Robot mutual identification and pose estimation method and system based on perception layer coding

Abstract

The application discloses a robot mutual identification and pose estimation method and system based on a perception layer coding, which comprises a coding inverse element, a laser scanning device, a segmented detection module, a sequence matching module and a pose estimation module. The coding inverse element is arranged along the circumference of the robot outer contour by high and low reflection code elements, and an angle point isolation section is arranged between adjacent side surfaces. The laser scanning device scans a target and obtains point cloud data. The segmented detection module extracts a target contour, identifies an isolation section according to echo intensity, divides a coding surface section, and sets an adjacent turning point of the isolation section as an anchor point. The sequence matching module binarizes code elements, matches a sliding window with a coding dictionary, obtains a target identification, an absolute phase index and physical parameters, and the pose estimation module determines a relative position according to distance data. When the anchor point is detected, the absolute phase index is converted into a geometric phase of the anchor point, and a heading angle is estimated in combination with an observation azimuth angle. Therefore, the robot mutual identification and pose estimation can be realized in a radio frequency mute environment.

Description

Technical Field

[0001] This invention relates to the field of mobile robot perception and collaborative control technology, and in particular to a method and system for mutual identification and relative pose estimation of multiple robots based on perception layer coding in a radio frequency silent environment. Background Technology

[0002] With the rapid development of intelligent manufacturing and autonomous transportation systems, the collaborative application of multiple mobile robots in warehousing and logistics, industrial production lines, and hazardous work environments is becoming increasingly widespread. In these scenarios, robot identification and relative pose estimation are the technological foundations for achieving collaborative path planning, dynamic obstacle avoidance, and efficient robot passing. Currently, multi-robot systems typically rely on active radio frequency communication methods such as WiFi, Bluetooth, ultra-wideband (UWB), and radio frequency identification (RFID) for information exchange to obtain the identification, geometric parameters, and real-time pose status of peer robots.

[0003] However, in special operating conditions such as confidential environments, environments with strong electromagnetic interference, flammable and explosive locations, and radio frequency silence, the stability and availability of active radio frequency communication are severely limited, and may even be prohibited due to security strategies. Under such conditions without radio frequency communication, if the robot cannot obtain the identity and posture information of peer robots through passive sensing, it will lead to delayed avoidance decisions, reduced passing efficiency, and conservative local path planning, seriously affecting the collaborative performance and operational safety of multi-robot systems.

[0004] To address the aforementioned issues, existing technologies have proposed passive optical recognition schemes based on visual markers, reflective pillars, or retroreflective tags. These schemes utilize visual sensors or lidar mounted on robots to perceive specific optical markers on the target surface to obtain the target's identity category, distance, or azimuth information. While these schemes reduce reliance on active communication to some extent, they still have some shortcomings:

[0005] For example, most existing identification technologies based on visual markers, reflective pillars, or retroreflective tags struggle to directly obtain the absolute phase information of the target robot relative to its own coordinate system without relying on active communication, making it difficult to further calculate a stable heading angle. Because the observation end cannot establish an absolute phase reference point in the target's own coordinate system, the completeness and accuracy of pose estimation are limited.

[0006] Secondly, existing labeling schemes have limited encoding capacity, making it difficult to simultaneously carry at least some identity attributes, including target identity, size category, priority, and other physical parameters used for avoidance decisions. This means that the identification results cannot directly provide structured physical parameter inputs for local avoidance planning, still requiring additional data links or prior databases for parameter queries, increasing system complexity and decision delay.

[0007] Therefore, it is evident that existing technologies cannot simultaneously achieve robust identification and relative pose estimation of peer robots using only two-dimensional LiDAR without radio frequency communication. Summary of the Invention

[0008] Therefore, the main objective of this invention is to provide a method and system for robot mutual recognition and pose estimation based on perceptual layer coding, which is suitable for radio frequency silent environments and can simultaneously realize peer robot identity recognition and pose estimation under the limited field of view of two-dimensional lidar.

