A fully automatic robot testing method based on visual analysis

Abstract

The application relates to the technical field of public health, in particular to a full-automatic robot testing method based on visual analysis, which can save the repeated mapping of each robot, is efficient, can automatically plan a path, is diversified, can comprehensively test, can obtain image recognition test results, is more accurate and objective, can automatically test in batches, can save manpower, can automatically archive and export test results, and is convenient to trace.

Description

Technical Field

[0001] This invention relates to the field of elevator technology, specifically to a fully automated robot testing method based on visual analysis. Background Technology

[0002] With the development of technology, robotics has gradually become a new and popular research field. People have been studying how to enable robots to replace human labor in various tasks. A robot itself does not have an intelligent unit, only an actuator and a sensing mechanism. It has the ability to process sensory information (including vision, hearing, touch, proximity, force, infrared, ultrasound, and laser) to achieve control and operation. Generally, robots are also controlled by an external computer. This external computer has an intelligent processing unit that processes various information collected by the controlled robot, as well as information about the robot's posture and trajectory, and then issues control commands to direct the robot's actions.

[0003] In the operating environment of robots, there are often obstacles, such as fixed supports and obstacles that temporarily stored robots cannot easily cross. When a robot moves from one location to another, it must avoid collisions with these obstacles and therefore needs to detour. How to enable a robot to automatically avoid obstacles and select a suitable route to continue moving is a very complex problem. Existing technologies have proposed many solutions. However, some of these solutions (artificial potential field method, VFH histogram method, etc.) are quite precise, but they have the drawbacks of being complex, difficult to implement and popularize, and making the overall cost of the robot very high. Other algorithms are simple and easy to implement, but their obstacle avoidance accuracy is not high. Moreover, different robots have different sizes, weights, travel speeds, and computing capabilities, so existing algorithms cannot achieve 100% obstacle avoidance. In summary, existing robot obstacle avoidance algorithms are difficult to meet users' needs for low cost and high accuracy, which has prevented the widespread adoption of robots. Summary of the Invention

[0004] The purpose of this invention is to provide a fully automated robot testing method based on visual analysis, in order to solve the problems mentioned in the background art, such as the need for separate mapping for batch robot testing; the need for manual path planning during the testing process; the lack of objectivity due to human observation of test results; the need for human assistance during the testing process; and the low efficiency.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a fully automated robot testing method based on visual analysis, the method comprising:

[0006] S1. Build a real-world testing environment based on the robot simulation testing environment;

[0007] S2. Generate an optimal route and a high-precision map for robot navigation and obstacle avoidance in the simulation environment;

[0008] S3. During robot testing, the robot's navigation trajectory and posture are recorded in real time using a vision-based analysis method.

[0009] S4. Add UWB module ranging to improve positioning accuracy;

[0010] S5. Test result output.

[0011] Furthermore, S1 includes:

[0012] S1-1. Develop a robot simulation test environment using robot simulation software, which includes obstacles and triggering conditions for triggering various robot sensors. These triggering conditions are consistent with the robot's navigation and obstacle avoidance triggering conditions.

[0013] S1-2. Based on the simulation environment, build a real test environment. The test environment should be a 1:1 replica of the simulation environment.

[0014] Furthermore, S1-1 includes:

[0015] S1-1-1, The obstacles and triggering conditions used for ultrasonic sensor triggering are: obstacles are less than 50cm to avoid being scanned by lidar; the triggering condition is that the obstacle avoidance threshold will be triggered when the distance is less than 15cm.

[0016] S1-1-2. The obstacle and triggering conditions used for the collision sensor are as follows: the obstacle is 4cm high and 1cm wide, avoiding detection by the ultrasonic sensor; the triggering condition is that after the collision contact, the anti-collision strip is pressed and outputs a high-level signal.

[0017] S1-1-3. The conditions for triggering the drop sensor are: there is a depression or step in front of it that is more than 3cm deep, and the drop sensor outputs a high-level signal after collecting data.

[0018] S1-1-4. The obstacles and triggering conditions used for laser fence sensor triggering are as follows: the obstacle is higher than 30cm and suspended below 30cm to avoid detection by ultrasonic and lidar; the triggering condition is: if the upward detection distance of the laser fence is less than 110cm, it is considered that there is a suspended object lower than the robot blocking the robot, and the laser fence sensor outputs a high-level signal.

[0019] Furthermore, S2 includes:

[0020] S2-1. In a simulation environment, the globally optimal path for robot navigation and obstacle avoidance is generated using Dijkstra's algorithm.

[0021] S2-2. In a simulation environment, the local optimal path is obtained using the TEB algorithm;

[0022] S2-3. In the simulation environment, the simulated 3D spatial structure is a high-precision 3D map. A high-precision 2D map of the robot is generated by using the 3D spatial section method: Based on the height of the robot, a top view is constructed from the 3D map portion within the height range that the robot can touch, resulting in a high-precision 2D map.

