Scene electromagnetic texture surveying and mapping method and system based on autonomous mobile robot

Abstract

The invention discloses a scene electromagnetic texture mapping method and system based on an autonomous mobile robot. The method comprises robot operation and development, radar sensor operation and DART implementation. The robot carries out data acquisition through a sensor and carries out data fusion and processing on the acquired information; the radar sensor operates to detect the surrounding environment through a signal transceiving mechanism, and the obtained signals are fused and processed; in the DART implementation step, visual results of different angles are provided through a visual angle synthesis technology and are finally output in a multi-visual-angle mode. Through a strong coupling mechanism of robot body motion control and radar scanning parameters, a personal error source is eliminated, high-precision motion control is realized based on a robot-radar cooperative control architecture, and based on a DART-NeRF hybrid modeling engine and a lightweight network architecture, model parameters are compressed through adaptive coding, and training and reasoning processes are accelerated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of surveying and mapping, in particular to a scene electromagnetic texture mapping method and system based on an autonomous mobile robot. BACKGROUND

[0002] The existing radar-based scene electromagnetic texture mapping methods mainly include two types: handheld radar-based scene electromagnetic texture mapping method and fixed radar array-based scene electromagnetic texture mapping method. Both of these two existing technologies have obvious application defects.

[0003] The traditional handheld DART data acquisition method has the problem of unstable speed control. The fluctuation of scanning speed caused by manual operation seriously affects the accuracy of Doppler shift calculation. Secondly, the vertical angle coverage range of handheld devices is limited, which cannot cover high altitude or complex structure areas (such as building roofs and pipeline tops), resulting in incomplete data acquisition. In addition, the randomness of the path of manual operation leads to uneven scanning density and poor data correlation. Finally, although the fixed radar array can achieve wide coverage, it has high deployment cost and poor flexibility, which is difficult to adapt to dynamic scene requirements (such as battlefield rapid deployment or complex industrial environments). The present application aims to solve the above problems by using an autonomous mobile robot system to realize high-precision and full-scene coverage of millimeter wave radar scene electromagnetic modeling. SUMMARY

[0004] (I) Technical problems solved In order to solve the above technical problems, the present application provides a scene electromagnetic texture mapping method and system based on an autonomous mobile robot.

[0005] (II) Technical solutions In order to solve the above technical problems and achieve the purpose of the present application, the present application is realized by the following technical solutions: A scene electromagnetic texture mapping method based on an autonomous mobile robot, comprising robot operation and development, radar sensor operation, and DART implementation; The robot operation and development step, the robot collects data through the sensor, and fuses and processes the collected information, establishes a map and locates based on SLAM; The radar sensor operation detects the surrounding environment through a signal transceiver mechanism, and fuses and processes the acquired signals; The DART implementation step includes selecting a suitable display space and constructing a world model based on it, providing visual results at different angles through view synthesis technology, and finally outputting in a multi-view mode.

[0006] Further, the robot includes a main body and a chassis, and the chassis is equipped with two steering motor driven wheels and a universal wheel.

[0007] Further, it also includes real-time coupling of robot chassis kinematics model and radar scanning mechanism.

[0008] Further, it also includes a coarse-scan-fine-scan dynamic triggering mechanism, which first quickly identifies high-concern areas in panoramic scanning mode, and then triggers local fine scanning mode based on laser radar point cloud density and millimeter wave radar RCS features, and dynamically adjusts parameters based on scene complexity evaluation model.

[0009] Further, it also includes millimeter wave radar point cloud data and laser radar scanning frame time stamp alignment, Kalman filter fusion by edge computing unit, and establishment of dynamic environment model.

[0010] Further, it also includes a series of calibration schemes, through specific calibration algorithms, to realize the spatio-temporal alignment of radar range-Doppler heat map, LiDAR point cloud and IMU attitude data.

[0011] Further, it is necessary to point out that through the multi-sensor spatio-temporal fusion method, a deep collaborative architecture of radar, laser radar (LiDAR) and inertial measurement unit (IMU) is constructed, which breaks through the technical bottleneck of time-space mismatch and motion distortion in traditional multi-source data fusion. Based on joint calibration technology and dynamic distortion correction engine, the sensor spatio-temporal synchronization problem is solved by a series of calibration schemes through specific calibration algorithms to realize the spatio-temporal alignment of radar range-Doppler heat map, LiDAR point cloud and IMU attitude data. On this basis, after the millimeter wave radar point cloud data and the laser radar scanning frame are aligned by time stamp, a dynamic environment model is established. And a real-time compensation engine for multi-modal data fusion is constructed, which dynamically adjusts the motion compensation parameters based on the LiDAR point cloud density distribution and radar features. At the same time, the algorithm is introduced to reconstruct the chassis motion trajectory by the encoder odometer historical data, and to non-linearly remap the radar echo timestamp, which to some extent eliminates the motion blur artifacts in the sudden stop working condition. The data redundancy and missed detection problems caused by sensor asynchrony in traditional solutions are solved, and a full-autonomous, zero-human intervention fusion perception solution for high-precision electromagnetic texture modeling in complex industrial environments is provided.