[0009] To achieve the above objectives, according to one aspect of the present invention, a robot mutual recognition and pose estimation system based on perceptual layer coding is provided, comprising:

[0010] Encoded reflective element: It consists of high and low reflective code elements, uses cyclic unique window encoding, is arranged circumferentially along the outer contour of the robot, and has corner isolation segments at the junction of adjacent sides;

[0011] Laser scanning device: Installed on the robot, used to scan and acquire point cloud data containing the target robot's coded reflective components;

[0012] Segmentation detection module: used to extract continuous target contours from point cloud data, identify corner isolation segments based on point cloud echo intensity, divide coded surface segments, and define the geometric turning points adjacent to corner isolation segments as anchor points;

[0013] Sequence matching module: It is used to binarize the corresponding symbols of the coded surface segment according to the echo intensity, and match them with the coded dictionary using a sliding window of preset length to obtain the target sequence identifier, the absolute phase index of the first symbol of the observation window, and the mapped physical parameters.

[0014] Pose estimation module: used to determine the relative position of the target based on the distance data of the point cloud; and when an anchor point is detected, to calculate the geometric phase of the anchor point based on the absolute phase index and global phase offset, and to estimate the relative heading angle of the target by combining the azimuth angle of the anchor point observation.

[0015] In a possible preferred embodiment, the system further includes: a geometric correction module, used to perform perspective distortion compensation on the observed symbol width based on the angle between the normal of the observation surface of the coded surface segment to be decoded and the incident direction of the laser.

[0016] In a possible preferred embodiment, the pose estimation module is further configured to, when no anchor point is detected and only a single observable plane exists, select at least two effective windows with different observation states on the same coded surface segment, solve for the initial value of the target heading by geometric constraints, and obtain the target relative heading angle by recursive filtering fusion.

[0017] In a possible preferred embodiment, the geometric constraints include at least one of the following: phase difference of absolute phase index, normal constraint of surface segment, and distance constraint of range value, and the constraints for solving the initial value of target heading also include at least one of the following: upper limit constraint of angular velocity, inter-frame continuity constraint, and minimum reprojection error constraint.

[0018] In a possible preferred embodiment, the cyclic unique window encoding and encoding dictionary are constructed using the de Bruin sequence, so that any consecutive symbol window on the same target robot that does not cross corner isolation segments and is within a preset length has a unique index; the encoding content includes: target sequence identification information, absolute phase index information, and associated physical parameter information for mapping including at least some of the target size category, priority, or safety boundary parameters.

[0019] In a possible preferred embodiment, the system further includes a preprocessing module for performing any of the following processes: performing filtering and denoising on the echo intensity sequence in the point cloud data, performing first valid echo retention on suspected multiple echo points, marking saturated region sampling points and restoring their center positions through neighborhood centroids, and performing suppression on distance anomalies.

[0020] In a possible preferred embodiment, the corner isolation segment is composed of a continuous low-reflection region with a width greater than the minimum symbol width, and no effective observation window crosses the corner isolation segment.

[0021] In a possible preferred embodiment, the system further includes an inter-frame fusion module for performing recursive filtering on pose observations of consecutive frames.

[0022] To achieve the above objectives, corresponding to the above system, according to another aspect of the present invention, a robot mutual recognition and pose estimation method based on perceptual layer coding is also provided, the steps of which include:

[0023] Step S1: Scan the peer robot and extract continuous target contours from the point cloud data;

[0024] Step S2 identifies corner isolation segments based on point cloud echo intensity and divides them into coded surface segments; the geometric turning points adjacent to the corner isolation segments are defined as anchor points;

[0025] Step S3 is based on the angle between the normal of the observation surface of the coded segment to be decoded and the incident direction of the laser, and performs perspective distortion compensation on the observed symbol width to obtain the local bitstream;

[0026] Step S4 binarizes the local bitstream based on the echo intensity, and matches it with the encoding dictionary using a sliding window of preset length to obtain the target sequence identifier, the absolute phase index of the first symbol of the observation window, and the mapped physical parameters.

[0027] Step S5 determines the target's relative position based on the distance data of the point cloud; the anchor point's geometric phase is calculated based on the absolute phase index and global phase offset, and the target's relative heading angle is estimated by combining the anchor point's observed azimuth angle.

