High-fidelity physical environment synthetic data generation method for embodied intelligent robots

Abstract

This invention relates to the field of data analysis technology, specifically a method for generating high-fidelity physical environment synthetic data for embodied intelligent robots. The method includes: dividing a glass panel into multiple spatial sub-regions; constructing the interlayer interface relationship between the film layer and the substrate; performing mapping processing in a simulation environment to obtain a partitioned physical structure representation; continuously tracking the reflection, transmission, and boundary abrupt changes of the laser beam in the film layer; merging multiple candidate echoes into a three-dimensional point cloud representation; independently and randomly perturbing the film thickness, surface roughness, and interface state of each spatial sub-region; randomly changing the film boundary morphology to form a point cloud perturbation sample set; combining this with robot motion navigation behavior simulation; and simultaneously encapsulating the three-dimensional point cloud representation as a unified output navigation training dataset. This invention solves the problem in existing technologies where robots repeatedly misjudge obstacle contours, leading to doorway recognition failures.

Description

Technical Field

[0001] This invention belongs to the field of data analysis, and more specifically, relates to a method for generating high-fidelity physical environment synthetic data for embodied intelligent robots. Background Technology

[0002] Current synthetic data generation methods for embodied intelligent robots generally simplify glass as a single material object with uniform or overall gradient optical properties when dealing with glass component environments. They lack the ability to effectively model the spatial partitions formed on the glass panel surface due to processes such as applying explosion-proof films, frosted films, silkscreen patterns, and decorative layers with local openings. Traditional technical solutions typically only perform global random perturbation on the reflectivity and transparency of the entire glass curtain wall. When a robot faces a glass panel with transparent areas, frosted areas, silkscreen areas, and film edge boundaries simultaneously, the transmission and reflection behavior of the laser beam in different sub-regions exhibits significant differences. Existing methods cannot reproduce these local optical abrupt changes within the same component in synthetic data. In real engineering scenarios, the edges of the film layers on the glass surface may curl or delaminate due to long-term use, and the silkscreen pattern areas may exhibit density variations and uneven thickness. These non-ideal interface states directly lead to complex phenomena such as point cloud breaks, virtual contours, and local high-density dot bands in the lidar echo near the boundary. However, existing synthetic data methods neglect to depict the degradation of film edge morphology, changes in interface adhesion state, and physical differences in transition zones between different process areas. This results in the generated training samples failing to reflect the true optical reflection characteristics of glass components at film boundaries and screen-printed areas. If robots rely on such synthetic data lacking details of local non-uniform reflection for navigation strategy learning, they are prone to problems such as obstacle contour misjudgment, failure to identify passable spaces, and path planning jitter near glass door handles, screen-printed waistline areas, or frosted film boundaries when entering scenarios such as glass partitions in office building conference rooms, shopping mall windows, or glass doors in bank lobbies. This fails to meet the urgent need for the authenticity of perception and navigation data in embodied intelligent systems in high-fidelity physical environments. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to overcome the aforementioned deficiencies and propose a method for generating high-fidelity physical environment synthetic data for embodied intelligent robots.

[0004] The present invention adopts the following technical solution; The first aspect of this invention discloses a method for generating high-fidelity physical environment synthesis data for embodied intelligent robots, as follows: S1: Divide the glass panel into multiple spatial sub-regions, construct the interlayer interface relationship between the film layer and the substrate in each spatial sub-region, and perform mapping processing in the simulation environment to obtain the partitioned physical structure expression of the local non-uniform glass component. S2: Based on the physical structure representation of partitions, the reflection, transmission and boundary abrupt behavior of the laser beam in the film are continuously tracked, and the candidate echoes of the first interface, the candidate echoes of internal reflection and the candidate echoes of abrupt bifurcation are merged and converted into a three-dimensional point cloud representation containing fracture and virtual contour features. S3: Based on three-dimensional point cloud representation, the film thickness, surface roughness and interface state of each spatial sub-region are independently and randomly perturbed, while the film boundary is randomly morphologically changed to form a point cloud perturbation sample set. S4: Based on the point cloud perturbation sample set, combined with the simulation of the robot's motion navigation behavior in the vicinity of the glass component, output the corresponding obstacle contour annotation, passable space annotation and path execution record; at the same time, it is uniformly encapsulated with the 3D point cloud representation and output as a navigation training dataset.

[0005] Specifically, S1 includes: S11: Divide the glass panel into multiple spatial sub-regions with different physical properties, and calculate the non-uniformity index of each spatial sub-region to obtain the region division result set; S12: For each spatial sub-region, establish a multi-layer structure in the thickness direction, and then calculate the equivalent refractive index of each spatial sub-region; S13: Based on the region division result set and equivalent refractive index, for different material layers in the multilayer structure of each spatial sub-region, the interface of each layer is corrected by introducing a rough interface reflection factor, and the transition width between adjacent spatial sub-regions is calculated to form the interlayer interface relationship between each spatial sub-region. S14: The region division result set, the equivalent refractive index of the multi-layer structure, and the interlayer interface relationship between each spatial sub-region are uniformly encapsulated to form a structured data object containing region identifiers, multi-layer parameters, and interface physical relationships, which serves as the physical structure expression unit. Then, in the simulation environment, each type of spatial sub-region is mapped to a corresponding physical material block according to the physical structure expression unit, and the physical material blocks of all spatial sub-regions are combined according to their spatial positions to obtain the partitioned physical structure expression of the local non-uniform glass component.

[0006] Specifically, S2 includes: S21: Extract the rough interface reflection factor from the physical structure representation of the partition and combine it with the lidar emission parameters to determine the spatial sub-region hit by each laser beam. Calculate its effective reflection energy at the first interface and its effective transmission energy into the layer after passing through the first interface. Then compare the effective reflection energy with the preset energy detection threshold. The echo corresponding to the laser beam that exceeds the energy detection threshold is taken as an independent candidate echo of the first interface.

[0007] Specifically, S2 also includes: S22: For the laser beam with effective transmitted energy that passes through the first interface and enters the layer, continuous propagation path tracking is performed in the multi-layer structure. During the tracking process, the echo corresponding to the laser beam whose interface reflection energy exceeds the preset detection threshold is output as the internal reflection candidate echo, and its remaining propagation energy and propagation time are calculated. S23: First, extract the transition width and equivalent refractive index from the physical structure representation of the partition to calculate the path mutation intensity. Then, in the film layer interface region, by introducing the propagation path mutation mechanism, generate mutation bifurcation candidate echoes based on the comparison results of the path mutation intensity and the preset mutation intensity threshold. At the same time, calculate the path mutation intensity of the laser beam in the boundary transition width and the additional time delay caused by the boundary mutation. The preset mutation intensity thresholds include: a lower threshold and an upper threshold; S24: Merge the candidate echoes of the first interface, the candidate echoes of internal reflections, and the candidate echoes of abrupt bifurcation into a complete set of candidate echoes, and then convert the candidate echo set into a 3D point cloud; Simultaneously, the path abruptness intensity and additional time delay in the candidate echoes of abrupt bifurcation are converted into the received intensity of the echo point, and the point is labeled with a category, resulting in a three-dimensional point cloud representation containing features of breaks, virtual contours, and high-density point bands.

[0008] Specifically, S3 includes: S31: Based on the three-dimensional point cloud representation, reverse partitioning and mapping are performed on each spatial sub-region of the glass component to infer its physical state and calculate the randomized intensity coefficient to obtain the partitioned perturbation reference set. S32: Based on the partitioned perturbation reference set, for each spatial sub-region, the film thickness, surface roughness state and interface state are independently perturbed, and the perturbed film thickness, surface roughness and interface state factors are encapsulated according to the spatial sub-region to form a partitioned randomized parameter set.