[0012] Further, the IMU high-frequency angular velocity data is combined with the Lyapunov stability criterion optimization to realize real-time compensation of radar beam deflection caused by robot motion.

[0013] Further, the DART implementation includes a three-dimensional reconstruction engine that introduces neural radiance field technology and a lightweight network model based on PyTorch.

[0014] Further, in order to optimize the radar data visualization effect, the system optimizes the radar turnout to improve the accuracy of information presentation. After data processing is completed, the DART system provides different angle visualization results through the new perspective synthesis technology, and finally outputs in a multi-perspective manner to enhance the user's interactive experience. The three-dimensional reconstruction engine innovatively introduces the NeRF technology, and through the lightweight network model realized by PyTorch, the radar point cloud and visual texture are jointly optimized. The network input layer adopts a double-branch structure: the radar branch processes the attenuation coefficient and scattering characteristics, and the optical branch analyzes the material reflection characteristics of the camera, and in the implicit space coding stage, the multi-modal feature fusion is realized through the cross-attention mechanism, and finally the high-fidelity three-dimensional model with electromagnetic properties is output. In order to improve the real-time performance, the system adopts a block-based progressive rendering strategy, divides the scene into 1m³ voxel units, dynamically loads the processing area through octree index, and the reconstruction speed is significantly improved compared with the traditional method.

[0015] The application also provides a scene electromagnetic texture mapping system based on an autonomous mobile robot, which comprises a device system and an implementation system. The device system comprises a robot system and a radar system, and the implementation system is composed of an FPGA, a DSP, a microprocessor and a server. The radar system adopts a dynamic layered scanning strategy, and performs fast panoramic scanning and local fine scanning through a coarse scanning-precise scanning mechanism.

[0016] In addition, in order to achieve the above-mentioned purposes, the application further provides a computer readable storage medium, which stores program instructions of a scene electromagnetic texture mapping method based on an autonomous mobile robot.

[0017] (Three) beneficial effects Compared with the prior art, the beneficial effects of the application are: The problems of speed fluctuation, limited view angle and data consistency in traditional handheld DART data acquisition are solved. The core advantages are as follows: high-precision motion control of the robot ensures constant scanning speed, combined with multi-degree-of-freedom gimbal to expand the vertical view angle to a large angle range, which can cover complex structures such as high-altitude pipelines and dense shelves; dynamic layered scanning strategy quickly locates high-reflectivity areas through coarse scanning and triggers local fine scanning, which improves distance resolution while shortening overall modeling time; multi-sensor spatio-temporal fusion technology realizes centimeter-level calibration of radar, lidar and IMU, effectively eliminating motion distortion noise; DART-NeRF hybrid modeling engine embeds radar physical properties into the neural radiance field framework, accelerates training through lightweight network, and supports real-time generation of penetrating tomographic images (such as metal corrosion and hidden obstacles). In addition, the system is compatible with multiple robot platforms such as tracked robots and unmanned aerial vehicles, which can adapt to various scenes such as battlefield reconnaissance, industrial detection and security inspection, significantly improving the reliability, efficiency and engineering application potential of electromagnetic texture modeling. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are included to provide a further understanding of the application, illustrate embodiments of the application and together with the description given below, serve to explain the application. The accompanying drawings are included as part of this application and, together with the specification, serve to further explain the application. In the drawings: Figure 1 is a scene electromagnetic texture mapping method flowchart based on an autonomous mobile robot according to an embodiment of the application; Figure 2 is a Doppler equivalent cone and distance equivalent sphere intersection diagram according to an embodiment of the application; Figure 3 is a system structure diagram according to an embodiment of the application. DETAILED DESCRIPTION

[0019] The embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0020] The embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0021] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concepts of the present disclosure in a schematic manner, and only show the components related to the present disclosure in the diagrams, not the number, shape and size of the components when actually implemented. The actual implementation of each component may be a random change in type, number and proportion, and the component layout pattern may also be more complex.

