning instructions are generated as a hierarchical planning strategy.

5. The robust control method for real-time obstacle avoidance of a robot in a dynamic environment according to claim 1, characterized in that: The adaptive fusion analysis is performed on the adaptive startup instruction, and the specific analysis method is as follows: The adaptive start command is identified to obtain the abnormal camera signal and the abnormal radar signal, which are respectively recorded as I pix and f r ; when I pix and f r When both are equal to 1, it means that abnormal camera signals and radar signals are detected; when I pix and f r When both are equal to 0, it means that no abnormal camera signal or radar signal is detected; When I pix =1, f r =0, indicating that an abnormal camera signal is detected but no abnormal radar signal is detected, and the corresponding sensor data ratio is generated to reduce the proportion of camera data and increase the proportion of radar data; When I pix =0, f r =1, it indicates that an abnormal radar signal is detected but no abnormal camera signal is detected, and the corresponding sensor data ratio is generated to reduce the proportion of radar data and increase the proportion of camera data.

6. The robust control method for real-time obstacle avoidance of a robot in a dynamic environment according to claim 5, characterized in that: When I pix =1, f r =1, obtain the overexposure effect ratio and global chromatic aberration distortion corresponding to the abnormal camera signal, as well as the ranging abnormal value and false alarm ratio corresponding to the abnormal radar signal; normalize the overexposure effect ratio, global chromatic aberration distortion, ranging abnormal value, and false alarm ratio and take their values; perform weighted calculation and summation on the overexposure effect ratio and global chromatic aberration distortion according to a pre-designed weight factor to obtain the total camera abnormality value corresponding to the abnormal camera signal; Similarly, the total radar anomaly value corresponding to the radar anomaly signal is obtained; the total camera impact value and the total radar anomaly value are respectively calculated by difference with the corresponding pre-set anomaly reference value to obtain the camera impact difference and the radar impact difference; the camera impact difference and the radar impact difference are proportionally calculated to obtain the difference ratio between the camera data and the radar data; and the corresponding sensor data ratio is obtained according to the difference ratio.

7. The robust control method for real-time obstacle avoidance of a robot in a dynamic environment according to claim 1, characterized in that: Path planning analysis is performed based on the environmental fusion data, and the specific analysis method is as follows: Obtain and identify obstacle avoidance planning instructions. When the obstacle avoidance planning instructions correspond to a high-precision obstacle avoidance strategy, obtain the environmental fusion data of each moving obstacle, perform precise path analysis on the environmental fusion data of each moving obstacle, and obtain the obstacle running route. Based on the obstacle running route corresponding to each moving obstacle, obtain the robot's obstacle avoidance path; When the obstacle avoidance planning instruction corresponds to a hierarchical planning strategy, the obstacle category and running direction are obtained according to the environmental fusion data of each moving obstacle, and the cluster standard range given by the design corresponding to each obstacle category is obtained. The cluster standard ranges corresponding to the moving obstacles with the same obstacle category and running direction are overlapped to obtain the cluster running range path of the corresponding obstacle cluster. The cluster behavior range path is analyzed to obtain the corresponding obstacle avoidance path.

8. A robust control system for real-time obstacle avoidance of a robot in a dynamic environment, characterized by A robust control method for implementing real-time obstacle avoidance of a robot in a dynamic environment as described in any one of claims 1 to 7, comprising a multimodal perception module, an anti-glare perception module, a dynamic risk assessment module, and a robust control decision module; The multimodal perception module collects environmental data through a multimodal sensor group integrated with the robot; The anti-glare perception module performs anti-glare execution analysis based on environmental data to obtain an anti-glare execution instruction; The dynamic risk assessment module obtains the dynamic risk of the obstacle and performs risk assessment to obtain the dynamic risk of the obstacle, and generates an obstacle avoidance decision instruction according to the dynamic risk of the obstacle; The robust control decision module performs obstacle avoidance analysis on environmental data corresponding to obstacles with high dynamic risk based on the obstacle avoidance decision instruction to obtain a corresponding obstacle avoidance path, and performs behavior control according to the obstacle avoidance path.

9. The robust control system for real-time obstacle avoidance of a robot in a dynamic environment according to claim 8, characterized in that: An anti-glare execution analysis is performed on the environmental data to obtain an anti-glare execution instruction, which is specifically: light environment data is obtained based on the collected environmental data, and the strong light illumination area corresponding to the multimodal sensor group is obtained through the light environment data, an anti-glare execution instruction is generated for the strong light illumination area, and execution is performed according to the anti-glare instruction.

Images