[0009] Specifically, S3 also includes: S33: Randomize the morphology of the membrane boundary region and calculate the morphological parameters of edge warping height and degumming length; The calculated warp height and degumming length, along with the resulting membrane boundary deformation geometry and the spatial distribution of the degumming region, are collectively output as the boundary morphology perturbation set. S34: Based on the partitioned randomization parameter set and the boundary morphology perturbation set, recalculate the echo intensity of each point on the basis of the baseline 3D point cloud, and generate a set of point cloud perturbation samples for multiple simulation rounds.

[0010] Specifically, S4 includes: S41: Based on the point cloud perturbation sample set, each point cloud perturbation sample is registered with the robot navigation scene, and samples that meet the requirements are selected according to the scene registration coefficient to obtain the navigation scene registration sample set. S42: In the registered scene, drive the robot to perform navigation behavior simulation in the area near the glass component, make real-time decisions by calculating the path risk intensity, and continuously record all the robot's behavior time series data, and encapsulate the complete behavior time series data into the behavior simulation result set corresponding to the sample. S43: Based on the robot's behavior simulation result set and combined with the 3D point cloud anomaly information, calculate the passability credibility and obstacle contour salience of each region, and generate a joint annotation result set.

[0011] Specifically, S4 also includes: S44: Encapsulate the 3D point cloud samples, behavior simulation result sets, and joint annotation result sets from each simulation round, calculate the sample validity, filter and sort them to generate the final navigation training dataset.

[0012] A second aspect of the present invention discloses a terminal, including a processor and a storage medium; characterized in that: The storage medium is used to store instructions;

[0013] The processor is configured to operate according to the instructions to execute the steps of the high-fidelity physical environment synthesis data generation method for embodied intelligent robots as described in the first aspect.

[0014] The third aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the steps of the method for generating high-fidelity physical environment synthetic data for embodied intelligent robots as described in the first aspect.

[0015] The beneficial effects of the present invention are as follows: Compared with the prior art, the present invention has the following advantages: 1. This technical solution divides the glass panel into multiple spatial sub-regions with independent physical structures and establishes a multi-layer structure model including the substrate, film layer, screen printing layer and debonding gap. It achieves high-fidelity characterization of local optical properties under different process states such as transparent area, frosted area, screen printing area and film boundary area. This allows the transmission and reflection behavior of the laser beam when passing through different areas to be simulated differently based on the real interlayer interface relationship and equivalent optical parameters. It solves the technical problem that the optical abrupt features inside the same component are difficult to reproduce in the synthetic data. 2. This technical solution explicitly tracks the continuous propagation path of the laser beam in the multilayer structure and combines independent random perturbations of key parameters such as film thickness, surface roughness, interface state factor, and film boundary warping height and delamination length. At the physical level, it reproduces complex anomalies such as point cloud fracture, virtual contours and local high-density dot bands caused by film warping, local delamination and changes in screen printing pattern density in real engineering scenarios. This improves the fidelity of the synthetic point cloud data in expressing the non-target interface state and boundary degradation morphology of glass components. 3. This technical solution constructs a training dataset that can realistically reflect the interaction between perceived noise and navigation decisions by jointly outputting obstacle contour annotation, passable space annotation, and path execution record. This enables the robot to effectively reduce the obstacle contour misjudgment rate and route jitter frequency when facing typical scenarios such as screen-printed waistlines, membrane edge junctions, and local virtual contours, thereby improving the robustness and safety of autonomous navigation in complex building environments. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the high-fidelity physical environment synthesis data generation method for embodied intelligent robots according to the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0019] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0020] Example 1: like Figure 1As shown in the figure, the method for generating high-fidelity physical environment synthesis data for embodied intelligent robots according to an embodiment of the present invention includes the following: S1: Divide the glass panel into multiple spatial sub-regions, construct the interlayer interface relationship between the film layer and the substrate in each spatial sub-region, form a physical structure expression unit, and perform mapping processing in the simulation environment to obtain the partitioned physical structure expression of the local non-uniform glass component. S1 includes: S11: Divide the glass panel into multiple spatial sub-regions with different physical properties, and calculate the non-uniformity index of each spatial sub-region to obtain the region division result set; It should be noted that in this embodiment, each spatial sub-region corresponds to a dominant process state; The process states include: transparent area, frosted area, screen printing area, film boundary area, and boundary transition area; It should be noted that, in this embodiment, the location where there is localized warping and delamination is defined as the boundary transition zone; For example, in this embodiment, a strategy for obtaining the non-uniformity index is provided, specifically as follows: ; In the formula, The index represents the non-uniformity of the r-th spatial sub-region, used to measure the degree of non-uniformity between a certain region and the glass reference region; This represents the surface reflectivity of the r-th spatial sub-region, with a value ranging from 0.04 to 0.92. This represents the reflectivity of the glass reference area, with a value ranging from 0.04 to 0.12. This represents the transmittance of the r-th spatial sub-region, with a value ranging from 0.10 to 0.95. This represents the transmittance of the glass reference region, with a value ranging from 0.75 to 0.92. This represents the surface roughness of the r-th spatial sub-region, with a value ranging from 0.05 micrometers to 20 micrometers; This represents the surface roughness of the glass reference area, with a value ranging from 0.01 micrometers to 0.20 micrometers. This represents the film thickness of the r-th spatial sub-region, with a value ranging from 0 to 0.50 mm. This indicates the thickness of the glass substrate, and its value ranges from 4 mm to 19 mm. It should be noted that, in this embodiment, the above formula is intended to unify the differences in reflection characteristics, transmission characteristics, surface roughness, and relative differences in film thickness into a single comprehensive index. The region division result set includes: the number, area, boundary length, process category, and adjacency relationship of each spatial sub-region; S12: For each spatial sub-region, establish a multi-layer structure in the thickness direction, and then calculate the equivalent refractive index of each spatial sub-region; For example, in this embodiment, the establishment of the multilayer structure specifically includes: for the transparent area, the multilayer structure includes a glass substrate layer and a film layer; for the screen printing area, the multilayer structure includes a glass substrate layer, a film layer and an ink layer; it should be noted that for the boundary transition area, a local air gap layer is added. For example, in this embodiment, a strategy for obtaining the equivalent refractive index is provided, specifically as follows: ; In the formula, denoted by the equivalent refractive index of the r-th spatial sub-region, reflecting the degree to which the average speed of light is slowed down when the beam passes perpendicularly through the entire multilayer structure; Indicates the thickness of the glass substrate; Let represent the refractive index of the film in the r-th spatial sub-region, with a value ranging from 1.20 to 1.80; This represents the refractive index of the glass substrate, with a value ranging from 1.45 to 1.60. This represents the film thickness of the r-th spatial sub-region; The value represents the refractive index of the screen printing layer in the r-th spatial sub-region, and its value ranges from 1.40 to 2.10. This represents the thickness of the silkscreen layer in the r-th region, with a value ranging from 0 to 0.20 mm. This represents the refractive index of air, and its value ranges from 1.0002 to 1.0004. This represents the thickness of the degumming gap in the r-th spatial sub-region, with a value ranging from 0 to 1.50 mm. It should also be noted that, in this embodiment, this step aims to further focus the regional differences on the differences in the layer structure in the thickness direction, so as to better distinguish the real differences in the propagation path between the transparent area, the screen printing area and the film boundary area, rather than staying at the level of surface color or label.

