Robotic multi-sensor integrated testing system and method

Abstract

The application provides a robot multi-sensor integrated test system and method. It comprises: a test box body for forming a test space with a fixed robot installation reference surface, and setting at least one group of ultrasonic acoustic reflection surface, laser radar geometric reflection surface and visual high flatness reference surface in the test space to provide standardized reference targets for the ultrasonic sensor, laser radar sensor and visual sensor of the robot; a multi-sensor test module arranged in the test box body and electrically connected with the robot; a central control module in bidirectional communication with the multi-sensor test module; and a result output module in signal connection with a data analysis unit for generating a robot multi-sensor performance evaluation report according to the performance evaluation results of each sensor generated by the data analysis unit. The application can achieve the technical effects of one-stop synchronous test, collaborative performance quantification evaluation and unified standardized result output of multi-sensors.

Description

Technical Field

[0001] This application relates to the field of robotics technology, and in particular to a multi-sensor integrated testing system and method for robots. Background Technology

[0002] Robots are automated equipment integrating cybernetics, mechatronics, computer technology, materials engineering, and bionics, and are widely used in diverse scenarios such as industrial production, medical care, modern agriculture, and national defense. To perform precision tasks in dynamic and complex environments, modern robots are typically equipped with multiple sensors, including lidar, ultrasonic sensors, top / bottom-view cameras, inertial measurement units (IMUs), barometers, and motor encoders. These sensors, through time synchronization and data fusion, provide high-precision environmental and self-state information to the motion control, path planning, map building, and task execution modules, forming the core foundation for ensuring the safe and stable operation of the robot.

[0003] To verify the accuracy and stability of sensor outputs, the industry commonly employs a distributed single-sensor testing method. LiDAR requires a dedicated turntable with a standard target plate to measure scanning accuracy; cameras rely on optical targets and a darkroom for endo- and extrinsic parameter calibration; IMUs are placed on a high-precision multi-axis turntable to evaluate zero bias and random drift; and motor encoders use laser interferometers or calibration racks to verify position and velocity output. Some manufacturers also attempt to collect robot operation data in the field and then statistically analyze error indicators offline, or to perform local simulations in the laboratory using a vibration table with superimposed finite working conditions. However, these testing processes are often based on a single sensor or local component, lacking a comprehensive testing platform that simultaneously examines the collaborative performance of multiple sensors at the overall system level. Furthermore, the industry has not yet established consistent performance evaluation dimensions and quality judgment thresholds.

[0004] Therefore, existing technologies still have the following problems: First, the testing process is fragmented and inefficient. Multiple independent sets of equipment and multi-station assembly / disassembly result in long testing cycles for a single robot, making it difficult to meet the pace of mass production. Second, single-sensor static testing cannot reproduce the complex working conditions faced by robots in actual operation, such as body vibration, temperature gradients, light changes, and electromagnetic interference. System-level collaborative errors and fusion mismatch risks are difficult to expose in a timely manner. Third, different sensor manufacturers and laboratories use their own custom accuracy, drift, and stability evaluation methods, lacking a unified quantitative benchmark, making it difficult to standardize quality control and compare and trace test results horizontally. Fourth, fault location relies on post-incident investigation. When positioning deviations or path planning errors occur online, it is impossible to quickly determine whether it is due to single sensor drift, hardware / software integration mismatch, or distortion of multi-source fusion algorithm parameters. Summary of the Invention

[0005] In view of this, embodiments of this application provide a robot multi-sensor integrated testing system and method to solve the problems of fragmented testing processes, low efficiency, difficulty in evaluating collaborative performance, and lack of unified performance evaluation standards in the prior art.

[0006] The first aspect of this application provides a robot multi-sensor integrated testing system, comprising: a test chamber for forming a test space with a fixed mounting reference surface for the robot under test, and at least one set of ultrasonic acoustic reflective surfaces, lidar geometric reflective surfaces, and visual high-level flatness reference surfaces arranged in the test space according to a preset spatial distribution, so as to provide standardized reference targets for the ultrasonic sensors, lidar sensors, and vision sensors of the robot under test respectively; a multi-sensor testing module, disposed in the test chamber and electrically connected to the robot under test, the multi-sensor testing module including an ultrasonic testing unit, a lidar testing unit, and a vision testing unit; a central control module for bidirectional communication with the multi-sensor testing module, the central control module including a test control unit, a data acquisition unit, and a data analysis unit; and a result output module, signal-connected to the data analysis unit, for generating a robot multi-sensor performance evaluation report based on the performance evaluation results corresponding to each sensor generated by the data analysis unit, so as to determine the consistency of sensors based on the robot multi-sensor performance evaluation report.

[0007] The second aspect of this application provides a multi-sensor integrated testing method based on a robot multi-sensor integrated testing system, comprising: positioning the robot under test on the mounting reference plane of the test box and completing the electrical connection with the multi-sensor testing module; using the test control unit in the central control module to synchronously issue start commands to the ultrasonic testing unit, the lidar testing unit, and the vision testing unit and establish a unified time reference; using the data acquisition unit in the central control module to acquire the raw data streams output by each testing unit in parallel under the unified time reference; performing processing on the acquired raw data streams associated with the corresponding reference plane or preset parameters, and using the data analysis unit to perform error calculation, stability statistics, and fit evaluation based on the processing results to obtain the performance evaluation results corresponding to each sensor; inputting the performance evaluation results to the result output module for summarization to generate a robot multi-sensor performance evaluation report containing the performance evaluation results of each sensor.

[0008] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects:

[0009] The test chamber forms a test space with a fixed mounting reference surface for the robot under test. At least one set of ultrasonic acoustic reflective surfaces, lidar geometric reflective surfaces, and visual high-level flatness reference surfaces are arranged in a preset spatial distribution within the test space to provide standardized reference targets for the ultrasonic sensors, lidar sensors, and vision sensors of the robot under test. A multi-sensor test module, housed within the test chamber and electrically connected to the robot under test, includes an ultrasonic test unit, a lidar test unit, and a vision test unit. A central control module, communicating bidirectionally with the multi-sensor test module, includes a test control unit, a data acquisition unit, and a data analysis unit. A result output module, signal-connected to the data analysis unit, generates a multi-sensor performance evaluation report for the robot based on the performance evaluation results generated by the data analysis unit for each sensor, facilitating sensor consistency determination based on the multi-sensor performance evaluation report. This application achieves the technical effects of one-stop synchronous testing of multiple sensors, collaborative performance quantitative evaluation, and unified standardized result output. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of the structural composition of the robot multi-sensor integrated testing system provided in the embodiments of this application;

[0012] Figure 2 This is a flowchart illustrating the multi-sensor integrated testing method based on a robot multi-sensor integrated testing system provided in this application embodiment. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] Currently, most robot sensor calibration adopts a "single-component separate testing" approach: lidar relies on independent turntables and target plates to measure scanning accuracy, ultrasonic sensors are compared at fixed distances on separate distance measuring fixtures, cameras undergo internal and external parameter calibration in an optical darkroom, and IMUs are placed on high-precision turntables to test zero bias and random drift. This dispersed deployment of multiple sets of equipment and multi-station assembly and disassembly not only lengthens the testing cycle but also makes it difficult to reproduce the real-world operating conditions of multiple sensors coupling together during robot operation. Furthermore, the evaluation indicators vary between laboratories, resulting in a lack of comparability and traceability of test results.