[0022] Referring to Figure 1 A scene electromagnetic texture mapping method based on an autonomous mobile robot includes robot design, radar detection, and DART implementation. Robot design includes data collection by the robot through sensors, and data fusion and processing of the collected information to obtain more accurate environmental perception capabilities. Subsequently, SLAM (Simultaneous Localization and Mapping) technology is used to establish a map and perform positioning to support autonomous navigation of the robot. After obtaining environmental information, the robot performs path planning, and finally the planned path is executed by the motion control system to achieve automatic navigation and motion control.

[0023] Radar detection provides environmental perception support for the robot based on radar sensors.

[0024] DART implementation builds a world model based on the exhibition space to provide a basis for subsequent visualization. Subsequently, network architecture design is performed to support complex rendering calculations.

[0025] The entire system interacts through a cooperative mechanism. The radar sensor operating part provides environmental data for the robot, enabling the robot to navigate and control based on accurate environmental perception. At the same time, the DART system utilizes the data of the radar and the robot to achieve visualization processing, enabling users to more intuitively understand the operation of the system. Through this cooperative mechanism, the system can achieve more efficient and intelligent operation, thereby improving overall performance.

[0026] Specifically includes: S1: Robot operation and development; The robot hardware composition includes a main body and a chassis, and the chassis is equipped with two servo motors to drive the wheels and a universal wheel, achieving good maneuverability and strong load capacity to meet the stable motion requirements in complex outdoor terrain.

[0027] The core of the dynamic coordination control of motion-scan parameters is to build a radar scan parameter adaptive adjustment system driven by motion state. Through real-time coupling of the robot chassis kinematics model and the radar scanning mechanism (FMCW signal modulation parameters), the dynamic matching of speed-resolution-scan range is realized: based on the real-time wheel speed output by the encoder, the system dynamically adjusts the pulse repetition frequency (PRF) of the millimeter wave radar, effectively eliminating the Doppler shift ambiguity caused by high-speed motion; at the same time, the roll angle and pitch angle data of the IMU are fused, and the four-degree-of-freedom gimbal scan range is adjusted in real time. When the robot climbing angle reaches a certain degree, the gimbal realizes automatic tilt compensation to maintain the radar beam horizontal coverage, and in the narrow channel scene, the laser radar point cloud density distribution is fused to trigger the radar horizontal scan range dynamic change, realizing the improvement of target azimuth angle resolution; in addition, when the navigation system detects that there is a dynamic obstacle in front, the radar working mode is switched through the state machine model, the dynamic target tracking mode is preferentially started, and the SLAM mapping task is suspended, and the computing resources are tilted to the motion prediction module to meet the obstacle avoidance response delay requirement, thereby forming a multi-dimensional parameter linkage mechanism of speed, attitude and obstacle avoidance scene, realizing the deep coordination optimization of radar scan parameters and robot motion state.

[0028] Dynamic hierarchical scanning strategy This embodiment quickly identifies high attention areas (such as metal joints, hidden corners, etc.) in panoramic scanning mode, and triggers local precision scanning mode based on laser radar point cloud density and millimeter wave radar, dynamically adjusts parameters in real time through a scene complexity dynamic evaluation model (complexity index = point cloud density x reflection intensity variance / motion speed), and at the same time, avoids obstacles through multi-objective optimization decision, dynamically plans the motion path according to the terrain roughness, realizes small path tracking accuracy error, and significantly improves the path repetition rate compared with manual scanning. In practical application, this strategy improves the modeling efficiency of the shielding plant detection task in the nuclear power plant compared with manual block scanning, significantly improves the completeness of feature extraction of high-reflectivity metal components, reduces the number of missed defects, and solves the efficiency-accuracy paradox of fixed scanning mode in complex environments, providing a high autonomy and low redundancy electromagnetic texture modeling solution for industrial inspection.

[0029] The deep learning inference layer is deployed on the NVIDIA Jetson AGX Xavier platform. The network introduces dense skip connection and attention gate mechanism based on traditional U-shaped structure, and is specially optimized for electromagnetic artifact suppression task. Specifically, the double-branch convolutional layer at the front end of the model processes the amplitude phase information and Doppler features respectively, and realizes cross-modal feature selection through channel attention weighting in the post-fusion stage, and finally outputs a three-channel segmentation map containing target categories (metal / non-metal), confidence and abnormal scattering area mask.