[0021] S13: Based on the region division result set and equivalent refractive index, for different material layers in the multilayer structure of each spatial sub-region, the interface of each layer is corrected by introducing a rough interface reflection factor, and the transition width between adjacent spatial sub-regions is calculated to form the interlayer interface relationship between each spatial sub-region. The different material layers include: a glass substrate layer, a film layer, an ink layer, and a local air gap layer; The interlayer interface relationship includes: the roughness interface reflection factor of the spatial sub-region, the layer interface position, and the transition width of the boundary between adjacent spatial sub-regions; It should be noted that in this embodiment, in order to reflect the local scattering enhancement phenomenon caused by film edge lifting, screen printing particles and frosted surface, the rough interface is corrected by calculating the rough interface reflectance factor. For example, in this embodiment, a strategy for obtaining the reflectivity of a rough interface is provided, specifically as follows: ; In the formula, Indicates the first Roughness interface reflectance factor of each spatial sub-region; Indicates the first The equivalent extinction coefficient of each spatial sub-region is obtained according to the material handbook of the film material, ink or frosted layer, and its value ranges from 0.1 to 50. Indicates the first The surface equivalent roughness of each spatial sub-region; This represents the laser incident angle, calculated based on the angle between the current beam emission direction and the local normal, with a value ranging from 0 to 1.40. This indicates the laser wavelength, with a value ranging from 0.85 micrometers to 1.55 micrometers.

[0022] It should be noted that the above formula first uses the refractive index difference and the equivalent extinction coefficient to calculate the reflectivity of the target interface, and then uses roughness attenuation to weaken the reflection of the target interface, so that the calculation result is closer to the engineering surface. A strategy for obtaining the transition width is also provided, specifically: ; In the formula, This represents the transition width between the r-th spatial sub-region and its adjacent spatial sub-region s; This represents the initial boundary width, with a value ranging from 0.1 to 2.0 millimeters. This is the degumming thickness magnification factor, with a value ranging from 0.5 to 3.0; This is the roughness amplification factor, and its value ranges from 0.01 to 0.20. This is the refractive index difference magnification factor, which ranges from 0.1 to 2.0. Represents the equivalent refractive index of the r-th spatial sub-region; This represents the equivalent refractive index of an adjacent spatial sub-region s; It should be noted that the above formula is used to transform the boundary from a non-line boundary into a perturbation band with width, which makes it easier to treat the membrane boundary as an independent physical region in the subsequent simulation. It should also be noted that in this embodiment, the point cloud fractures, virtual contours and local high-density point bands are caused not only by the different materials within the region, but also by the non-ideal interfaces and transition zones between regions. The interlayer interface relationship established in this step allows subsequent simulations to not only obtain different blocks, but also boundary anomalies.

[0023] S14: The region division result set, the equivalent refractive index of the multi-layer structure, and the interlayer interface relationship between each spatial sub-region are uniformly encapsulated to form a structured data object containing region identifiers, multi-layer parameters, and interface physical relationships, which serves as the physical structure expression unit. Then, in the simulation environment, each type of spatial sub-region is mapped to a corresponding physical material block according to the physical structure expression unit, and the physical material blocks of all spatial sub-regions are combined according to their spatial positions to obtain the partitioned physical structure expression of the local non-uniform glass component.

[0024] Understandably, a physical structure representation unit is a structured encapsulation result of a single spatial sub-region, including a set of region division results, multi-layer structural parameters, and inter-layer interface relationships, describing only local physical properties; while a partitioned physical structure representation is the result of mapping multiple physical structure representation units to corresponding physical material blocks according to their spatial positions and combining them as a whole in the simulation environment, used to describe the spatial partitioned physical structure of the entire glass.

[0025] It should be noted that, in this embodiment, the physical material block includes: a transparent area, a frosted area, a screen-printed area, and a film boundary area.

[0026] The mapping is intended to allow physical material blocks to exist in a partitioned manner on the same piece of glass; S2: Based on the physical structure representation of partitions, the reflection, transmission and boundary abrupt behavior of the laser beam in the film are continuously tracked, and the candidate echoes of the first interface, the candidate echoes of internal reflection and the candidate echoes of abrupt bifurcation are merged and converted into a three-dimensional point cloud representation containing fracture and virtual contour features. S2 includes: S21: Extract the rough interface reflection factor from the physical structure representation of the partition and combine it with the lidar emission parameters to determine the spatial sub-region hit by each laser beam, and calculate its effective reflection energy at the first interface and its effective transmission energy into the layer after passing through the first interface. Then, the effective reflected energy is compared with the preset energy detection threshold, and the echo corresponding to the laser beam that exceeds the energy detection threshold is taken as an independent candidate echo of the first interface. The laser radar emission parameters include: emission height, scanning angle range, angular resolution, pulse width, emission power, wavelength, and sampling period; It should be noted that, in this embodiment, when the lidar is a mechanically rotating three-dimensional lidar, the angular resolution ranges from 0.05 degrees to 0.40 degrees; when the lidar is a solid-state scanning lidar, a discrete beam sequence is constructed according to the geometric distribution of the emission units given by the device. It should be noted that, in this embodiment, since the same piece of glass has been divided into a transparent area, a frosted area, a screen-printed area, and a film boundary area, even if the angle difference between two adjacent beams is very small, they may hit different areas, thus exhibiting different energy distribution results on the first interface. The energy distribution results include: effective reflected energy and effective transmitted energy at the first interface; For example, in this embodiment, discrete laser beams are emitted in the lidar coordinate system according to a preset scanning angle range; each laser beam i is intersected with the first interface of the partition physical structure expression to determine the spatial sub-region where its first hit point is located; It should be noted that, in this embodiment, the intersection of the first interface can be determined by testing the intersection points of a ray with a partitioned surface or a ray with a partitioned triangular facet. For example, in this embodiment, an effective strategy for obtaining reflected energy is provided, specifically as follows: ; In the formula, This represents the effective reflected energy of the i-th laser beam at the first interface; This represents the emitted energy of the i-th laser beam, with a value ranging from 0.1 to 10 microjoules; This represents the roughness reflection factor of the spatial sub-region hit by the i-th laser beam; This represents the equivalent attenuation coefficient of the hit area, with a value ranging from 0.001 to per millimeter; This represents the propagation distance of the i-th laser beam from the emission point to the first interface; It should be noted that the above formula takes the transmitted energy as a benchmark, first allocates the theoretical reflectable energy through the first interface reflectivity, and then multiplies it by the two-way exponential attenuation caused by propagation along the path, so as to obtain the effective reflected energy of the first interface that can actually return to the receiver. The comparison between the effective reflected energy and the preset energy detection threshold is specifically as follows: When the effective reflected energy exceeds the energy detection threshold, the echo corresponding to the laser beam is taken as an independent candidate echo of the first interface, and its time, spatial coordinates and intensity are recorded, and it does not propagate into the multi-layer structure. When the effective reflected energy is lower than the energy detection threshold, it is determined that it cannot be effectively detected by the receiver, and no candidate echo is generated. The preset energy detection threshold ranges from 0.5% to 5% of the emitted energy. An effective strategy for obtaining transmission energy is provided, specifically as follows: ; In the formula, Indicates the first The effective transmission energy of a laser beam after passing through the first interface and entering the layer; It should be noted that the above formula treats the energy not reflected by the first interface as the energy that can propagate into the layer, and then obtains the effective transmission energy for subsequent interlayer propagation based on the one-way attenuation.