[0023] Furthermore, in S2-3, after the robot builds a map in a real-world testing environment, the map is compared with a simulated high-precision map to verify the accuracy of the map building.

[0024] Furthermore, S3 includes:

[0025] S3-1. High-definition cameras are placed above the test area to record the robot's path planning and obstacle avoidance.

[0026] S3-2. Record the robot's trajectory and posture using visual analysis methods to generate the robot's actual walking route.

[0027] Furthermore, S3-1 includes:

[0028] S3-1-1. The robot's walking path is analyzed through visual analysis. The method is as follows: First, the robot's position is located through the SSD or YOLO visual detection model. Then, the robot is tracked using vision-based object tracking methods such as KCF to obtain the robot's walking path.

[0029] S3-1-2, Analyzing the robot's real-time posture: By attaching a QR code to the top of the robot, visual analysis is used to analyze the robot's posture;

[0030] Furthermore, in S3-2, the robot's real-time walking route is compared with the simulated route in the simulation environment to determine whether the robot's navigation and obstacle avoidance are qualified. The judgment condition is that the robot does not collide with obstacles and the difference between the route planning result and the path planning result in the simulation environment is within a certain range.

[0031] Furthermore, in S4, the UWB module is a centimeter-level positioning UWB module. Its module distribution is such that there are three fixed UWB modules in the test environment, and one UWB at the top center of the robot. The robot's real-time positioning is achieved by calculating the distance from the robot to each UWB module.

[0032] Furthermore, during the testing process, the robot automatically completes the testing steps and can perform repeated tests.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] This invention utilizes multiple sensors to detect obstacles accurately, then calculates the specific parameters of the obstacles based on the collected sensor data. It then identifies routes where the robot cannot continue due to the obstacle's presence, and replans the corresponding sections of these routes. The robot's movement is then controlled based on the replanned route, improving the accuracy of obstacle detection and obstacle avoidance. Furthermore, this embodiment can acquire obstacles in the current path using a 3D map, enabling the robot to successfully avoid obstacles during movement. This not only ensures the robot reaches the target location smoothly and moves stably, but also eliminates the need for repetitive mapping for each robot, increasing efficiency. It features fully automated path planning, diverse and comprehensive testing, image recognition for higher accuracy and objectivity, fully automated batch testing for high efficiency and labor savings, and automatic archiving and export of test results for easy traceability. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the Dijkstra algorithm of the present invention;

[0036] Figure 2 This is a flowchart illustrating the TEB algorithm of the present invention;

[0037] Figure 3 This is a schematic diagram of the positioning principle of the UWB module of the present invention;

[0038] Figure 4 This is a flowchart illustrating the fully automated robot testing method based on visual analysis according to the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example 1

[0041] like Figure 4 As shown, a fully automated robot testing method based on visual analysis is proposed, which includes:

[0042] S1. Build a real-world testing environment based on the robot simulation testing environment.

[0043] Furthermore, S1 includes:

[0044] S1-1. Develop a robot simulation test environment using robot simulation software, which includes obstacles and triggering conditions for triggering various robot sensors. These triggering conditions are consistent with the robot's navigation and obstacle avoidance triggering conditions.

[0045] Specifically, S1-1 includes:

[0046] S1-1-1, The obstacles and triggering conditions used for ultrasonic sensor triggering are: obstacles are less than 50cm to avoid being scanned by lidar; the triggering condition is that the distance is less than 15cm to trigger the obstacle avoidance threshold.

[0047] S1-1-2. The obstacle and triggering conditions used for the collision sensor are as follows: the obstacle is 4cm high and 1cm wide, avoiding detection by the ultrasonic sensor; the triggering condition is that after the collision contact, the anti-collision strip is pressed and outputs a high-level signal.

[0048] S1-1-3 The conditions for triggering the drop sensor are: there is a depression or step in front of it that is more than 3cm deep, and the drop sensor outputs a high-level signal after collecting data.

[0049] S1-1-4. The obstacles and triggering conditions used for laser fence sensor triggering are as follows: the obstacle is higher than 30cm and suspended below 30cm to avoid detection by ultrasonic and lidar; the triggering condition is: if the upward detection distance of the laser fence is less than 110cm, it is considered that there is a suspended object lower than the robot blocking the robot, and the laser fence sensor outputs a high-level signal.

[0050] S1-2. Based on the simulation environment, build a real test environment. The test environment should be a 1:1 copy of the simulation environment.

[0051] S2. Generate an optimal route and a high-precision map for robot navigation and obstacle avoidance in the simulation environment.

[0052] Furthermore, S2 includes:

[0053] S2-1. In a simulation environment, the globally optimal path for robot navigation and obstacle avoidance is generated using Dijkstra's algorithm.