[0030] Clock synchronization circuit In the deepening design of the space-time synchronization system, an intelligent clock architecture with multi-modal collaboration is constructed, and high-precision time reference transmission is realized in complex scenarios through deep coupling of the physical layer and the protocol layer. The core clock source adopts a dual-redundancy configuration, and the high-stability oven-controlled crystal oscillator (OCXO) tamed by GPS can achieve long-term frequency stability when the satellite signal is good, while the built-in rubidium atomic clock serves as a backup source. Dynamic switching between the two avoids the time sequence breakage problem caused by traditional hard switching.

[0031] S2: Collaborative driving of robot platform and millimeter wave radar; The core of collaborative driving is to establish a physical-level coupling between millimeter wave radar data and the kinematic model of the robot.

[0032] In the design of intelligent robot systems for dynamic and complex environments, the deep collaboration of motion compensation and radar-mechanical coupling optimization technology achieves high-precision perception and stable execution. The motion compensation system uses a deep fusion of multi-source data fusion and active control mechanism to calculate the robot's three-dimensional attitude angle (pitch, roll, yaw) and angular velocity information in real time at a high sampling rate. The built-in sensor fusion coprocessor directly outputs the temperature drift compensated attitude matrix through a hardwired interface. This data stream is transmitted to the radar control unit through a serial channel. In the coordinate transformation acceleration engine built in the FPGA, the radar beam pointing vector and the IMU attitude quaternion are subjected to tensor multiplication operation to achieve sub-millimeter level beam pointing dynamic calibration. The specially designed abnormal motion suppression strategy introduces a frequency domain analysis module: when the IMU frequency spectrum analysis detects high-frequency vibration components (corresponding to the typical bump frequency of gravel road), it automatically triggers the radar scan timing rescheduling, dynamically adjusts the linear frequency modulation pulse repetition interval (PRI), and effectively avoids the distance gate leakage caused by motion artifacts.

[0033] S3: DART experimental verification based on robot platform; DART technology uses a data-driven approach to build an environment model. Data-driven methods use real sensor scans to build an environment model. Sparse methods use constant false alarm rate (CFAR) to detect discrete reflectors in the environment. At the same time, the dense method divides the environment into an explicit voxel grid and infers the radar properties of each cell. The dense method can be further divided into coherent and incoherent aggregation.

[0034] Unlike explicitly defining an inverse imaging algorithm that recovers a scene representation from sensor readings, neural radiance fields implicitly invert a forward rendering function using stochastic gradient descent. This requires the following components: Scene model: NeRF defines a scene as an RGB color and transparency at each position and view angle; subsequent work has extended it to handle anti-aliasing, different cameras, and lighting.

[0035] Scene representation: In addition to neural networks or voxel grids, recent work has also explored spatial hash tables as well as functional decomposition for view-dependent.

[0036] Rendering function and model inversion: NeRF models each pixel as a ray and performs raytracing on the radiance field. The invertibility of this rendering function is crucial: by assuming each pixel is a ray, NeRF is “supervised” by one RGB image pixel per ray, allowing NeRF to “solve” for a few opaque points along the ray.

[0037] DART applies the NeRF technique to millimeter wave radar. By applying the NeRF technique to radar, as well as the extensive body of literature on neural radiance fields, the potential of neural implicit representations is simultaneously unlocked.

[0038] The invention chooses a radar measurement representation space—range-doppler—which overcomes the problem of poor spatial resolution for compact radars. To obtain better angular resolution, radars can instead exploit the Doppler effect: objects moving at different relative speeds with respect to the radar have different Doppler velocities, which are measured by examining the residual phase of the range-azimuth-elevation angle heat map. The key is that in a static scene, these relative speeds depend not only on the relative speed between the radar and the world but also on the relative azimuth and elevation angles between the object and the radar, each Doppler corresponding to a cone in space. Due to the fine range and Doppler resolution, Doppler greatly reduces the ambiguity of each cell in three-dimensional space, reducing it to a thin ring, as shown in Figure 2 We further simplify the integration by performing thinness arguments on the range and Doppler axes to simplify to a circle for radar rendering.

[0039] DART models with a data-driven approach, using a viewpoint-dependent neural network method to represent reflectance and transmittance.