[0027] S2 also includes: S22: For the laser beam with effective transmitted energy that passes through the first interface and enters the layer, continuous propagation path tracking is performed in the multi-layer structure. During the tracking process, the echo corresponding to the laser beam whose interface reflection energy exceeds the preset detection threshold is output as the internal reflection candidate echo, and its remaining propagation energy and propagation time are calculated.

[0028] For example, in this embodiment, the continuous propagation path tracking in the multilayer structure adopts a depth-first light path extension, which is as follows: the first interface enters the film layer or substrate layer, passes through the current layer and intersects with the interface of the next layer. Then, the reflected energy and transmitted energy are recalculated according to the new interface conditions. If the reflected energy is greater than the preset detection threshold, an internal reflection candidate echo is generated. Tracking stops when the remaining propagation energy is lower than the tracking lower limit or the maximum tracking order is reached. The propagation time, spatial coordinates and echo intensity of the internal reflection candidate echo are recorded. The maximum tracking order ranges from 3 to 8. If there are screen printing layers or debonded air gaps in a local area, the laser beam will undergo additional absorption by the ink layer or weak constraint propagation through the air gap, thus forming a propagation chain different from that of ordinary transparent glass. For example, in this embodiment, the remaining propagation energy of the i-th laser beam after passing through the m-th layer is provided. The acquisition strategy is as follows: ; In the formula, ; ; In the formula, This represents the transmission factor of the i-th laser beam at the j-th layer interface, and its value ranges from 0.01 to 0.99. This represents the volume attenuation coefficient of the i-th laser beam in the j-th layer, with a value ranging from 0.001 to 5.000 per millimeter. This represents the propagation length of the i-th laser beam within the j-th layer; Let represent the reflection factor of the i-th laser beam at the q-th interface.

[0029] It should be noted that the above formula integrates the intralayer transmission chain and the multiple interface reflection chain into the same energy expression. The former describes the remaining propagation energy after penetrating to a deeper layer, and the latter describes the contribution of the echo energy that may return in each layer, thus providing a unified energy basis for subsequent judgment of real points, virtual contour points and high-density point bands. For example, in this embodiment, a strategy for obtaining the propagation time of candidate echoes is provided, specifically as follows: ; In the formula, This represents the propagation time of a candidate echo from the i-th laser beam; This indicates the total number of layers traversed corresponding to the candidate echo; c represents the speed of light in vacuum, and its value ranges from... per second; Represents the refractive index of the i-th laser beam in the m-th layer; This represents the propagation length of the i-th laser beam in the m-th layer.

[0030] It should be noted that the above formula sums the actual optical path lengths of each layer and then divides by the speed of light to obtain the echo arrival time. Because the layer structure of different regions is different, the echo time will vary discontinuously even if the spatial locations are very close. The set of continuous propagation paths within the region includes the propagation length of each laser beam in the glass substrate layer, film layer, screen printing layer and air gap layer, the remaining propagation energy between layers, the exit position of each interface and the propagation time of the candidate echo. It should also be noted that, in this embodiment, this step aims to extend the glass impact from a single event to a continuous propagation process within the layer; true local non-uniform reflection cannot be represented by simply changing the reflectivity of the surface, but rather by the fact that the attenuation, reflection, delay, and return position of the beam will differ after entering different layers, and these differences must be explicitly preserved through continuous path tracing. S23: First, extract the transition width and equivalent refractive index from the physical structure representation of the partition to calculate the path mutation intensity. Then, in the membrane layer interface region, by introducing the propagation path mutation mechanism, generate mutation bifurcation candidate echoes based on the comparison results of the path mutation intensity and the preset mutation intensity threshold, and calculate the additional time delay caused by the boundary mutation. The preset mutation intensity thresholds include: a lower threshold and an upper threshold; It should be noted that, in this embodiment, the path abrupt change does not refer to a complete loss of path control, but rather to a jump in the echo direction, energy, and arrival time of the laser beam within a very short spatial range due to the combined effects of interface roughness, refractive index difference, changes in adhesion state, and abrupt changes in interlayer thickness. For example, in this embodiment, a strategy for obtaining the intensity of path mutation is provided, specifically as follows: ; In the formula, This represents the intensity of the path abrupt change in the boundary region for the i-th laser beam; The larger the value, the greater the likelihood that the beam will be deflected, bifurcated, or time-jumped at the membrane boundary. This represents the equivalent refractive index of the region before the boundary; Indicates the equivalent refractive index of the region after the boundary; Indicates the thickness of the local air gap at the interface; This represents the transition width of the i-th laser beam at the boundary; This represents the relevant propagation length of the beam in the boundary zone, calculated based on the boundary crossing distance; Indicates the surface roughness of the region after the boundary; Indicates the surface roughness of the region before the boundary; It should be noted that the above formula integrates the abrupt change in refractive index, local debonding thickness, transition width, relevant propagation time, and the roughness ratio before and after the interface to construct an interface path disturbance intensity index. Will The criteria for determining a path fork switch are as follows: Then the path mutation intensity is compared with the preset mutation intensity threshold; when Less than the lower threshold When, it should be handled as a routine regional transmission; when When it is in the middle range, an additional weak echo branch is formed, which serves as a candidate echo for abrupt bifurcation; when Greater than the upper limit threshold At that time, multiple boundary bifurcation echoes are formed, which serve as multiple abrupt bifurcation candidate echoes, and their echo time, additional time delay, spatial coordinates and echo intensity are recorded. It should be noted that the lower threshold... and upper limit threshold Calibration is performed based on the measured echo at the glass door boundary; lower threshold. The value range is 0.05-0.20, with an upper limit threshold. The value range is 0.30-0.80; For example, in this embodiment, an acquisition strategy with an additional time delay is provided, specifically as follows: ; In the formula, This represents the additional time delay introduced by the abrupt change in the boundary path of the i-th laser beam; This represents the converted value of the speed of light on a millimeter scale.

[0031] It should be noted that the above formula reflects that the thicker the air gap, the greater the refractive index difference, and the narrower the boundary band, the more obvious the additional time delay; when the boundary band is wider, the abrupt change is smoothed out, and the time jump is weakened. It should also be noted that, in this embodiment, local breaks, virtual contours and high-density dot bands are not naturally and uniformly generated within the ordinary transparent area, but are mostly concentrated at the film edge, the boundary of the screen-printed waistline, the boundary of the frosted film, etc.; therefore, only by analyzing the abrupt changes in the propagation path separately can the simulation results show boundary anomalies.

[0032] S24: Merge the candidate echoes of the first interface, the candidate echoes of internal reflections, and the candidate echoes of abrupt bifurcation into a complete set of candidate echoes, and then convert the candidate echo set into a 3D point cloud; Understandably, candidate echoes from the first interface originate from the effective reflected echoes generated when the laser beam first hits the surface of the spatial sub-region; candidate echoes from the internal reflection originate from the echoes formed by multiple reflections at the interlayer interfaces after the beam enters the multilayer structure; and candidate echoes from abrupt bifurcation originate from the bifurcation echoes generated by abrupt changes in the path at the membrane boundary, reflecting differences in direction, energy, or time jumps.

[0033] Simultaneously, the path abruptness intensity and additional time delay in the candidate echoes of abrupt bifurcation are converted into the received intensity of the echo point, and the point is labeled with a category, resulting in a three-dimensional point cloud representation containing features of breaks, virtual contours, and high-density point bands.