[0054] Specifically, such as Figure 1 As shown, the basic process of Dijkstra's algorithm is as follows: Dijkstra's algorithm first initializes the distance from the source node to all nodes as infinity, except for the source node itself, which is set to 0. Then it initializes a node priority queue, where the priority of each node is its distance to the source.

[0055] The algorithm then enters a loop until the priority queue is empty. In each iteration, it extracts the node with the smallest distance to the source from the priority queue and checks its neighbors. For each neighbor, it calculates the distance from the source to that neighbor using the current node; if the distance is less than the current distance, it updates the distance. It also updates the priorities of the neighbors in the priority queue. Finally, the algorithm returns the distances from the source to all nodes as the result.

[0056] S2-2. In the simulation environment, the local optimal path is obtained through the TEB (Time Elastic-Band) algorithm.

[0057] Specifically, such as Figure 2 As shown, the input parameters of the TEB algorithm include the robot's current state, target state, list of obstacles, and a set of parameters controlling the algorithm's behavior (e.g., maximum speed, maximum acceleration, etc.). The algorithm begins by initializing the current and target states as elastic bands. It then enters a loop that iteratively optimizes the shape of the elastic bands and checks for collisions with obstacles. If a collision is detected, the algorithm replans the trajectory using the new elastic band. The loop continues until the elastic bands converge, at which point the optimized elastic band is returned as the planned trajectory.

[0058] S2-3. In the simulation environment, the simulated 3D spatial structure is a high-precision 3D map. A high-precision 2D map of the robot is generated by using the 3D spatial section method: Based on the height of the robot, a top view is constructed from the 3D map portion within the height range that the robot can touch, resulting in a high-precision 2D map.

[0059] Furthermore, in S2-3, after the robot builds a map in a real-world testing environment, the map is compared with a simulated high-precision map to verify the accuracy of the map building.

[0060] Specifically, commonly used 2D map quality assessment methods include normalized nearest neighbor distance, mean square error, structural similarity, and sum of side lengths. Commonly used 3D map quality assessment methods include absolute trajectory error, relative trajectory error, and root mean square error. The process for the map quality assessment method based on mean square error is as follows:

[0061]

[0062] Where M is the total number of pixels in image x; N is the total number of pixels in image y; and x and y are the generated map and the ground reality map, respectively. The smaller the MSE, the smaller the difference between the generated map and the ground reality map.

[0063] S3. During robot testing, the robot's navigation trajectory and posture are recorded in real time using a vision-based analysis method.

[0064] Furthermore, S3 includes:

[0065] S3-1. High-definition cameras are placed above the test site to record the robot's path planning and obstacle avoidance.

[0066] Furthermore, S3-1 includes:

[0067] S3-1-1. The robot's walking path is analyzed through visual analysis. The method is as follows: First, the robot's position is located through the SSD or YOLO visual detection model. Then, the robot is tracked using vision-based object tracking methods such as KCF to obtain the robot's walking path.

[0068] S3-1-2. Analyze the robot's real-time posture: By attaching a QR code to the top of the robot, visual analysis is used to analyze the robot's posture.

[0069] S3-2. Record the robot's trajectory and posture using visual analysis methods to generate the robot's actual walking route.

[0070] Furthermore, in S3-2, the robot's real-time walking route is compared with the simulated route in the simulation environment to determine whether the robot's navigation and obstacle avoidance are qualified. The judgment condition is that the robot does not collide with obstacles and the difference between the route planning result and the path planning result in the simulation environment is within a certain range.

[0071] S4. Adding a UWB module for ranging improves positioning accuracy.

[0072] Furthermore, in S4, the UWB module is a centimeter-level positioning UWB module. Its module distribution is such that there are three fixed UWB modules in the test environment, and one UWB at the top center of the robot. The robot's real-time positioning is achieved by calculating the distance from the robot to each UWB module.

[0073] Specifically, three fixed UWB modules were deployed on-site, and the coordinates of each UWB module were recorded.

[0074] The robot reads its distance from three UWBs in real time during operation and calculates its real-time coordinates to achieve assisted positioning.

[0075] like Figure 3 As shown, the coordinates of point P are obtained using the trilateration method:

[0076]

[0077] Among them, (X) A ,Y A (X) represents the coordinates of point A; B ,Y B (X) represents the coordinates of point B;C ,Y C (X) represents the coordinates of point C; P ,Y P Let P be the coordinates of point P; L1 be the distance from point P to point A; L2 be the distance from point P to point B; and L3 be the distance from point P to point C.

[0078] S5. Test result output.

[0079] Furthermore, during the testing process, the robot automatically completes the testing steps and can perform repeated tests.