[0028] In a possible preferred embodiment, the method steps further include:

[0029] In step S6, when no anchor point is detected and only a single observable plane exists, select at least two effective windows with different observation states on the same coded surface segment and obtain their absolute phase index and ranging value respectively; based on the phase difference of the absolute phase index, the normal constraint of the surface segment, and the distance constraint of the ranging value, solve for the candidate value set of the target heading; introduce at least one of the constraints of upper limit of angular velocity, inter-frame continuity, and minimum reprojection error, and select the initial heading value with the best consistency from the candidate value set, and obtain the target relative heading angle through recursive filtering fusion.

[0030] The robot mutual identification and pose estimation method and system based on perception layer coding provided by this invention ingeniously designs a scheme for passive mutual identification and pose estimation in a radio frequency silent environment. This scheme uses high and low reflection symbols to form coded reflections, combined with laser scanning perception to obtain distance and echo intensity information. This innovatively utilizes the global uniqueness of local subsequences in cyclic unique window coding, enabling simultaneous identification of target identity attributes and the absolute phase index of the observation window on the global coding ring, even when only the local outer surface of the target is observed. This establishes a stable benchmark for relative pose estimation, thereby achieving global identity recognition and absolute phase index recovery under locally visible conditions. This solves the problem in existing technologies where stable heading angles cannot be calculated due to a lack of absolute phase information. Therefore, this scheme enables robots to mutually identify each other's identity attributes and estimate poses in a radio frequency silent environment, providing effective support for subsequent avoidance decisions. Attached Figure Description

[0031] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0032] Figure 1 This is a schematic diagram of the robot mutual recognition and pose estimation system based on perceptual layer coding of the present invention;

[0033] Figure 2 This is a schematic diagram of a multi-robot mutual recognition scenario, which is a system example of the present invention, where R≤5m represents the recognition distance;

[0034] Figure 3 This is a schematic diagram of the layout of the rectangular robot after the coded reflectors are unfolded on its four sides in global phase order in the system of the present invention.

[0035] Figure 4 This is a schematic representation of the encoding mapping of the encoding dictionary in the system of the present invention, to illustrate the relationship between the echo intensity signal and the local window matching principle;

[0036] Figure 5 This is a schematic diagram of the geometric relationship for anchor point mode pose estimation in the system of the present invention;

[0037] Figure 6 This is a schematic diagram illustrating the steps of the robot mutual recognition and pose estimation method based on perceptual layer coding of the present invention.

[0038] Explanation of reference numerals in the attached figures

[0039] 100. This robot; 110. Two-dimensional lidar; 120. Scanning plane; 200. Peer robot; 210. Encoded reflection device; 221. First side view; 222. Second side view; 223. Third side view; 224. Fourth side view; 230. Corner isolation segment; 240. Global phase index direction; 310. Observation window; 320. Encoded dictionary mapping table; 410. Anchor point. Detailed Implementation

[0040] To enable those skilled in the art to better understand the technical solutions of the present invention, the specific technical solutions of the present invention will be clearly and completely described below in conjunction with embodiments, so as to help those skilled in the art further understand the present invention. Obviously, the embodiments described in this application are merely some embodiments of the present invention, and not all embodiments. It should be noted that, for those skilled in the art, the embodiments and features in the embodiments of this application can be combined with each other without departing from the concept of the present invention and without conflict. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the disclosure and protection scope of the present invention.

[0041] Furthermore, the terms "first," "second," "S1," "S2," etc., used in the specification, claims, and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such features can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those described herein. At the same time, the stages described in each step are not necessarily to be implemented in the same step; it should be understood that the implementation order of the contents of each step stage can be adjusted and interchanged without violating the inventive concept, so that embodiments of the invention described herein can be implemented in orders other than those described herein. Additionally, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. Unless otherwise expressly specified and limited, the terms "set," "arrange," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or a connection within two elements. Those skilled in the art can understand the specific meaning of the above terms in this case based on the specific circumstances and in conjunction with existing technology.