[0034] It should be noted that, in this embodiment, the propagation time of each candidate echo is first used as the basis for the determination. The distance is calculated, and the spatial coordinates are restored by combining the transmission angle and the receiving direction to form a three-dimensional point. The path change intensity and the additional time delay are then converted into the received intensity of the echo point. Finally, each three-dimensional point is labeled with a category to form a three-dimensional point cloud representation. The marker categories include: main echo, boundary bifurcation echo, and weak ghost point; The weak ghosting point is specifically defined as: branches with echo energy below the threshold but abnormal delay can be selectively retained as weak ghosting points to enhance the simulation results' ability to express the anomalies of real glass boundaries. The three-dimensional point cloud representation includes: the spatial coordinates of the echo points, the received intensity, the time, the marker category, and local anomaly distribution information; Exemplarily, in this embodiment, a first The strategy for obtaining the received intensity of the laser beam at the corresponding echo point is as follows: ; In the formula, This represents the received intensity of the echo point corresponding to the i-th laser beam; This represents the receiver gain coefficient, which ranges from 1 to 500. This indicates the effective echo energy of the echo branch; This represents the equivalent width of the transmitted pulse, and its value ranges from 1 nanosecond to 20 nanoseconds. This indicates the propagation time of the echo; This represents the additional delay suppression coefficient, and its value range is... per second; This represents the additional time delay introduced by the abrupt change in the boundary path of the i-th laser beam; This represents the intensity of the path abrupt change in the boundary region for the i-th laser beam; It should be noted that the above formula shows that the echo intensity is determined by the effective energy and is also affected by propagation time attenuation and abrupt change delay suppression; however, if the path abrupt change is strong, it will tend to form echo enhancement or dense distribution in the local boundary region. Therefore, a term related to the path abrupt change intensity is used. The relevant enhancement terms preserve the high-density point band features at the boundary.

[0035] S3: Based on three-dimensional point cloud representation, the film thickness, surface roughness and interface state of each spatial sub-region are independently and randomly perturbed, while the film boundary is randomly morphologically changed to form a point cloud perturbation sample set. S3 includes: S31: Based on the three-dimensional point cloud representation, reverse partitioning and mapping are performed on each spatial sub-region of the glass component to infer its physical state and calculate the randomized intensity coefficient to obtain the partitioned perturbation reference set. First, establish the reference film thickness, reference surface roughness, and reference interface state factor for each spatial sub-region; the reference film thickness... The value range is 0.02-0.5 mm; the reference surface roughness The value range is 0.01-20 micrometers; the reference interface state factor The value range is 0.1-1.0; Then calculate the randomization intensity coefficient for each spatial sub-region; For example, in this embodiment, a strategy for obtaining the randomization intensity coefficient is provided, specifically as follows: ; In the formula, This represents the randomization intensity coefficient of the r-th spatial sub-region; The length density of the high-density point band within a unit area of ​​the r-th spatial sub-region; This represents the baseline value for the length density of high-density spot zones, with a value range of 0.05-0.50. This represents the proportion of virtual contour points within the r-th spatial sub-region. In this embodiment, It is obtained based on statistics of echo marker categories; This represents the baseline value for the proportion of the virtual outline, and its value ranges from 0.01 to 0.20. This represents the average distance of the r-th spatial sub-region from the nearest membrane boundary centerline; This indicates the attenuation scale of the boundary effect, and its value ranges from 5 to 50 millimeters; This represents the weighting index of high-density points, with a value range of 0.5-2.0; This represents the virtual contour weight index, with a value range of 0.5-2.0; It should be noted that, in this embodiment, the randomization intensity coefficients of each spatial sub-region are... After normalization, its value ranges from 0.1 to 3.0; If a region is stable in a 3D point cloud, only a weak perturbation is applied; if a region shows obvious anomalies, the amplitude of subsequent perturbations is increased, but should not be increased indefinitely to avoid deviating from the feasible engineering physics range. It should be noted that the above formula converts the abnormal echo phenomena that have already appeared into the weighting basis for the random perturbation of subsequent parameters; the more obvious the high-density dot band, the more virtual contours, and the closer the region is to the membrane boundary, the stronger the randomization should be applied to that region in the next round. Finally, the reference film thickness, reference surface roughness, reference interface state factor, and corresponding randomization intensity coefficient of each spatial sub-region are synthesized as the output of the partitioned perturbation reference set.

[0036] S32: Based on the partitioned perturbation reference set, for each spatial sub-region, the film thickness, surface roughness state and interface state are independently perturbed, and the perturbed film thickness, surface roughness and interface state factors are encapsulated according to the spatial sub-region to form a partitioned randomized parameter set.

[0037] It should be noted that in this embodiment, the film thickness, surface wear, and interlayer adhesion state do not necessarily change synchronously during the process. Specifically, the thickness of the glass film is basically stable, but the roughness of the edge may increase due to the erosion of the cleaning agent; the thickness of the screen printing area does not change much, but local delamination will change the interface adhesion state. Therefore, each parameter must be perturbed independently, and the same random variable cannot be scaled as a whole. Exemplarily, in this embodiment, a first The strategy for obtaining the film thickness after perturbation of each spatial sub-region is as follows: ; In the formula, Indicates the first The thickness of the film layer after perturbation of each spatial sub-region; This represents the first-order thickness perturbation coefficient, with a value ranging from 0.02 to 0.30. This represents the second-order thickness perturbation coefficient, with a value range of 0.01-0.20; Indicates the first The arc length from the current sampling position to the starting boundary of the region within each spatial sub-region; Indicates the first The characteristic length of each spatial sub-region along the main extension direction; Indicates the first The local attachment energy attenuation index of each spatial sub-region is obtained from the interface state assessment. This represents the baseline value for adhesion energy attenuation, with a range of 0.2-2.0. It should be noted that the above formula uses a sine term to express the continuous fluctuations within the region and an exponential term to express the local thickness amplification caused by local adhesion weakening, so that the film thickness has both spatial continuous variation and local abrupt enhancement. Provide a first The strategy for obtaining the surface roughness of each spatial sub-region after perturbation is as follows: ; In the formula, This represents the surface roughness of the r-th spatial sub-region after perturbation; This represents the sensitivity coefficient of surface roughness to local wear, and its value ranges from 0.05 to 2.0 micrometers. This represents the surface wear level of the r-th sub-region, with a value ranging from 0 to 10; This represents the baseline value for the wear level, and its value ranges from 1 to 5. This represents the roughness amplification factor due to changes in film thickness; This represents the thickness normalization constant, with a value range of 0.05 to 0.50 mm; It should be noted that the above formula maps surface wear and local undulations of the film layer together as roughness growth; because the more uneven the film layer, the stronger the surface micro-undulations are often, roughness cannot be completely regarded as independent of thickness. It also provides a first The strategy for obtaining the interface state factors after perturbation of each spatial sub-region is as follows: ; In the formula, This represents the interface state factor after the perturbation of the r-th sub-region; This represents the coefficient of influence of randomization intensity on interface degradation, with a value range of 0.05-1.50. This represents the roughness attenuation coefficient of the interface state, and its value ranges from 0.01 to 0.50. This represents the normalized roughness reference value, which ranges from 0.1 to 5.0 micrometers. This represents the local compensation coefficient, with a value ranging from 0.01 to 0.30. Indicators representing the integrity of localized adhesion; This represents the attachment integrity scale parameter, with a value range of 0.2-1.5; For example, in this embodiment, a partitioned independent sampling plus constraint truncation method is adopted: the parameters of each spatial sub-region are first calculated according to the above formula to calculate the candidate perturbation value, and then it is checked whether it falls within the engineering feasible range; if it exceeds the range, it is folded back to the vicinity of the boundary value. It should be noted that the above formula reflects that the interface state will deteriorate due to increased disturbance and roughness, and will also be locally compensated in some areas with relatively intact adhesion. Therefore, fractional attenuation and exponential compensation are used together to describe this in this embodiment.