[0015] Distributed testing leads to the following problems: 1. Low testing efficiency, making it difficult to support mass production; 2. Inability to evaluate the collaborative performance of multiple sensors in the laboratory; 3. Lack of unified quantitative standards, making quality control difficult; 4. Long fault location cycle and high cost. Therefore, there is an urgent need for a comprehensive testing solution that can simultaneously detect multiple types of sensors in a single device and output a unified performance report.

[0016] This application proposes a robot multi-sensor integrated testing system, including:

[0017] Test chamber: Based on the fixed robot mounting reference surface, it integrates ultrasonic acoustic reflective surface, lidar geometric reflective surface and visual high flatness reference surface according to the preset layout, providing standardized reference targets for the three types of sensors;

[0018] Multi-sensor testing module: Ultrasonic, lidar, vision and optional IMU testing units are set up in the chamber to collect raw data and compare it with the corresponding reference surface or preset parameters;

[0019] Central control module: includes test control, data acquisition and data analysis units, issues start commands with a unified clock and pulls data from each sensor in parallel, performs error calculation, stability statistics and fit evaluation;

[0020] Results output module: Summarizes the performance evaluation results of each sensor and generates a structured test report with timestamps and device identifiers for use by the manufacturing execution system or maintenance platform.

[0021] This application utilizes an integrated enclosure and a unified control platform to achieve one-stop synchronous testing of multiple sensors, enabling quantitative evaluation of collaborative performance within the laboratory. The testing process is reduced from multi-station disassembly and assembly to single-station one-click execution, significantly improving efficiency and reducing equipment and labor costs. At the same time, it outputs unified and standardized results, facilitating batch quality control, fault tracing, and in-service monitoring.

[0022] The specific structure and function of the robot multi-sensor integrated testing system provided in this application will be described in detail below with reference to the accompanying drawings and specific embodiments. Figure 1This is a schematic diagram of the structural composition of the robot multi-sensor integrated testing system provided in the embodiments of this application, as shown below. Figure 1 As shown, the robot multi-sensor integrated testing system may specifically include the following modules:

[0023] The test chamber 101 is used to form a test space with a fixed mounting reference surface for the robot under test, and at least one set of ultrasonic acoustic reflective surfaces, lidar geometric reflective surfaces and visual high-flatness reference surfaces are set in the test space according to a preset spatial distribution, so as to provide standardized reference targets for the ultrasonic sensor, lidar sensor and visual sensor of the robot under test respectively.

[0024] The multi-sensor test module 102 is set inside the test chamber and electrically connected to the robot under test. The multi-sensor test module includes an ultrasonic test unit, a lidar test unit and a vision test unit.

[0025] The central control module 103 communicates bidirectionally with the multi-sensor test module. The central control module includes a test control unit, a data acquisition unit, and a data analysis unit.

[0026] The result output module 104 is connected to the data analysis unit and is used to generate a robot multi-sensor performance evaluation report based on the performance evaluation results of each sensor generated by the data analysis unit, so as to determine the consistency of sensors based on the robot multi-sensor performance evaluation report.

[0027] In some embodiments, the test chamber has an acoustic reflective surface on each of the left and right sides of the mounting reference surface of the robot under test, and each acoustic reflective surface maintains a preset fixed distance from the mounting reference surface of the robot under test, and forms a non-reflective area in front of the mounting reference surface of the robot under test, so as to provide the ultrasonic sensor with left and right fixed distance reference targets and unobstructed reference targets in front.

[0028] Specifically, this embodiment, based on the aforementioned robot multi-sensor integrated testing system, provides a detailed description of the benchmark scene construction and measurement process for ultrasonic sensors, focusing on the specific structure, installation method, and testing steps of the acoustic reflecting surface and non-reflective area within the test chamber.

[0029] I. Test chamber and mounting reference surface

[0030] The test chamber uses a one-piece welded aluminum frame with an antistatic powder coating on the inner wall, measuring 2000mm × 1600mm × 1500mm. A 10mm thick granite mounting reference surface is provided at the bottom of the chamber. This reference surface is precision ground to ensure a flatness error within 0.02mm / 1000mm, guaranteeing consistent posture of the robot under test during repeated placement. Cross-shaped dovetail grooves are arranged on the reference surface, which, in conjunction with a quick-positioning fixture, enable rapid locking of robots with different wheelbases or track spacings.

[0031] II. Acoustic Reflector Structure and Positioning

[0032] On each of the left and right sides of the mounting reference surface, a 600mm×800mm×8mm aluminum alloy acoustic reflector is fixed. The reflectors are treated with T6 hard anodizing, with a surface roughness Ra≤0.8µm, effectively ensuring the ultrasonic mirror reflection characteristics. The two reflectors are connected to the side wall of the housing via adjustable slide rails, with an adjustment range of ±30mm to accommodate different vehicle widths. At the factory, the horizontal distance from the inner side of the two reflectors to the center line of the mounting reference surface is calibrated to 25cm, with a corresponding error tolerance of ±2cm. After calibration, stop pins are installed on the slide rails to prevent accidental displacement.

[0033] III. Design of Non-Reflective Areas

[0034] To create an "unobstructed" environment in the ultrasonic forward channel, a 400mm wide sound-absorbing strip is installed in front of the reference plane. This strip is constructed from layers of 35mm thick polyurethane sound-absorbing foam with a 90% porosity, covered with flame-retardant fleece fabric. Sound-absorbing foam is also applied to the corresponding area on the top panel of the enclosure, while the side walls are covered with tapered sound-absorbing wedges to reduce leakage reflections. A yellow warning label is affixed to the leading edge of the sound-absorbing strip to help operators keep the channel clean during material loading.

[0035] IV. Hardware Connection of the Ultrasonic Testing Unit

[0036] The ultrasonic testing unit uses shielded cables for continuous communication with the ultrasonic sensors of the robot under test. The system captures the analog echo values ​​from the left, right, and forward sensors via a four-channel differential amplifier board and uploads the data in real time to the data acquisition unit of the central control module. To avoid ground loop interference, an independent isolation module is installed between the sensor ground and the testing unit ground.

[0037] V. Test Procedure

[0038] The operator drives the robot into the housing, using the dovetail guide to align the robot's centerline with the reference surface's centerline; then, the operator presses down the quick-positioning fixture to complete the mechanical locking and connect the main line of the testing system.

[0039] After the test control unit reads the fixture status and confirms the locking, it automatically issues an "ultrasonic test start" command and simultaneously writes a synchronization timestamp to the three-channel sensor.

[0040] The data acquisition unit samples the left, right, and forward ranging values ​​in parallel at a period of 0.5ms, with each sampling window lasting 100ms; the data bus bandwidth is 10Mbps, which meets the real-time requirements.