[0042] To achieve simultaneous peer robot identification and pose estimation in a radio frequency silent environment, relying on the limited field of view of a two-dimensional LiDAR. For example... Figures 1 to 5 As shown, this invention provides a robot mutual recognition and pose estimation system based on perceptual layer coding, examples of which include:

[0043] Encoded reflective element: It consists of high and low reflective code elements, uses cyclic unique window encoding, is arranged circumferentially along the outer contour of the robot, and has corner isolation segments at the junction of adjacent sides.

[0044] Laser scanning device: This example uses a two-dimensional lidar, which is mounted on the robot to scan and acquire point cloud data containing the target robot's coded reflective components.

[0045] Specifically, such as Figure 1 , Figure 2 As shown, the coded reflector is positioned on the outer contour of the peer robot (in practice, both robots can be configured to recognize each other), and preferably at the scanning height of the 2D LiDAR. The recognition distance R between the robots is preferably less than or equal to 5m. The coded reflector consists of high-reflectivity and low-reflectivity symbols arranged sequentially along the outer contour of the target robot. The high-reflectivity symbols are preferably made of retroreflective film, and the low-reflectivity symbols are preferably made of diffuse-reflective matte material. The coded reflector is distributed in a closed-loop height along the outer contour of the target robot on the 2D LiDAR scanning plane.

[0046] In this case, considering that for robots with rectangular or multi-plane chassis, the arrangement of such coded reflective elements around the robot's outer contour can easily lead to cross-plane errors and sequence crosstalk at corners during observation, this example preferably includes a corner isolation segment at the junction of adjacent sides of the coded reflective element. This corner isolation segment is constructed from a low-reflectivity material to form a continuous low-intensity area, and its width is preferably set to be greater than the width of a single minimum symbol, so that no sliding window used for identification crosses the corner isolation segment.

[0047] By setting up corner isolation segments, for a rectangular robot, the encoded face segments on each side are logically part of the same global phase index system, but physically separated by corner isolation segments. This prevents mutual interference between the encoding of adjacent sides while maintaining the logical continuity of the global phase index. It ensures that the sliding window completes independent decoding only on a single side, thereby eliminating cross-face errors and sequence crosstalk at corners, and significantly improving decoding stability in rectangular chassis scenarios.

[0048] Compared to existing technologies where similar markings mostly default to continuous circular or single-sided labels, making it difficult to handle crosstalk and corner breakage issues across rectangular chassis, the solution of this invention not only retains the unique identification capability of local windows but also adapts to rectangular chassis, making it highly suitable for practical engineering implementation.

[0049] Furthermore, in this example, the cyclic unique window encoding and encoding dictionary used by the encoding reflector are preferably constructed using a de Bruin sequence. Examples of the encoded content of the encoding reflector include:

[0050] Target sequence identification information is used to distinguish different robot models, size categories, or functional categories;

[0051] Absolute phase index information is used to represent the absolute position of the current observed code segment in the target robot's peripheral coding;

[0052] The associated parameter information is used to read the corresponding length, width, height, priority, or security boundary physical parameters from the local dictionary after the sequence identifier is identified.

[0053] Furthermore, in other alternative implementations, other cyclic code sequences that satisfy the uniqueness of local windows can be used instead, so that any consecutive symbol windows on the same target robot that do not cross corner isolation segments and are within a preset length have a unique index; for example, binary encoding can be used and a window length n can be selected. For the same target robot, any consecutive symbol window that does not cross corner isolation segments and has a length of n has a unique index in its corresponding encoding dictionary. Thus, when the two-dimensional lidar identifies at least one complete window, the encoding sequence to which the target belongs, the absolute phase index corresponding to the first symbol of the window, and the physical parameters associated with the encoding sequence can be determined.

[0054] The segmented detection module is used to extract continuous target contours from point cloud data, identify corner isolation segments based on point cloud echo intensity, and divide coded surface segments; the geometric turning points adjacent to the corner isolation segments are defined as anchor points.

[0055] Specifically, the raw data output by a 2D LiDAR scanner is a point cloud sequence, and each point contains at least three pieces of information:

[0056] Distance: The distance from this point to the radar;

[0057] Angle: The azimuth angle of this point in the radar coordinate system;

[0058] Intensity: The energy intensity of the echo at that point.