[0038] S3 also includes: S33: Randomize the morphology of the membrane boundary region and calculate the morphological parameters of edge warping height and degumming length; The calculated warp height and degumming length, along with the resulting membrane boundary deformation geometry and the spatial distribution of the degumming region, are collectively output as the boundary morphology perturbation set. It should be noted that the geometric morphology of membrane boundary deformation includes the warping shape change of the membrane boundary; the spatial distribution of degummed areas includes the discontinuous position of the membrane boundary and the range of local openings. It should be noted that in this embodiment, after completing the parameter perturbation in the spatial sub-region, the membrane boundary region is also the part most prone to generating abnormal echoes, so the membrane boundary itself is morphologically randomized. For example, in this embodiment, a strategy for obtaining the warp height is provided, specifically as follows: ; In the formula, This represents the warp height of the k-th membrane boundary; This indicates the reference warp height of the boundary, with a value ranging from 0 to 0.30 millimeters. This represents the strong disturbance uplift coefficient, with a value ranging from 0.01 to 1.50 mm. This represents the randomization intensity coefficient corresponding to the boundary neighborhood; This indicates the service life of the boundary membrane material, with a value ranging from 0 to 120 months. This represents the time decay baseline value, which ranges from 6 to 36 months. The boundary period fluctuation coefficient ranges from 0.01 to 0.80 mm. Indicates the position coordinates along the boundary direction; This represents the boundary principal fluctuation period, with a value ranging from 10 to 300 millimeters. This indicates the spatial attenuation scale, with the unit being millimeters, and its value ranges from 20 to 500 millimeters. It should be noted that the above formula reflects that the boundary warp is not a constant value, but is formed by the superposition of the overall rise and local wave undulation caused by long-term use, and the undulation has periodicity and decay in space. For example, in this embodiment, a strategy for obtaining the degumming length is provided, specifically as follows: ; In the formula, Indicates the length of degumming after disturbance; This indicates the baseline degumming length, with a value ranging from 0 to 100 mm. This represents the effect of randomization intensity on degumming expansion coefficient, with a value ranging from 0.05 to 1.20. This represents the magnification factor of the warping height to the degumming length, and its value ranges from 0.10 to 2.00. This represents the normalized height of the warp edge, with a value ranging from 0.05 to 0.50 millimeters. This represents the edge crack growth factor, with a value ranging from 0.1 to 10 mm. Indicates the number of times local cracks occur at the boundary; It should be noted that the above formula reflects that the expansion of the debonding length is driven by both the overall boundary disturbance and local edge warping, and will extend further as the number of cracks increases; It should be noted that in this embodiment, a boundary reference curve is first generated, and then the curve is determined based on the warp height. and degumming length The geometric structure around the boundary is locally rewritten, turning the originally straight membrane boundary into a non-target boundary with warping, discontinuity and local openings. This is then fed back into the 3D point cloud reprojection to form a high-density point band and local fracture features at the boundary. It should be noted that, in this embodiment, this step further advances boundary anomalies from parameter-level randomization to morphological-level randomization; S34: Based on the partitioned randomization parameter set and the boundary morphology perturbation set, recalculate the echo intensity of each point on the basis of the baseline 3D point cloud, and generate a set of point cloud perturbation samples for multiple simulation rounds.

[0039] Understandably, the partitioned randomization parameter set describes the perturbation results of parameters such as film thickness, surface roughness, and interface state in each spatial sub-region; the boundary morphology perturbation set describes the film boundary warping height, debonding length, and their geometric morphological changes; and the behavior simulation result set describes the robot's pose, speed, and path execution process in the registration scenario, etc. These three correspond to the parameter layer, morphology layer, and behavior layer, respectively. It should be noted that in this embodiment, the partitioned randomization parameter set and the boundary morphology perturbation set are remapped onto the 3D point cloud representation. By recalculating the echo intensity, each round of samples presents a different local non-uniform combination morphology, and then new point cloud samples are generated under multiple simulation rounds. For example, in this embodiment, a strategy for obtaining echo intensity is provided, specifically as follows: ; In the formula, This represents the echo intensity after the i-th point is disturbed in the j-th simulation. This indicates the reference echo intensity before the disturbance at that point; This represents the influence coefficient of film thickness, with a value range of 0.05-1.00; This represents the roughness influence coefficient, with a value range of 0.01-0.80; This represents the warping suppression coefficient, which ranges from 0.1 to 10 per millimeter. Indicates the normalized height of the warp edge; This represents the degumming reinforcement coefficient, with a value ranging from 0.01 to 0.50. This represents the normalized baseline value for degumming length, with a range of 5-100 mm. It should be noted that the above formula shows that changes in film thickness and roughness will adjust the echo intensity baseline, edge warping will cause local echo attenuation, and debonding expansion may trigger additional strong or dense echoes near specific boundaries, thus forming a new point cloud intensity distribution. Finally, the simulation samples from each round are tagged and recorded synchronously, including region number, film type, boundary anomaly type, edge warping level, debonding level, and virtual contour distribution level, forming a point cloud perturbation sample set output; S4: Based on the point cloud perturbation sample set, combined with the simulation of the robot's motion navigation behavior in the vicinity of the glass component, output the corresponding obstacle contour annotation, passable space annotation and path execution record; at the same time, it is uniformly encapsulated with the 3D point cloud representation and output as a navigation training dataset.