[0041] In some examples, the data analysis unit executes the following sequentially:

[0042] a) Compare the average left and right distance measurements with the 25cm fixed distance threshold; a deviation of less than ±2cm is considered passing.

[0043] b) Compare the average forward distance with the "barrier-free" threshold configured in the system. If it returns to infinity or exceeds 2m, it is considered passed.

[0044] c) Check the variance of the three-channel data. If any variance exceeds the preset jitter threshold, it is deemed unqualified and the corresponding channel number is recorded.

[0045] If all three comparisons pass, the data analysis unit generates an "ultrasonic sensor qualified" flag and writes it to the test report buffer; otherwise, it writes an "unqualified" flag and anomaly details.

[0046] After all sensor tests are completed, the results output module aggregates the information and generates a structured evaluation report containing sensor type, sampling timestamp, mean distance, variance, and pass / fail indicators, which is then uploaded to the manufacturing execution system.

[0047] VI. Maintenance and Calibration

[0048] Every 500 complete tests, the acoustic reflector and sound-absorbing strip require a visual beam scan calibration. The calibration tool uses a robotic vision sensor to identify the pixel positions at the edge of the reflector, automatically calculates the offset from the center line of the reference surface, and updates the system configuration file after correction. If the offset exceeds ±0.5mm, manual tightening and recalibration are performed.

[0049] Through the above embodiments, rapid determination of ultrasonic left and right distance and unobstructed conditions in front can be achieved at a single workstation, providing a highly repeatable benchmark scenario for multi-sensor collaborative testing of the whole machine, while ensuring that the test results are traceable, easy to maintain and quantifiable.

[0050] In some embodiments, the test chamber has multiple geometric reflective surfaces in front of the mounting reference face of the robot under test. The multiple geometric reflective surfaces are distributed horizontally in the central region and the opposite left and right side regions to provide a multi-directional point cloud reference target for the lidar sensor.

[0051] Specifically, the distance between the front wall of the test chamber and the robot's mounting reference surface is set at 800mm, and three geometric reflective surfaces are arranged in this clearance area. All three reflective surfaces are made of birch wood copper-clad sheet, 10mm thick, with a diffuse reflection suppression layer sprayed on the surface, retaining only the high-reflectivity copper foil on the front to enhance laser echo. The three reflective surfaces, arranged horizontally from left to right, are the "left reflective surface," "center reflective surface," and "right reflective surface." The "center reflective surface" faces the robot's forward-looking direction. The angle between the "left reflective surface" and the "center reflective surface" is θ1, and the angle between the "right reflective surface" and the "center reflective surface" is θ2. θ1 and θ2 are uniformly set to be equal during the initial manufacturing process based on the horizontal field of view of the laser radar of the robot under test. They can be finely adjusted within ±5° using a pivot-type limiting mechanism to match different radar models. A dovetail slide rail is installed at the lower end of the reflective surface, along with a scale, to achieve micro-adjustment of the front and rear distances. The scale resolution is 1mm, used to control the point cloud landing point distance parameter.

[0052] In some examples, the installation and debugging process is as follows:

[0053] During initial installation, the operator inserts the three reflective surfaces into the corresponding slide rails, moves them along the scale to the calibrated distance L0, and secures them with butterfly locking screws.

[0054] Adjust the left and right reflector shafts to ensure that the horizontal angle between them and the central reflector meets the set value θ (the fixture is equipped with a laser angle gauge for reading).

[0055] A high-precision handheld laser rangefinder was used to measure the vertical distance from the front edge of the three reflectors to the reference plane, and the error was controlled within ±2mm; if the error exceeded the tolerance, the slide rail was finely adjusted and relocked.

[0056] The test system's automatic calibration module is activated, enabling the robot's LiDAR to scan in low-speed mode. The central control module captures the point cloud and quickly determines the projection of the three reflective surfaces. If the profile deviates from the system's stored reference template by more than a threshold, a manual reset is prompted. After successful calibration, the status "Geometric Reference OK" is written.

[0057] Furthermore, the test control unit issues a "LiDAR test start" command, and the radar rotation speed is set to the standard factory mode; the data acquisition unit acquires 360° raw point cloud frames at a 5MHz streaming rate.

[0058] The data analysis unit first divides the point cloud into three segments—left, middle, and right—based on the horizontal angle interval, and then applies a line fitting algorithm based on random sampling consistency to each segment to obtain a local planar model. Here, "based on random sampling consistency" means randomly sampling a subset of points from that angle segment and iterating repeatedly to find the line parameters with the minimum residual, thereby reducing outlier interference and improving fitting stability.

[0059] For each point set, the "average vertical distance" and "vertical distance variance" indices are calculated to evaluate the radar's planar detection error and scanning stability, respectively. At the same time, the "angle between the fitted line and the horizontal reference" is recorded to infer the radar's elevation installation error.

[0060] If the average vertical distance of all three segments is less than 1 cm and the vertical distance variance is less than the manufacturing preset threshold, and the radar elevation installation error is within ±0.5°, then the lidar is deemed to have passed the test; otherwise, the reason for failure and the specific out-of-tolerance index are recorded.

[0061] The "multi-azimuth point cloud reference target" uses three reflective surfaces with defined geometric relationships within the horizontal plane to acquire point cloud data segments from different azimuths in a single scan. Compared to traditional single-target reciprocating movement or turntable-driven methods, this arrangement eliminates the need for mechanical movement to complete multi-angle point cloud sampling, simplifying the tooling and eliminating motion errors. Combined with random sampling consistency fitting and piecewise error statistics, it can not only detect the radar's horizontal scanning accuracy but also calculate the elevation installation angle within the same data stream, achieving integrated, static-condition multi-parameter verification.

[0062] In practical applications, after every 1000 radar tests, the system automatically prompts for a "geometric reflector re-inspection". The operator uses a laser rangefinder and angle gauge to recheck the L0 and θ values ​​according to the installation and calibration procedures and takes photos for archiving; if any index exceeds the tolerance, the reflector is replaced or the slide rail zero position is recalibrated to ensure long-term test consistency and traceability.

[0063] Through the above embodiments, the multi-geometric reflective surface layout at the front of the test chamber provides a stable and adjustable multi-directional point cloud reference target without the need for complex mechanisms. This provides a highly repeatable benchmark scenario for the rapid detection of the planar detection accuracy and installation angle of the lidar sensor, and works in synergy with ultrasonic, vision and other testing units to enable the testing of multiple sensors of the whole machine to be completed in a single station.

[0064] In some embodiments, the multi-sensor testing module includes:

[0065] The ultrasonic testing unit is used to collect ranging data output by the ultrasonic sensor and compare it with the preset distance parameters corresponding to the ultrasonic acoustic reflecting surface.

[0066] The lidar test unit is used to collect point cloud data output by lidar sensors and compare it with preset geometric parameters corresponding to the lidar geometric reflective surface.

[0067] The vision testing unit is used to acquire depth images output by the vision sensor and compare them with preset flatness parameters corresponding to the visual high flatness reference surface.