[0040] One of the most challenging factors in radar view synthesis is modeling the radio frequency reflectance and millimeter wave material interactions. From the radar’s perspective, a point in space has two key attributes: reflectance (the proportion of energy reflected back) and transmittance (the proportion of energy that continues past). However, millimeter waves also interact differently with objects depending on the angle of incidence; for example, a metallic surface can be specularly reflective, potentially invisible from certain viewpoints. Therefore, it is necessary to model a reflectance and transmittance value for each physical point, each value dependent on the angle of incidence. We formalize this as (1.1); The model takes reflectance and transmittance as a function of the position and the angle of incidence functions and allows DART to simulate a range of radar phenomena such as partial occlusion, specular reflections, and ghost reflections.

[0041] To effectively train and learn the neural implicit mapping for radar, DART utilizes a network architecture for adaptive grid-world representation, designs a range-Doppler rendering method, and proposes key rendering optimizations. and , the mapping can be based on the given The network generates a multi-antenna range-Doppler heatmap; we call this a radar rendering. Unlike visual NeRF, DART must account for a range of physical effects beyond occlusion, including path attenuation, antenna gain patterns, and radar-specific Doppler axes.

[0042] From radar position and direction (rotation matrix) emits a single "ray" at an angle of incidence When the ray passes through space to the maximum range of the processed (range, Doppler, antenna) image, each range point Receive an amplitude of signal, which will be lost due to free space path loss. Then each point reflects a signal with an amplitude of signal, and continues to propagate with an amplitude of When the reflected signal returns to the radar, the signal is attenuated by an additional factor , and will also be affected Within the range of the processed heat map, the distance interval along the antenna Discrete sampling, for rays In the distance range and antennas Radar return amplitude yes: (1.2); in It's an antenna Point at angle (relative to the radar direction The antenna beamforming gain is specified).

[0043] For a given attitude, including radar position ,speed and direction , in each range-Doppler-antenna cell Evaluate the return value , synthesize a specific perspective, multi-antenna range-Doppler image Since the Doppler velocity is We integrate the return value along the thin ring corresponding to each cell As shown below: (1.3); Need to divide by speed and multiply by To correct the width of the discrete points as a function of range and radar speed.

[0044] Approximate this integral as Random directions The sum of , multiplied by an additional Factor to correct for Add the range-Doppler intersection circumference. This will give the result: (1.4); because The field function must be evaluated for every sample, so efficient sampling is crucial for computational efficiency. Therefore, the naive approach of treating each (range, Doppler, antenna) "pixel" as a separate sample, as is standard in NeRF, is computationally infeasible, requiring the field to be sampled (range, Doppler, antenna, range integral, Doppler integral) times to render a single image. Therefore, we actively reuse the bins by rendering all bins with the same Doppler at the same time. and The training is done using stochastic gradient descent with the Adam optimizer. field function, and use (mean absolute error) loss.

[0045] To calibrate the estimated velocity, the present invention proposes a velocity estimation algorithm that uses a simple threshold check to identify the maximum velocity detected in the range-Doppler image. Assuming the scene is static, this then indicates the velocity of the radar. By using this estimate, the parameters applied to the SLAM position estimate are adjusted to produce the final velocity estimate. In the dataset, errors or sudden relocalization can also lead to unstable velocity estimates. Therefore, as a final check, the acceleration of the final velocity estimate is calculated experimentally, and if the acceleration exceeds a certain level, these frames are excluded. Finally, the dataset containing high-frequency SLAM system failures is recollected or excluded.

[0046] Based on the above experiment for the problem of speed can not be accurately estimated, a millimeter wave radar data acquisition device based on TurtleBot platform is constructed. As a mature open robot platform, TurtleBo has rich speed estimation methods and position positioning means, which can improve the accuracy and stability of the data set under the premise of meeting the requirements of DART technology for data set.

[0047] Embodiment 2 As Figure 3 The application also proposes a scene electromagnetic texture mapping system based on autonomous mobile robot; The robot system part includes power system, inertial navigation, GPS, host computer, bus system and camera. The power system provides the energy required for the robot to run; the inertial navigation (IMU) is used to measure the acceleration and angular velocity of the robot and other motion parameters, helping it to perceive its attitude and motion state. GPS provides global positioning information, enabling the robot to determine its geographical location in space. The host computer can be used for input of control instructions for the robot, as well as data processing and analysis functions. The bus system serves as a data transmission channel, responsible for transmitting information between the internal components of the robot. The camera is used to obtain visual image data of the surrounding environment, providing visual reference for the robot to perform tasks.

[0048] The radar system part covers millimeter wave radar and laser radar, which can detect the distance, speed and other information of surrounding objects by emitting and receiving electromagnetic waves or laser, realizing environment perception. The storage system is used to store the data generated during the work of the radar and related programs, etc. The signal processor processes and analyzes the signals received by the radar to extract useful target information.