[0040] S4 includes: S41: Based on the point cloud perturbation sample set, each point cloud perturbation sample is registered with the robot navigation scene, and samples that meet the requirements are selected according to the scene registration coefficient to obtain the navigation scene registration sample set. It should be noted that, in this embodiment, the registration specifically involves: first, coarse registration is performed according to the installation boundary of the glass component, and then fine registration is performed according to local features such as the door handle area, the silkscreen waistline area, and the film boundary area; For example, in this embodiment, a strategy for obtaining scene registration coefficients is provided, specifically as follows: ; In the formula, Represents the scene registration coefficient for the j-th sample; This represents the average deviation between the centerline of the glass component in this sample and the centerline of the target installation in the scene. This represents the normalized scale of positional deviation, with a value ranging from 20 to 200 millimeters. This represents the percentage of overlap between the boundary anomaly region and the target doorway / partition boundary in the j-th sample. This represents the baseline value for the overlap ratio, and its value ranges from 0.05 to 0.50. This represents the percentage of the passable surface covered by invalid point clouds in the j-th sample; α represents the invalid coverage baseline value, with a value range of 0.02-0.20; β represents the boundary overlap enhancement index, with a value range of 0.5-2.0; β represents the invalid coverage suppression index, with a value range of 0.5-2.0. It should be noted that the above formula reflects that the closer the location, the more the abnormal boundary coincides with the real glass boundary, and the fewer invalid point clouds there are, the more suitable the sample is as a high-fidelity input for the current navigation scenario. when Below the preset registration lower limit threshold If the sample is not found, then discard it; when Reaching the preset registration lower threshold If so, then retain the sample; It should be noted that, in this embodiment, the registration lower limit threshold is... The value range is 0.35-0.70; All samples that reach the lower limit of registration are combined as the navigation scene registration sample set for output. S42: In the registered scene, drive the robot to perform navigation behavior simulation in the area near the glass component, make real-time decisions by calculating the path risk intensity, and continuously record all the robot's behavior time series data, and encapsulate the complete behavior time series data into the behavior simulation result set corresponding to the sample. The behaviors include at least: walking along the edge, passing through glass doorways, walking around the boundaries of shop windows, turning in front of glass partitions in conference rooms, and exploring obstacle avoidance in local virtual outline areas; It should be noted that, in this embodiment, the path update can adopt a discrete time-step progression method; the sampling period ranges from 20 milliseconds to 200 milliseconds; the linear velocity ranges from 0.1 meters per second to 1.5 meters per second; and the angular velocity ranges from 0.1 radians per second to 1.5 radians per second. The behavioral time-series data includes: pose, velocity, decision commands, and traversal results; The behavioral simulation result set includes: the robot's pose sequence, velocity sequence, turning sequence, obstacle proximity response events, and local avoidance trajectory in each registration sample; For example, in this embodiment, a strategy for obtaining path risk intensity is provided, specifically as follows: ; In the formula, This represents the risk intensity of the path corresponding to the j-th registered sample; n_j represents the total number of discrete sampling points on this path; This represents the proportion of high-density points in the neighborhood of the i-th sampling point; This represents the proportion of virtual contour points in the neighborhood of the i-th sampling point; This represents the distance from the robot's outer contour to the nearest actual boundary of the glass component; This represents the distance regularization constant, with a value ranging from 0.02 to 0.20 meters. This represents the distance from the i-th sampling point to the center line of the nearest doorway, in meters; This indicates the attenuation scale of the impact of the doorway, with a value ranging from 0.2 to 2.0 meters. , , These represent the weights of the high-density point band, the virtual contour, and the near-boundary distance, respectively, with values ​​ranging from 0.2 to 3.0. It should be noted that the above formula reflects that the more high-density dots, the stronger the virtual contour, and the closer the robot is to the glass boundary, the higher the risk; while the farther away from the doorway, the weaker the decision value of passing through the doorway. Therefore, exponential decay is used to reflect the semantic differences in local passage. The real-time decision-making process includes: assessing the path risk intensity. If the value exceeds the set threshold, the robot will be instructed to decelerate, deflect, or pause briefly; if a doorway candidate area is detected at the same time, the doorway crossing strategy will be adopted. It should be noted that, in this embodiment, this step can transform the perceived anomalies caused by local non-uniform reflection of the glass into the robot's actual path adjustment, pausing, avoidance, and crossing behaviors. S43: Based on the robot's behavior simulation result set and combined with the 3D point cloud anomaly information, calculate the passability credibility and obstacle contour salience of each region, and generate a joint annotation result set.

[0041] The joint annotation result set includes: obstacle outline annotation, passable space annotation, effective boundary annotation of doorway, and annotation of anomalous echo affected area; It should be noted that, in this embodiment, after obtaining the timing of robot behavior, the point cloud region in the scene is reliably labeled based on the results of the robot passing through, being blocked, detouring, and stopping. It should be noted that in this embodiment, the scene is divided into a passage area, a trial area, and a blockage area based on the robot trajectory, and then the abnormal point cloud density and the actual glass installation boundary are superimposed to generate a multi-category annotation layer; For example, in this embodiment, a strategy for obtaining passability credibility is provided, specifically as follows: ; In the formula, This represents the passability confidence level of the z-th local region; This indicates the number of times the robot successfully traversed the area; This represents the number of valid events in which the robot passes through the edge without colliding; This indicates the number of times the robot slowed down and stopped in that area; This indicates the number of times the robot detoured and abandoned the area. This represents the smoothing constant, which ranges from 1 to 10.

[0042] It should be noted that the above formula reflects that successful passage and stable edge-hugging passage will increase the reliability of passage, while frequent stops and detours will decrease the reliability of passage; thus, what is obtained is not an empty space in a purely geometric sense, but a safe passage space that is closer to the actual navigation. For example, in this embodiment, a strategy for obtaining the saliency of obstacle contours is provided, specifically as follows: ; In the formula, Indicates the saliency of the obstacle contour in the z-th local region; This indicates the total length of the candidate obstacle boundaries within the region; Indicates the area of ​​the region; This indicates the proportion of anomalous echo points within the region; This represents the baseline value for the proportion of abnormal echoes, and its value ranges from 0.01 to 0.30. This indicates the credibility of passage in the area; This represents the passability credibility decay scale, with a value ranging from 0.2 to 0.8; It should be noted that the above formula reflects that the greater the boundary length density, the more anomalous echoes, and the lower the passability confidence, the more the area should be marked as a strong obstacle contour area. S4 also includes: S44: Encapsulate the 3D point cloud samples, behavior simulation result sets, and joint annotation result sets from each simulation round, calculate the sample validity, filter and sort them to generate the final navigation training dataset.

[0043] It should be noted that in this embodiment, the 3D point cloud samples, behavior simulation result set and joint annotation result set in each simulation round are uniformly encapsulated to form data entries that can be directly used for navigation training. The data entries include: local non-uniform reflection point cloud, obstacle contour annotation, passable space annotation, effective boundary annotation of doorway, and robot path execution record; The robot path execution record includes: timestamp, robot pose, linear velocity, angular velocity, risk intensity, whether a pause was triggered, whether a detour was triggered, and whether the robot successfully passed through the doorway. For example, in this embodiment, a strategy for obtaining sample validity is provided, specifically as follows: ; In the formula, Indicates the first Sample validity of each data entry; This indicates that the valid path length has been marked in this entry; This represents the baseline value for path length, ranging from 1.0 to 20.0 meters. This indicates the length of abnormal glass boundary coverage in this entry; This represents the baseline value for abnormal coverage, and its range is 0.1-5.0 meters. This indicates the number of valid behavioral events in the entry; This represents the baseline value for the behavioral event, and its value ranges from 1 to 0. This indicates the number of conflicts marked in this entry; This represents the baseline value for conflict penalty, which ranges from 1 to 5.

[0044] It should be noted that the above formula reflects that the more complete the path, the more sufficient the boundary anomaly coverage, and the richer the behavioral samples, the more valuable the entry is for training; if there are too many annotation conflicts, its effectiveness will be reduced.

[0045] Finally, all data entries are filtered and sorted according to sample validity, low-quality entries are removed, and samples covering different glass process areas, different boundary anomaly types, and different traffic behavior patterns are retained to form the final navigation training dataset output. It should also be noted that, in this embodiment, by jointly recording the local non-uniform reflection point cloud and the robot's behavioral trajectory, a navigation training dataset that can reflect the optical abrupt change characteristics of real glass components is formed, thereby realizing the generation of high-fidelity physical environment synthetic data for embodied intelligent robots.

[0046] Example 2: This embodiment provides an electronic device, including: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor executes the aforementioned method for generating high-fidelity physical environment synthesis data for embodied intelligent robots by calling computer programs stored in memory.

[0047] The electronic device can vary considerably depending on its configuration or performance. It may include one or more Central Processing Units (CPUs) and one or more memories, wherein the memory stores at least one computer program that is loaded and executed by the processor to implement the high-fidelity physical environment synthesis data generation method for embodied intelligent robots provided in the above-described method embodiments. The electronic device may also include other components for implementing its functions; for example, it may have wired or wireless network interfaces and input / output interfaces for data input and output. Details will not be elaborated upon in this embodiment.

[0048] Example 3: This embodiment proposes a computer-readable storage medium on which an erasable and rewritable computer program is stored. When the computer program runs on the computer device, it causes the computer device to execute the above-described method for generating high-fidelity physical environment synthetic data for embodied intelligent robots.