[0068] Specifically, the multi-sensor test module is installed on the top back panel inside the test enclosure. The back panel integrates high-density aviation sockets and connects to the central control module via a multi-core hybrid cable. The hybrid cable internally houses a high-speed Ethernet twisted-pair cable, a four-core CAN FD harness, and an isolated power supply line, providing bandwidth, synchronization clock, and regulated power to the three types of test units at the physical level. The central control module has twelve programmable logic input / output ports, which can be expanded to subsequent test units such as barometers and motor encoders via pluggable terminals.

[0069] In some examples, the ultrasonic testing unit includes:

[0070] Hardware Interface: The ultrasonic testing unit includes a three-channel differential amplifier board. The input terminals of the board are connected to the analog echo ports of the left, right, and forward ultrasonic sensors via shielded cables, and the output terminals are connected to the CAN FD bus via an ADC sampling module.

[0071] Data acquisition: The ADC module synchronously acquires the ranging values ​​of the three channels at a sampling rate of 2kHz; the sampling clock is frequency divided and phase-locked by the central control module to ensure time consistency across units.

[0072] Threshold management: The system configuration file sets a fixed target distance threshold D1 for the left and right directions and a "barrier-free" judgment threshold D2 for the forward passage. The thresholds are stored in the EEPROM of the central control module and can be updated online via the user interface.

[0073] Comparison process: The data analysis unit calculates the average distance measurement value ΔL within each sampling window in real time. l ΔL r And ΔLf, and compare it with D1, D2 respectively; if |ΔL1-D1|≤δ, |ΔLf ... r If -D1|≤δ and ΔLf≥D2, then mark the result of this round as "PASS", where δ is given by the system-level accuracy requirements.

[0074] In some examples, the lidar test unit includes:

[0075] Hardware Interface: The lidar test unit is connected to the central control module via gigabit Ethernet direct connection to avoid the blockage of high-speed rotating scanning data in the fieldbus.

[0076] Data Acquisition: The radar outputs a complete 360° point cloud frame, and the data acquisition unit splits the point cloud into three buffer areas (left, middle, and right) in real time according to the horizontal angle threshold table.

[0077] Geometric parameters: When the system leaves the factory, the spatial coordinates of the geometric reflective surfaces are written into the "geometric parameter table" through the 3D CAD model, including the normal vector of each reflective surface, the relationship between the reference edge line and the box coordinate system.

[0078] Comparison process: The data analysis unit performs random sampling consistency line fitting on each segment of the cached point cloud to obtain the fitting model M. i ; Calculate the distance from point M i The mean distance ε i With variance σ i And compare it with the upper tolerance limits εmax and σmax in the geometric parameter table; at the same time calculate M i Offset Δθ in the vertical direction i And compare it with the allowable deviation of pitch angle θmax. If all three segments satisfy ε i ≤εmax、σ i ≤σmax、|Δθ i If |≤θmax, then the result of this round is recorded as "PASS".

[0079] In some examples, the visual test unit includes:

[0080] Hardware Interface: The vision testing unit consists of an RGB-D camera module and a synchronous trigger control board. The trigger board receives synchronization commands via CAN FD and drives the camera to complete the exposure. The depth image is transmitted back to the central control module in real time via a USB-3.0 interface.

[0081] Reference surface parameters: A glass-ceramic plate with a flatness of ≤0.5mm is fixed at the top and bottom of the enclosure. The central control module stores its normal vector and position matrix, which are used as the flatness comparison reference.

[0082] Data processing: The data analysis unit calls the depth map to point cloud algorithm to convert the depth data of each frame into a point cloud P; after deleting NaN and outliers, the number of invalid points n_invalid and the total number of points N are counted, and the hole rate ρ=n_invalid / N is calculated; at the same time, the plane fit η=(λ1+λ2) / (λ1+λ2+λ3) is calculated through the eigenvalues ​​λ1, λ2, and λ3 of the covariance matrix of P.

[0083] Comparison conditions: If ρ≤ρmax and η≥ηmin, then the frame is considered to meet the flatness requirements; if K consecutive frames pass, "Visual Sensor PASS" is output. ρmax, ηmin and K are set by the user in the system configuration interface.

[0084] This embodiment introduces a "unified threshold management and parallel data channel" mechanism. Thresholds, geometric parameters, and reference surface normals are all written into a central configuration table, which is retrieved uniformly by the data analysis unit during testing, avoiding the inconsistency risks caused by each testing unit independently maintaining parameters. Simultaneously, ultrasonic testing uses a CAN FD bus, LiDAR uses Gigabit Ethernet, and the vision unit uses USB-3.0. All three parallel data channels are driven by a synchronous clock from the central control module, achieving millisecond-level alignment across sensors and establishing a unified time base for subsequent system fusion judgment or trend prediction software.

[0085] After the operator initiates the test with a single click, the test control unit issues a synchronous trigger. The three test units upload raw data in parallel via their respective communication links. The data analysis unit, relying on a unified parameter table, completes the comparison of ultrasonic distance thresholds, calculation of lidar point cloud fitting errors and pitch angles, and determination of visual hole rate and fit degree within the same cycle. The results are then used to generate a pass / fail flag for each sensor through logical AND operations and written to the test report cache. Finally, the central control module aggregates the timestamp, threshold version number, and pass / fail flag, outputting a standardized XML file via USB or network interface for automatic parsing by the production line quality system.

[0086] Through the above embodiments, the multi-sensor testing module achieves synchronous acquisition and unified judgment of ultrasonic, lidar and vision sensors without hardware interference, which fully reflects the technical concept of "one-stop multi-sensor collaborative testing" of this application and its system architecture that can be industrialized.

[0087] In some embodiments, the ultrasonic testing unit is configured as follows:

[0088] Simultaneously acquire distance measurement data from the ultrasonic sensor in the left, right, and forward directions;

[0089] The distance measurement data in the left and right directions are compared with the corresponding left and right distance thresholds, respectively.

[0090] Compare the distance measurement data in the forward direction with the accessibility threshold;

[0091] When the ranging data in all three directions meets the corresponding threshold, the ultrasonic sensor outputs a qualified signal.

[0092] Specifically, the ultrasonic testing unit consists of three parts: a sensor interface board, a synchronous sampling module, and a threshold determination module. The sensor interface board is connected to the analog echo ports of the left, right, and forward ultrasonic sensors of the robot under test via shielded cables; the synchronous sampling module has a built-in 3-channel 12-bit ADC with a sampling resolution of 1mm and a maximum sampling rate of 2kHz; the threshold determination module uses an FPGA logic unit to implement data buffering, threshold comparison, and qualified signal generation, and its output is connected to the central control module via CANFD.

[0093] When the test control unit issues the "ultrasonic test start" command, the synchronous sampling module triggers its internal phase-locked loop, locking the sampling clocks of the three ADC channels to the same rising edge. Each frame acquisition window is set to 100ms, and the sampling module continuously acquires the left-direction ranging value L within the window. l Distance measurement value L in the right direction r and forward distance measurement value L f And calculate the average value ΔL in each direction. l ΔLr ΔL f Synchronous sampling ensures that data from the three directions are compared on the same time base, avoiding instantaneous errors caused by environmental noise or robot micro-vibrations.