[0080] This fully automated robot testing method based on vision analysis uses multiple sensors to detect accurate sensor data of obstacles. Then, based on the collected sensor data, it calculates the specific parameters of the obstacles and determines which routes cannot continue due to the presence of obstacles. For the sections of the routes that cannot continue, a new route is planned, and the robot is controlled again based on the replanned route, thus improving the accuracy of obstacle detection and obstacle avoidance. Furthermore, this embodiment can obtain obstacles in the current path based on a 3D map, enabling the robot to successfully avoid obstacles during movement. This not only ensures the robot reaches the target location smoothly and moves stably, but also eliminates the need for repetitive mapping for each robot, increasing efficiency. Fully automated path planning allows for diverse and comprehensive testing. Test results are obtained through image recognition, resulting in higher accuracy and objectivity. Fully automated batch testing saves manpower. Test results are automatically archived and exported for easy traceability.

[0081] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A fully automated robot testing method based on visual analysis, characterized in that, The method comprises: S1, according to the robot simulation test environment, a real environment test is built; S1 includes: S1-1, develop a robot simulation test environment through a robot simulation software, including obstacles for triggering various sensors of the robot and triggering conditions, the triggering conditions are consistent with the robot navigation obstacle avoidance triggering conditions; S1-1 includes: S1-1-1, the obstacle and triggering condition for ultrasonic sensor triggering are: the obstacle is lower than 50cm, avoiding laser radar scanning; the triggering condition is that the distance is lower than 15cm, which will trigger the obstacle avoidance threshold; S1-1-2, the obstacle and triggering condition for collision sensor triggering are: the obstacle is 4cm high and 1cm wide, avoiding ultrasonic sensor detection; the triggering condition is that the collision contact presses the anti-collision strip to output a high level signal; S1-1-3, the condition for drop sensor triggering is: there is a depression or step of more than 3cm in front, and the drop sensor outputs a high level signal after collecting data; S1-1-4, the obstacle and triggering condition for laser fence sensor triggering are: the obstacle is higher than 30cm and suspended below 30cm to avoid ultrasonic and laser radar detection; the triggering condition is: the upward detection distance of the laser fence is lower than 110cm, then it is considered that there is a suspended object below the robot blocking the robot, and the laser fence sensor outputs a high level signal; S1-2, according to the simulation environment, a real test environment is built, and the test environment is required to be 1:1 copied with the simulation environment; S2, an optimal route and a high-precision map of robot navigation obstacle avoidance are generated in the simulation environment; S3, when the robot is tested, the navigation trajectory and attitude of the robot are recorded in real time through a method based on visual analysis; S3 includes: S3-1, the situation when the robot avoids obstacles through path planning is recorded by arranging a high-definition camera above the test site; S3-2, the robot running track and attitude are recorded by a visual analysis method, and the actual walking route of the robot is generated; S3-1 includes: S3-1-1, the walking route of the robot is analyzed by visual analysis, which is: first, the position of the robot is located by an SSD or YOLO visual detection model, and then the KCF visual-based object tracking method is used to track the robot, so as to obtain the walking route of the robot; S3-1-2, the real-time attitude of the robot is analyzed by pasting a two-dimensional code on the top of the robot, and the attitude of the robot is analyzed by visual analysis; S4, the positioning accuracy is improved by adding a UWB module for distance measurement; S5, the test result is output.

2. A fully automated robotic testing method based on visual analysis as claimed in claim 1, wherein: S2 includes: S2-1, in the simulation environment, a global optimal path of robot navigation obstacle avoidance is generated by Dijkstra algorithm; S2-2, in the simulation environment, a local optimal path is obtained by TEB algorithm; S2-3, in the simulation environment, the simulated 3D space structure is a high-precision 3D map, and a high-precision 2D map of the robot is generated by 3D space section method: according to the height of the robot, the 3D map part in the touchable height range of the robot is constructed into an overhead view to obtain a high-precision 2D map.

3. A fully automated robotic testing method based on visual analysis as claimed in claim 2, wherein: In S2-3, the robot compares the map with the simulation high-precision map after mapping in the real test environment to verify the mapping accuracy.

4. The fully automated robotic testing method based on visual analysis of claim 1, wherein: In S3-2, the robot real-time walking route is compared with the simulation route in the simulation environment to determine whether the current robot navigation and obstacle avoidance are qualified, and the judgment condition is that no collision with obstacles occurs and the difference between the route result of path planning and the path planning result in the simulation environment is within a certain range.

5. A fully automated robotic testing method based on visual analysis as claimed in claim 1, wherein: In S4, the UWB module is a centimeter-level positioning UWB module, and the module distribution is that three UWB modules are fixed in the test environment, and one UWB is arranged at the center of the top of the robot. The real-time positioning of the robot is realized by calculating the distance from the robot to each UWB module.

6. A fully automated robotic testing method based on visual analysis as claimed in claim 1, wherein: During the test, the robot automatically completes the test link and can be tested in a loop.

Images