[0059] In the LiDAR scanning scenario, in addition to the target robot, other obstacles such as walls and the ground may also be visible. The system first needs to segment and extract a continuous sequence of points belonging to the outer contour of the target robot from the entire frame point cloud. Each point in the extracted target contour point sequence has an echo intensity value. At the same time, since the corner isolation segment in this example uses a low-reflection material to form a continuous low-intensity area, the segment detection module scans this intensity curve to find segments whose intensity is continuously lower than the segmentation threshold, thereby detecting whether a corner isolation segment exists. If it is detected, the two sides of it are regarded as independent coded surface segments, and the geometric turning position adjacent to the corner isolation segment is defined as a candidate anchor point.

[0060] Therefore, the corner isolation segment design avoids the decoding window from simultaneously reading the encoded segments of two intersecting sides, thus solving the cross-surface crosstalk problem. Furthermore, by identifying the corner isolation segment through intensity threshold detection, it is ensured that each encoded surface segment is decoded independently and cleanly. This design is a key condition supporting the "local window unique encoding" principle on non-cylindrical robots.

[0061] Sequence matching module: It is used to binarize the corresponding symbols of the coded surface segment according to the echo intensity, and match them with the coded dictionary using a sliding window of preset length to obtain the target sequence identifier, the absolute phase index of the first symbol of the observation window, and the mapped physical parameters.

[0062] Specifically, such as Figure 4 As shown, after obtaining the local bitstream corresponding to the encoded segment, a sliding window of length n is used to match it with a pre-established encoding dictionary. During matching, either a perfect match or a fault-tolerant match allowing a limited number of bit errors can be used. When a match is successful, the following is obtained:

[0063] 1. The target sequence identifier corresponding to the current window;

[0064] 2. The absolute phase index of the first symbol of the current window on the global coding ring. ;

[0065] 3. Target physical parameters mapped from the target sequence identifier.

[0066] in It indicates the starting position of the bitstream in the current observation window, rather than the phase of the corner point itself.

[0067] At this point, the target's identity information, as well as the target robot's size, priority, and safety boundary physical parameters, have been obtained based on the matching results, thus completing the identity attribute recognition. At the same time, the distance information (distance from the point to the radar) and angle information (azimuth angle of the point in the radar coordinate system) have also been obtained based on the point cloud data. Only the heading angle is needed to determine the target robot's pose.

[0068] Pose estimation module: used to determine the relative position of the target based on the distance data of the point cloud; and when an anchor point is detected, to calculate the geometric phase of the anchor point based on the absolute phase index and global phase offset, and to estimate the relative heading angle of the target by combining the azimuth angle of the anchor point observation.

[0069] Specifically, since the sequence matching module has obtained the absolute phase index information of the target robot based on the encoding dictionary matching, and has defined the geometric turning point adjacent to the corner isolation segment as the anchor point based on the detection results of the segment detection module, the anchor point mode means that the geometric turning point of two adjacent sides has been observed. At this time, there is an absolute phase reference: that is, through the absolute phase index of the observation window and the global phase offset, the geometric phase of the anchor point in the target's own coordinate system can be uniquely determined; therefore, combined with the observation azimuth angle of the anchor point in the radar coordinate system of this robot, it can be directly calculated.

[0070] For example, such as Figure 5 As shown, in anchor point mode, set:

[0071] 1. The target anchor point is the observation azimuth angle in the robot's radar coordinate system;

[0072] 2. The absolute phase index of the first symbol of the matched window;

[0073] 3. Phase from the first symbol of the current observation window to the predefined anchor point The global phase offset;

[0074] 4. The known geometric phase angle of the anchor point in the target robot's own coordinate system.

[0075] Then first according to and Calculate the absolute phase corresponding to the anchor point, and then convert it to the target's own coordinates. This allows for the estimation of the target's heading angle. In other words, the anchor point observation angle and the target's own geometric phase should correspond to the same reference point to avoid directly mixing the observation window phase with the corner point phase.

[0076] The above relationship can be denoted as:

[0077]

[0078] in, Absolute phase index of the observation window Global phase shift The conversion yields, The relative heading angle of the target robot.