[0094] The system configuration file presets left and right distance thresholds D1 and obstacle-free thresholds D2. D1 is set based on the calibrated distance from the acoustic reflective surface of the test chamber to the robot's outer shell, while D2 is set to exceed the upper limit of the measurement range based on the "no echo" characteristic of ultrasonic waves. The threshold determination module receives ΔL1 and ΔL... r ΔL f Then execute in sequence:

[0095] Judgment |ΔL l -D1|≤δ and|ΔL r -D1|≤δ, where δ is the allowable deviation, defaulting to 2cm;

[0096] Judge ΔLf≥D2;

[0097] If both of the above conditions are met, the PASS flag will be set in this acquisition window; otherwise, the FAIL flag will be set and the out-of-tolerance direction and error value will be recorded.

[0098] In some examples, when the PASS flag is true for N consecutive frames within the acquisition window, the threshold determination module generates an "ultrasonic sensor qualified" signal U_PASS, along with a timestamp T and threshold version number V, and sends it to the central control module via a CAN FD message. Upon receiving U_PASS, the central control module writes it to the test report buffer; if FAIL is detected in any window, the system immediately terminates the ultrasonic test and records the fault details.

[0099] This embodiment implements "multi-directional synchronous sampling + unified threshold determination" through FPGA hardware. Compared with the traditional polling test method, it can complete the alignment of three-directional data and one-time determination within a millisecond time window, avoiding the error introduced by changes in environmental conditions when testing in different directions. At the same time, the continuous N-frame verification mechanism improves the robustness against noise, ensuring that qualified signals are only output when the threshold is stably met, providing a reliable foundation for subsequent multi-sensor collaborative evaluation.

[0100] In some embodiments, the lidar test unit is configured as follows:

[0101] The point cloud data collected by the lidar sensor is divided into multiple data segments according to the orientation;

[0102] A linear fitting algorithm based on random sampling consistency is applied to each data segment to obtain the corresponding fitting model;

[0103] Based on the distance distribution from the midpoint of each data segment to the corresponding fitted model, determine whether the planar detection error of the lidar sensor falls into the error threshold.

[0104] Based on the vertical offset of the fitted model for each data segment, the pitch installation angle of the lidar sensor is calculated and compared with the installation angle threshold.

[0105] Specifically, the lidar test unit is directly connected to the central control module via a gigabit Ethernet cable to avoid congestion of high-frequency point clouds in the fieldbus. The lidar operates in standard rotation mode, outputting 360° two-dimensional point cloud frames at a scanning frequency of 10Hz, with more than 9000 points per frame. The central control module records the first frame as a baseline immediately after starting the test, and subsequent frames are analyzed under the same time baseline.

[0106] Furthermore, the central control module has a pre-set azimuth segmentation table, defining the horizontal angle from -30° to -10° as the left segment, -10° to 10° as the center segment, and 10° to 30° as the right segment. Other azimuth data are ignored to reduce wall clutter interference. After receiving the complete point cloud frame, the module allocates the point cloud to the corresponding buffer based on the horizontal angle interpolation index of each point, achieving zero-copy segmentation without duplication.

[0107] Furthermore, each azimuth segment independently invokes the RANSAC straight-line fitting algorithm. The algorithm parameters are set by a central configuration file, including a maximum of 300 iterations, a sample size of 2, and an interior point threshold of 3mm. The algorithm flow is as follows:

[0108] Randomly select two points within the segment to calculate candidate lines and count the number of points inside the line.

[0109] Repeat the iteration until the proportion of interior points or the number of iterations meets the exit condition;

[0110] Perform least squares regression on the optimal set of interior points to obtain the final fitted model;

[0111] After fitting, the coefficients of the linear equation and the root mean square error of the interior points are output for subsequent error evaluation.

[0112] In some examples, the central control module reads the distance distribution from the interior points to the fitted model for each segment, and calculates the average distance d_avg and the standard deviation d_std. The system presets a plane error threshold d_max and a jitter threshold σ_max. If d_avg ≤ d_max and d_std ≤ σ_max, the segment is considered qualified; otherwise, "plane error exceeds limits" is recorded, and the segment number and the out-of-tolerance value are marked.

[0113] The vertical offset Δh between the center segment and the left and right segments of the straight line is compared by fitting the model to calculate the elevation angle α = arctan(Δh / Δx) of the radar scanning surface, where Δx is the horizontal distance between the reflector and the vertical projection of the radar. α is compared with the system-set allowable deviation α_max of the elevation angle. If |α| ≤ α_max, the elevation angle is acceptable; otherwise, "elevation angle exceeds limit" is recorded and the angle value is given.

[0114] In some examples, when the planar errors of the three azimuth segments are all qualified and the pitch angle judgment is passed, the lidar test unit generates a "lidar sensor qualified" signal L_PASS with a timestamp and fitting error statistics, and uploads it to the central control module via Ethernet; if any indicator fails, the system immediately outputs a fault report and stops subsequent point cloud sampling.

[0115] The orientation segmentation strategy avoids edge distortion caused by overfitting of the whole frame point cloud, and focuses the detection on key reflection areas; random sampling consistent line fitting can robustly extract the reflection surface features in the presence of outliers and local occlusions, and is more fault-tolerant in multipath noise scenarios than traditional global least squares fitting; the pitch angle is calculated by inter-segment vertical offset rather than single-point angle measurement, and the installation angle can be checked without additional attitude sensors, reducing tooling complexity and improving detection speed.

[0116] In some embodiments, the vision testing unit is disposed on the top and bottom of the test chamber, and on upper and lower reference surfaces whose surface flatness meets a preset flatness tolerance; the vision testing unit is configured as follows:

[0117] The depth images acquired by the visual sensor are converted into point cloud data, and invalid and outlier points are removed.

[0118] The number of invalid points is counted and compared with the total number of points to obtain the void ratio, which is then compared with the void ratio threshold.

[0119] The plane fit coefficient is calculated based on the eigenvalues ​​of the covariance matrix of point cloud data and compared with the fit threshold to generate the visual sensor judgment result.

[0120] Specifically, a 1000mm × 800mm glass-ceramic reference surface is fixed to the top and bottom of the test chamber. The flatness tolerance of both reference surfaces is controlled within 0.5mm after interferometry testing. Both reference surfaces are coated with anti-glare black paint, and flexible LED light strips are installed around them to provide uniform diffuse illumination and avoid interference from strong reflection points. The RGB-D camera module is mounted on a top-view bracket directly above the robot's mounting reference surface and a bottom-view frame directly below it. The bracket angle is adjusted via a precision lead screw to ensure that the camera's optical axis is perpendicular to the corresponding reference surface. The camera connects to the central control module via a USB-3.0 interface.