[0049] The implementation system is composed of FPGA, DSP, microprocessor and server. These components work together with the device system. FPGA has the characteristics of flexible programmability and can quickly realize parallel data processing, which is used to process some data with high real-time requirements; DSP focuses on digital signal processing and can efficiently process signals generated by radar and other devices; microprocessor as the processing core can perform various complex operations and control tasks. The server can be used for data storage, analysis and interaction with external systems and other functions. The device system and the implementation system work together to complete complex tasks.

[0050] Embodiment 3 The present invention also proposes a computer-readable storage medium, on which program instructions for a method for electromagnetic texture mapping of a scene based on an autonomous mobile robot are stored. The program instructions for electromagnetic texture mapping of a scene based on an autonomous mobile robot can be executed by one or more processors to implement the steps of the method for electromagnetic texture mapping of a scene based on an autonomous mobile robot as described above.

[0051] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A scene electromagnetic texture mapping method based on an autonomous mobile robot, characterized in that: Including robot operation and development, radar sensor operation and DART implementation; In the robot operation and development steps, the robot collects data through sensors, fuses and processes the collected information, and performs map building and positioning based on SLAM; The radar sensor detects the surrounding environment through a signal transmission and reception mechanism, and fuses and processes the acquired signals; The DART implementation steps include selecting a suitable display space, building a world model based on it, providing visualization results from different angles through perspective synthesis technology, and finally outputting them in a multi-perspective manner.

2. The scene electromagnetic texture mapping method based on an autonomous mobile robot according to claim 1, characterized in that: The robot comprises a main body and a chassis, wherein the chassis is equipped with two steering gear drive wheels and a universal wheel.

3. The scene electromagnetic texture mapping method based on an autonomous mobile robot according to claim 2, characterized in that: It also includes real-time coupling of the robot chassis kinematic model with the radar scanning mechanism.

4. The scene electromagnetic texture mapping method based on an autonomous mobile robot according to claim 1, characterized in that: It also includes a dynamic trigger mechanism for coarse scanning and fine scanning. First, it quickly identifies high-attention areas in panoramic scanning mode, and then triggers local fine scanning mode based on the lidar point cloud density and millimeter-wave radar RCS characteristics, combined with a dynamic scene complexity evaluation model to adjust parameters in real time.

5. The scene electromagnetic texture mapping method based on an autonomous mobile robot according to claim 4, characterized in that: It also includes the alignment of millimeter-wave radar point cloud data and lidar scanning frames through timestamps, and the edge computing unit performs Kalman filter fusion to establish a dynamic environment model.

6. The scene electromagnetic texture mapping method based on an autonomous mobile robot according to claim 5, characterized in that: It also includes a tightly coupled calibration scheme using a GPS-disciplined rubidium atomic clock and a federated Kalman filter, which achieves spatiotemporal alignment of radar range-Doppler heat maps, LiDAR point clouds, and IMU attitude data through hardware-level timestamp synchronization and a multi-sensor joint calibration algorithm.

7. The scene electromagnetic texture mapping method based on an autonomous mobile robot according to claim 6, characterized in that: The IMU high-frequency angular velocity data is combined with the Lyapunov stability criterion optimization to compensate for the radar beam deflection caused by the robot movement in real time.

8. The scene electromagnetic texture mapping method based on an autonomous mobile robot according to claim 1, characterized in that: The DART implementation includes a 3D reconstruction engine that introduces neural radiation field technology and a lightweight network model implemented based on PyTorch.

9. A system based on the scene electromagnetic texture mapping method based on an autonomous mobile robot according to any one of claims 1 to 8, comprising an apparatus system and an implementation system; in, The device system includes a robot system and a radar system; the implementation system is composed of FPGA, DSP, microprocessor and server; motion control is achieved based on the robot-radar collaborative control architecture; The radar system adopts a dynamic layered scanning strategy, and performs rapid panoramic scanning and local refined scanning through a coarse-fine scanning mechanism; and realizes the spatiotemporal synchronization and coordinate system alignment of the radar, lidar, and IMU through joint calibration technology.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program instructions for a scene electromagnetic texture mapping method based on an autonomous mobile robot, and the program instructions for scene electromagnetic texture mapping based on an autonomous mobile robot can be executed by one or more processors to implement the steps of the scene electromagnetic texture mapping method based on an autonomous mobile robot as described in any one of claims 1-8.

Images