[0049] For example, computer-readable storage media can be read-only memory (ROM), random access memory (RAM), compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage devices.

[0050] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes 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 application.

[0051] It should be understood that determining B based on A does not mean determining B solely based on A; it also means determining B based on A and / or other information.

[0052] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired network and / or wireless network. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives (SSDs).

[0053] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0054] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0055] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0056] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0057] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0058] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0059] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating high-fidelity physical environment synthesis data for embodied intelligent robots, characterized in that, The method includes the following: S1: Divide the glass panel into multiple spatial sub-regions, construct the interlayer interface relationship between the film layer and the substrate in each spatial sub-region, and perform mapping processing in the simulation environment to obtain the partitioned physical structure expression of the local non-uniform glass component. S2: Based on the physical structure representation of partitions, the reflection, transmission and boundary abrupt behavior of the laser beam in the film are continuously tracked, and the candidate echoes of the first interface, the candidate echoes of internal reflection and the candidate echoes of abrupt bifurcation are merged and converted into a three-dimensional point cloud representation containing fracture and virtual contour features. S3: Based on three-dimensional point cloud representation, the film thickness, surface roughness and interface state of each spatial sub-region are independently and randomly perturbed, while the film boundary is randomly morphologically changed to form a point cloud perturbation sample set. S4: Based on the point cloud perturbation sample set, combined with the simulation of the robot's motion navigation behavior in the vicinity of the glass component, output the corresponding obstacle contour annotation, passable space annotation and path execution record; at the same time, it is uniformly encapsulated with the 3D point cloud representation and output as a navigation training dataset.

2. The method for generating high-fidelity physical environment synthesis data for embodied intelligent robots according to claim 1, characterized in that, S1 includes: S11: Divide the glass panel into multiple spatial sub-regions with different physical properties, and calculate the non-uniformity index of each spatial sub-region to obtain the region division result set; S12: For each spatial sub-region, establish a multi-layer structure in the thickness direction, and then calculate the equivalent refractive index of each spatial sub-region; S13: Based on the region division result set and equivalent refractive index, for different material layers in the multilayer structure of each spatial sub-region, the interface of each layer is corrected by introducing a rough interface reflection factor, and the transition width between adjacent spatial sub-regions is calculated to form the interlayer interface relationship between each spatial sub-region. S14: The region division result set, the equivalent refractive index of the multi-layer structure, and the interlayer interface relationship between each spatial sub-region are uniformly encapsulated to form a structured data object containing region identifiers, multi-layer parameters, and interface physical relationships, which serves as the physical structure expression unit. Then, in the simulation environment, each type of spatial sub-region is mapped to a corresponding physical material block according to the physical structure expression unit, and the physical material blocks of all spatial sub-regions are combined according to their spatial positions to obtain the partitioned physical structure expression of the local non-uniform glass component.

3. The method for generating high-fidelity physical environment synthesis data for embodied intelligent robots according to claim 2, characterized in that, S2 include: S21: Extract the rough interface reflection factor from the physical structure representation of the partition and combine it with the lidar emission parameters to determine the spatial sub-region hit by each laser beam. Calculate its effective reflection energy at the first interface and its effective transmission energy into the layer after passing through the first interface. Then compare the effective reflection energy with the preset energy detection threshold. The echo corresponding to the laser beam that exceeds the energy detection threshold is taken as an independent candidate echo of the first interface.

4. The method for generating high-fidelity physical environment synthesis data for embodied intelligent robots according to claim 3, characterized in that, S2 also includes: S22: For the laser beam with effective transmitted energy that passes through the first interface and enters the layer, continuous propagation path tracking is performed in the multi-layer structure. During the tracking process, the echo corresponding to the laser beam whose interface reflection energy exceeds the preset detection threshold is output as the internal reflection candidate echo, and its remaining propagation energy and propagation time are calculated. S23: First, extract the transition width and equivalent refractive index from the physical structure representation of the partition to calculate the path mutation intensity. Then, in the film layer interface region, by introducing the propagation path mutation mechanism, generate mutation bifurcation candidate echoes based on the comparison results of the path mutation intensity and the preset mutation intensity threshold. At the same time, calculate the path mutation intensity of the laser beam in the boundary transition width and the additional time delay caused by the boundary mutation. The preset mutation intensity thresholds include: a lower threshold and an upper threshold; S24: Merge the candidate echoes of the first interface, the candidate echoes of internal reflections, and the candidate echoes of abrupt bifurcation into a complete set of candidate echoes, and then convert the candidate echo set into a 3D point cloud; Simultaneously, the path mutation intensity and additional time delay in the candidate echoes of mutation bifurcation are converted into the received intensity of the echo point, and the point is labeled with a category, resulting in a three-dimensional point cloud representation containing features of breakage, virtual contour, and high-density point bands.

5. The method for generating high-fidelity physical environment synthesis data for embodied intelligent robots according to claim 4, characterized in that, S3 include: S31: Based on the three-dimensional point cloud representation, reverse partitioning and mapping are performed on each spatial sub-region of the glass component to infer its physical state and calculate the randomized intensity coefficient to obtain the partitioned perturbation reference set. S32: Based on the partitioned perturbation reference set, for each spatial sub-region, the film thickness, surface roughness state and interface state are independently perturbed, and the perturbed film thickness, surface roughness and interface state factors are encapsulated according to the spatial sub-region to form a partitioned randomized parameter set.

6. The method for generating high-fidelity physical environment synthesis data for embodied intelligent robots according to claim 5, characterized in that, S3 also includes: S33: Randomize the morphology of the membrane boundary region and calculate the morphological parameters of edge warping height and degumming length; The calculated warp height and degumming length, along with the resulting membrane boundary deformation geometry and the spatial distribution of the degumming region, are collectively output as the boundary morphology perturbation set. S34: Based on the partitioned randomization parameter set and the boundary morphology perturbation set, recalculate the echo intensity of each point on the basis of the baseline 3D point cloud, and generate a set of point cloud perturbation samples for multiple simulation rounds.

7. The method for generating high-fidelity physical environment synthesis data for embodied intelligent robots according to claim 6, characterized in that, S4 include: S41: Based on the point cloud perturbation sample set, each point cloud perturbation sample is registered with the robot navigation scene, and samples that meet the requirements are selected according to the scene registration coefficient to obtain the navigation scene registration sample set. S42: In the registered scene, drive the robot to perform navigation behavior simulation in the area near the glass component, make real-time decisions by calculating the path risk intensity, and continuously record all the robot's behavior time series data, and encapsulate the complete behavior time series data into the behavior simulation result set corresponding to the sample. S43: Based on the robot's behavior simulation result set and combined with the 3D point cloud anomaly information, calculate the passability credibility and obstacle contour salience of each region, and generate a joint annotation result set.

8. The method for generating high-fidelity physical environment synthesis data for embodied intelligent robots according to claim 7, characterized in that, S4 also includes: S44: Encapsulate the 3D point cloud samples, behavior simulation result sets, and joint annotation result sets from each simulation round, calculate the sample validity, filter and sort them to generate the final navigation training dataset.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for generating high-fidelity physical environment synthetic data for embodied intelligent robots as described in any one of claims 1-8.

10. An electronic device, characterized in that, include: Memory, used to store instructions; A processor is configured to execute the instructions, causing the device to perform operations that implement the high-fidelity physical environment synthetic data generation method for embodied intelligent robots as described in any one of claims 1-8.

Images