[0121] In some examples, the central control module sends exposure commands to the synchronization trigger control board via CANFD. The upper and lower cameras simultaneously acquire a depth image frame with a resolution of 640×480 and a quantization step of 1mm. The depth image is transmitted in real time to the shared memory area of ​​the central control module via USB-3.0 to ensure that the time base of the two perspectives is consistent.

[0122] In some examples, the data analysis unit calls a depth map to convert the depth value of each pixel into 3D coordinates, obtaining a point cloud P. Two preprocessing steps are then performed:

[0123] Remove invalid points: If the depth value is 0 or exceeds the measurement limit, mark it as NaN and delete it from P;

[0124] Outlier removal: A radius filtering method is used, with the search radius set to 30mm. When the number of neighbors of a point is less than the threshold n_min, it is identified as an outlier and deleted. n_min is preset based on the camera resolution and the reference distance.

[0125] In some examples, after preprocessing, the number of invalid points n_invalid is counted and compared with the total number of point clouds N, and the hole rate ρ = n_invalid / N is calculated. The system configuration file sets the hole rate threshold ρ_max; if ρ ≤ ρ_max, it is judged as "hole rate qualified"; otherwise, it is recorded as "unqualified" and proceeds to the next frame for evaluation.

[0126] In some examples, the covariance matrix C is calculated for the point cloud P, and the eigenvalues ​​λ1≤λ2≤λ3 are extracted. The plane fit η is obtained using the formula η=(λ1+λ2) / (λ1+λ2+λ3). The system sets a fit threshold η_min; if η≥η_min, then the fit is considered "qualified". The central control module requires K consecutive frames to simultaneously meet the hole rate and fit qualification criteria within one test cycle. K is usually set to 5 to suppress occasional noise.

[0127] In some examples, after K consecutive frames are deemed successful, the data analysis unit generates a "vision sensor qualified" signal V_PASS, along with the hole rate ρ, fit η, timestamp T, and threshold version number V, which is written back to the central control module via USB-3.0. If ρ > ρ_max or η < η_min is detected in any frame, the system immediately marks "vision sensor unqualified" and records the abnormal frame number and corresponding statistical value.

[0128] This embodiment employs a dual-metric evaluation of depth image quality: "hole rate + plane fitting degree." The hole rate directly reflects the completeness of depth perception, while the plane fitting degree measures the accuracy of plane reconstruction. These two metrics complement each other, simultaneously capturing quality degradation caused by insufficient illumination and structured light distortion. A continuous frame sliding window mechanism enhances decision robustness, preventing misjudgments due to occasional errors in a single frame. Furthermore, strict control of the flatness of the upper and lower reference planes, combined with uniform diffuse illumination, minimizes the impact of high-brightness areas and shadows on depth measurement, achieving stable and repeatable verification of the visual sensor's performance.

[0129] In some embodiments, the multi-sensor test module further includes an IMU test unit, which is configured to:

[0130] The system collects raw data from the gyroscope and accelerometer, calculates the sampling frequency deviation, zero-bias drift, and acceleration jump index, and outputs an IMU sensor qualification signal when all three indicators meet their respective thresholds.

[0131] Specifically, the IMU test unit is mounted on a vibration-damping tray below the reference surface of the test chamber. The tray is isolated from the bottom plate of the chamber by rubber damping pads, and its pulse resonance frequency is below 20Hz, which can shield against ground micro-vibrations. The IMU of the robot under test is connected to the signal acquisition board of the test unit via a high-speed SPI bus; the acquisition board has a built-in 48MHz precision crystal oscillator, which provides an independent time base for subsequent frequency deviation calculations.

[0132] In some examples, after the test control unit sends an "IMU test start" command, the acquisition board begins to retrieve raw triaxial data from the gyroscope and accelerometer at a set nominal sampling rate f_nom (e.g., 400Hz), and adds a hardware timestamp t_i (in 1µs) to each data point. The continuous sampling time window is set to 60s to ensure statistically significant drift assessment.

[0133] In some examples, the metric calculation includes:

[0134] 1. Sampling frequency deviation Δf

[0135] Calculate the difference between adjacent timestamps Δt_i = t_i - t_{i-1};

[0136] Calculate the average sampling period of the window and obtain the actual sampling rate f_act=1 / ,\overline{Δt};

[0137] The sampling frequency deviation Δf = |f_act - f_nom| / f_nom.

[0138] 2. Zero bias drift β_g, β_a

[0139] If the robot is kept stationary, the theoretical gyroscope angular velocity is 0° / s;

[0140] Calculate the three-axis mean of the gyroscope (ω) x ,ω y ,ω z The largest absolute value is β_g;

[0141] For the accelerometer, the theoretical output is (0,0,±g), and the value with the largest absolute difference between the mean value of each axis and the theoretical value is taken as β_a;

[0142] Zero bias drift β = max(β_g, β_a).

[0143] 3. Acceleration jump δ_a

[0144] Calculate the acceleration difference Δa_j = |a_{j} - a_{j-1}| between adjacent samples within a sliding window w = 25ms;

[0145] When Δa_j > δ_thr, it is counted as a "jump event", and δ_thr is preset according to the sensor noise density;

[0146] Count the number of jump events N_spike per unit time, and then convert it into the jump rate δ_a = N_spike / 60s.

[0147] In some examples, the system configuration file specifies the following thresholds: sampling frequency deviation threshold Δf_max = 0.5%, zero-bias drift threshold β_max = 0.05° / s (or 5mg), and jump rate threshold δ_a_max = 2 times / 60s. The data analysis unit compares Δf ≤ Δf_max, β ≤ β_max, and δ_a ≤ δ_a_max. If all conditions are met, an "IMU sensor qualified" signal I_PASS is generated at the end of the test; otherwise, "unqualified" is output along with the out-of-limit indicators and their values.

[0148] This embodiment uses an independent crystal oscillator timestamp to eliminate the impact of robot main control clock jitter on frequency deviation calculation. Simultaneously, a sliding window is used to statistically analyze the jump rate, distinguishing between occasional high-frequency spikes and systematic gyroscope jitter. By fixing the robot to a vibration-damping tray and locking the actuator, the theoretical gyroscope output is ensured to be zero, allowing direct estimation of zero-bias drift using the mean method without the need for an additional rotary table. Sampling frequency deviation focuses on time base reliability, zero-bias drift measures static accuracy, and acceleration jumps capture high-frequency noise. The combination of these three allows for a comprehensive assessment of IMU key performance in a single static measurement, enabling rapid sorting of production line products without the need for lengthy Allan variance testing.

[0149] In some embodiments, the central control module includes:

[0150] The test control unit is used to synchronously issue start commands to each test unit and maintain a uniform time reference during the test.

[0151] The data acquisition unit is used to acquire the raw data streams of each test unit in parallel under the scheduling of the test control unit;

[0152] The data analysis unit performs error calculation, stability statistics, and fit evaluation on the raw data stream to generate performance evaluation results for each sensor.

[0153] Specifically, the central control module adopts an integrated industrial computer, with the motherboard integrating a six-core 1.6GHz ARM processor, 2GB DDR4 ECC memory, and a 256GB NVMe solid-state drive. It features three onboard communication interfaces: Gigabit Ethernet for high-speed point cloud processing with LiDAR, dual-channel CANFD for ultrasonic and IMU data, and USB-3.0 for visual depth image acquisition. The motherboard also incorporates a 10MHz temperature-compensated crystal oscillator and a GPS-PPS pulse capture circuit, which, via a phase-locked loop, generate a system clock tree, providing a nanosecond-level synchronization reference for the test control unit, data acquisition unit, and data analysis unit.

[0154] In some examples, the test control unit runs on the real-time Linux kernel Space-1, receiving PPS pulses via hardware interrupts and correcting system clock drift of ≤50ns per second. The unit employs a finite state machine process: INIT → WAIT_SYNC → RUN → FINISH. Upon entering the RUN state, a start message is broadcast to each test unit once, carrying a 64-bit absolute timestamp T0; each test unit then counts down to start accordingly, ensuring synchronized start of the test activities within a ±200µs window. During operation, the test control unit monitors the heartbeat packets of each test unit; if a heartbeat times out, a safety shutdown is triggered.

[0155] In some examples, the data acquisition unit deploys three zero-copy circular buffer channels in user space: EtherRing, CanRing, and UsbRing. The driver layer uses DMA to directly write Ethernet frames, CANFD frames, and USB-3.0 Bulk packets into the corresponding circular buffers, avoiding inter-core copying. The acquisition thread adopts a lock-free single-producer, multi-consumer model, stuffing data into a unified MetaQueue in ascending order of packet arrival timestamps. Each record includes a channel ID, timestamp, payload pointer, and length field, ensuring cross-channel data alignment on a single timeline.

[0156] In some examples, the data analysis unit implements an "algorithm plugin container" framework, with a lightweight scheduler as its core, capable of dynamically loading .so plugin files. Each type of sensor corresponds to one algorithm chain.

[0157] Ultrasonic chain: Original distance → Mean and variance calculation → Threshold comparison;

[0158] LiDAR chain: point cloud segmentation → RANSAC fitting → error statistics → elevation angle calculation;

[0159] Visual chain: Depth to point cloud → Hollow rate statistics → Plane fit calculation;

[0160] IMU chain: Frequency deviation detection → Zero-bias drift estimation → Jump rate statistics.

[0161] The scheduler allocates independent threads and CPU affinity to each chain, and uses timestamp indices in the circular buffer to retrieve raw data, ensuring that the algorithm input has no dropped frames and that the time base is consistent. The end of the chain outputs a triplet of error E, stability S, and fit F within the sampling period, and writes it to the ResultBuffer.

[0162] In some examples, when the test control unit detects that all acquisition windows have ended, it hands over the ResultBuffer to the ReportBuilder thread. The ReportBuilder combines the sensor ID, threshold version number, and timestamp to generate a JSON structured entry, which is then uploaded to the result output module via TLS or written to a local XML backup.

[0163] Based on the method described in this embodiment, this embodiment has the following key features:

[0164] Unified Time Base for Multiple Buses: The GPS-PPS and temperature-compensated crystal oscillator composite clock tree unifies the asynchronous data from Ethernet, CANFD, and USB-3.0 buses onto the same nanosecond-level time axis.

[0165] Lock-Free Circular Buffer: DMA direct write + Lock-Free dequeue eliminates inter-core mutual exclusion, ensuring no packet loss of gigabit point cloud and high-frequency IMU data under full load.

[0166] Plug-in algorithm chain: Algorithms are hot-loaded in the form of plug-ins, and sensor algorithms can be added or removed without restarting, enabling rapid iteration and maintenance of the production line.

[0167] The three-dimensional evaluation uses a unified approach: error E reflects accuracy, stability S reflects noise characteristics, and fit F reflects model consistency. The three-dimensional evaluation outputs a PASS / FAIL table, balancing the comprehensiveness of the indicators with the ease of analysis.

[0168] With the above configuration, the central control module completes millisecond-level multi-sensor synchronous control, microsecond-level data alignment, and second-level result aggregation on the single-board hardware, providing high-precision, high-throughput, and easy-to-maintain core scheduling and analysis capabilities for the robot multi-sensor integrated testing system.

[0169] The above embodiments have described in detail the specific module structure and functions of the robot multi-sensor integrated testing system of this application. The implementation process of the multi-sensor integrated testing method of this application will be described in detail below with reference to specific embodiments. Figure 2 This is a flowchart illustrating the multi-sensor integrated testing method based on a robot multi-sensor integrated testing system provided in this application embodiment, as shown below. Figure 2 As shown, this multi-sensor integrated testing method may specifically include the following steps:

[0170] S201, Position the robot under test on the mounting reference surface of the test box and complete the electrical connection with the multi-sensor test module;

[0171] S202, using the test control unit in the central control module to synchronously send start commands to the ultrasonic test unit, lidar test unit and vision test unit and establish a unified time reference;

[0172] S203 utilizes the data acquisition unit in the central control module to acquire the raw data streams output by each test unit in parallel under a unified time reference;

[0173] S204, the acquired raw data stream is processed and associated with the corresponding reference surface or preset parameters, and the data analysis unit performs error calculation, stability statistics and fit evaluation based on the processing results to obtain the performance evaluation results of each sensor.

[0174] S205: Input the performance evaluation results into the result output module for aggregation, and generate a robot multi-sensor performance evaluation report containing the performance evaluation results of each sensor.

[0175] Specifically, the robot under test first drives into the test chamber and stops at the granite mounting reference surface. The positioning fixture guides the robot's centerline to align with the reference surface's centerline via a dovetail slide, and then a four-point clamping mechanism locks the chassis to prevent slight movement. Simultaneously, the robot's power supply busbar, Ethernet, CANFD, and USB 3.0 adapter ports are inserted into the reserved sockets of the multi-sensor test module, forming an electrical closed loop. After locking is complete, the test control unit reads the fixture limit switches and interface status, confirming "mechanical completion, communication completion" signals before proceeding to the next step, ensuring the repeatability and safety of the test environment.

[0176] Furthermore, the test control unit generates an absolute timestamp T0 based on the system clock tree and simultaneously sends a start command via Ethernet broadcast frames and CANFD synchronization frames. Upon receiving the command, the ultrasonic, lidar, vision, and optional IMU test units start their internal countdown logic to ensure that their respective ADCs, Ethernet acquisition units, and camera trigger lines operate concurrently within the T0±200µs window. This unified time base allows the raw data reported from different buses to be directly aligned at the back end without the need for subsequent resampling.

[0177] Furthermore, the data acquisition unit is configured with zero-copy ring buffers for the three communication links. Gigabit LiDAR traffic is written to EtherRing via DMADescriptor, high-frequency frames from ultrasound and IMU are written to CANRing via CANRing, and depth images are written to UsbRing via USB 3.0 Bulk packets. The acquisition thread encapsulates the data header pointers of each ring buffer into MetaQueue entries in timestamp order. This parallel acquisition process eliminates inter-core lock contention, ensuring no frame loss even when the peak bandwidth of the LiDAR point cloud is close to 90MBps.

[0178] Furthermore, the data analysis unit loads the corresponding algorithm chain based on the sensor type. For ultrasonic data, it performs a three-way mean and threshold comparison, outputting the distance error Eus and stability Sus; for the lidar point cloud, it performs azimuth segmentation and RANSAC line fitting, calculating the plane error Elidar, standard deviation Slidar, and pitch angle Flidar; after converting the visual depth map into a point cloud, it calculates the hole rate Evision, calculates the covariance matrix eigenvalues ​​to obtain the fitting degree Fvision, and uses a frame sliding window method to calculate the stability Svision; if the IMU chain is enabled, it generates the frequency deviation Eimu, zero-bias drift Fimu, and jump rate Simu based on the hardware timestamp. The data analysis unit writes the <E,S,F> triplet and the pass / fail flag in a unified queue, forming mutually referential performance evaluation results.

[0179] Furthermore, the results output module calls ReportBuilder to read all sensor results and automatically assembles them into a JSON-formatted performance evaluation report. The report includes the test number, timestamp T0, threshold version V, values ​​for each <E, S, F>, and PASS / FAIL flag. The system writes this file to an NVMe backup and pushes it via HTTPS to the Manufacturing Execution System (MES) and the after-sales cloud platform, achieving seamless integration of production quality control and in-service traceability. When any sensor is marked as non-compliant, ReportBuilder immediately triggers an alarm and highlights the anomaly on the control panel, facilitating troubleshooting personnel to locate the cause and retest.

[0180] Through the above five steps, this method enables the robot under test to simultaneously collect data from multiple sensors in a single workstation, determine unified thresholds, and output standardized reports. This significantly shortens the testing cycle and ensures the comparability of horizontal data, providing a reliable basis for batch delivery and subsequent maintenance.

[0181] It should be understood that the sequence number of each step in the above method embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this disclosure.

[0182] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although the technical solutions of this application have been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A robot multi-sensor integrated testing system, characterized in that, include: The test chamber is used to form a test space with a fixed mounting reference surface for the robot under test, and at least one set of ultrasonic acoustic reflective surfaces, lidar geometric reflective surfaces and visual high-flatness reference surfaces are set in the test space according to a preset spatial distribution, so as to provide standardized reference targets for the ultrasonic sensors, lidar sensors and vision sensors of the robot under test respectively. A multi-sensor testing module is installed inside the test chamber and electrically connected to the robot under test. The multi-sensor testing module includes an ultrasonic testing unit, a lidar testing unit, and a vision testing unit. The central control module communicates bidirectionally with the multi-sensor test module. The central control module includes a test control unit, a data acquisition unit, and a data analysis unit. The result output module is signal-connected to the data analysis unit and is used to generate a robot multi-sensor performance evaluation report based on the performance evaluation results of each sensor generated by the data analysis unit, so as to determine sensor consistency based on the robot multi-sensor performance evaluation report. The test chamber is provided with acoustic reflective surfaces on the left and right sides of the mounting reference surface of the robot under test, and each acoustic reflective surface maintains a preset fixed distance from the mounting reference surface of the robot under test, and forms a non-reflective area in front of the mounting reference surface of the robot under test, so as to provide the ultrasonic sensor with left and right fixed distance reference targets and unobstructed reference targets in front. The test chamber has multiple geometric reflective surfaces in front of the mounting reference face of the robot under test. These multiple geometric reflective surfaces are distributed horizontally in the central area and the opposite left and right side areas to provide multi-directional point cloud reference targets for the lidar sensor. The multi-sensor testing module includes: The ultrasonic testing unit is used to collect the ranging data output by the ultrasonic sensor and compare it with the preset distance parameters corresponding to the ultrasonic acoustic reflecting surface. The lidar test unit is used to collect point cloud data output by the lidar sensor and compare it with the preset geometric parameters corresponding to the lidar geometric reflective surface. The visual testing unit is used to acquire the depth image output by the visual sensor and compare it with the preset flatness parameters corresponding to the visual high flatness reference surface. The ultrasonic testing unit is configured as follows: The ultrasonic sensor simultaneously acquires ranging data in the left, right, and forward directions; The distance measurement data in the left and right directions are compared with the corresponding left and right distance thresholds, respectively. Compare the distance measurement data in the forward direction with the accessibility threshold; When the ranging data in all three directions meet the corresponding thresholds, the ultrasonic sensor outputs a qualified signal. The lidar test unit is configured as follows: The point cloud data collected by the lidar sensor is divided into multiple data segments according to the orientation; A linear fitting algorithm based on random sampling consistency is applied to each data segment to obtain the corresponding fitting model; Based on the distance distribution from the midpoint of each data segment to the corresponding fitted model, determine whether the planar detection error of the lidar sensor falls within the error threshold; Based on the vertical offset of the fitted model for each data segment, the pitch installation angle of the lidar sensor is calculated and compared with the installation angle threshold. The visual testing unit is disposed on the top and bottom of the testing box, and its surface flatness meets the preset flatness tolerance on the upper and lower reference surfaces; the visual testing unit is configured as follows: The depth image acquired by the visual sensor is converted into point cloud data, and invalid points and outliers are removed. The number of invalid points is counted and compared with the total number of points to obtain the void ratio, which is then compared with the void ratio threshold. The plane fit coefficient is calculated based on the eigenvalues ​​of the covariance matrix of the point cloud data and compared with the fit threshold to generate the visual sensor judgment result.

2. The system according to claim 1, characterized in that, The multi-sensor testing module further includes an IMU testing unit, which is configured to: The system collects raw data from the gyroscope and accelerometer, calculates the sampling frequency deviation, zero-bias drift, and acceleration jump index, and outputs an IMU sensor qualification signal when all of these indicators meet their corresponding thresholds.

3. The system according to claim 1, characterized in that, The central control module includes: The test control unit is used to synchronously issue start commands to each test unit and maintain a uniform time reference during the test. A data acquisition unit is used to acquire the raw data streams of each test unit in parallel under the scheduling of the test control unit; The data analysis unit is used to perform error calculation, stability statistics and fit evaluation on the raw data stream, and generate performance evaluation results for each sensor.

4. A multi-sensor integrated testing method based on a robot multi-sensor integrated testing system as described in any one of claims 1 to 3, characterized in that, include: Position the robot under test on the mounting reference surface of the test box and complete the electrical connection with the multi-sensor test module; The test control unit in the central control module synchronously sends start commands to the ultrasonic test unit, lidar test unit, and vision test unit and establishes a unified time reference. The data acquisition unit in the central control module acquires the raw data streams output by each test unit in parallel under the unified time reference. The acquired raw data stream is processed and associated with the corresponding reference surface or preset parameters. The data analysis unit then performs error calculation, stability statistics, and fit evaluation based on the processing results to obtain the performance evaluation results for each sensor. The performance evaluation results are input into the result output module for aggregation, generating a robot multi-sensor performance evaluation report that includes the performance evaluation results of each sensor.

Images