[0079] For example, for a robot with a rectangular chassis, coded reflective devices are attached to its four outer surfaces at the same height as the two-dimensional LiDAR. Several high-reflectivity and low-reflectivity symbols are arranged on each side, and corner isolation segments are set between adjacent sides. The coded surface segments are numbered according to a uniform global order, and the corner isolation segments themselves are not included in the symbol index.

[0080] This robot scans a peer robot using a 2D LiDAR scanner. When at least one corner isolation segment is detected, the system uses the corner's neighborhood point cloud to fit the normal direction of the corresponding side and performs perspective compensation and local window matching on the observation stream of that side. The target sequence identifier obtained from the matching is used to read the target size and priority, and the absolute phase index of the matching observation window is used. Used for converting anchor point geometric phase Finally, the azimuth angle observed from the anchor point was combined. Calculate the target heading angle .

[0081] As can be seen from the above pose calculation method, this scheme does not directly subtract the observation angle from the observation window phase, but first uses the absolute phase index of the observation window. Global phase shift Calculate the geometric phase of the anchor point Then by To determine the heading, this scheme, unlike traditional passive marking schemes which typically only provide angles or distances and do not establish a calculation chain of "window phase -> anchor point phase -> heading angle," directly converts the identification result into a calculable absolute heading, thus moving from "identification" to "pose estimation."

[0082] On the other hand, when the two-dimensional lidar can only observe a single side of the target robot and the corner isolation section is not visible, it can also perform heading angle estimation in the anchor-free degradation mode.

[0083] In an optional implementation, the system can still perform identification on a local window on that side to obtain the absolute phase index and corresponding ranging value at multiple times. Then, based on the phase difference, contour normal constraint, and adjacent frame continuity constraint, the initial target heading value is calculated and input into a recursive filter to converge frame by frame to estimate the target's relative heading angle. This implementation is suitable for the initial stage of meeting oncoming traffic, partial obstruction, or long-distance oblique observation conditions.

[0084] For example, at least two valid windows with different observation distances or different observation times are selected on the same coded surface segment, and the corresponding absolute phase index and range value are obtained respectively. Based on the phase difference, surface segment normal constraint, and range constraint, a set of candidate values ​​for the target heading is solved.

[0085] In addition, to reduce the risk of multiple solutions, at least one of the following constraints may be introduced in the alternative implementation:

[0086] 1. The normal to the robot chassis should be orthogonal to the local tangent of the observed surface segment;

[0087] 2. The change in target heading between adjacent time points does not exceed the preset upper limit of angular velocity;

[0088] 3. The candidate solution should be continuous with the filtering result of the previous frame;

[0089] 4. The candidate solution has the smallest reprojection error.

[0090] Therefore, the initial heading value with the best consistency is selected from the candidate value set, and the target's relative heading angle is obtained by recursive filtering and fusion. The filter can be a Kalman filter, an extended Kalman filter, a particle filter, or other recursive estimators suitable for smoothing the attitude of low-speed targets.

[0091] Based on the above examples, the present invention can simultaneously identify the identity of a peer robot and estimate its pose in a radio frequency silent environment, relying on the limited field of view of a two-dimensional lidar.

[0092] Furthermore, the window length n, minimum symbol width, corner isolation segment width, echo intensity threshold, and allowable number of bit errors in the above examples can all be selected based on the target size, lidar angular resolution, maximum identification distance, and material reflectivity.

[0093] In a preferred embodiment, the corner isolation segment width is greater than the minimum symbol width to ensure that local windows do not cross the corner isolation segment. The minimum symbol width is preferably greater than the projection width of a single laser spot on the target surface at the target working distance, and a certain sampling redundancy is reserved for oblique observations. For different chassis sizes, cyclic unique window encoding of different lengths can be selected to balance encoding capacity, manufacturing complexity, and recognition robustness.

[0094] On the other hand, considering that echo saturation, multiple echoes, and coding breaks near corners can reduce recognition robustness in high-reflectivity material applications, a preprocessing module can be further included in the system in an optional implementation to perform the following preprocessing steps:

[0095] 1. Perform low-pass filtering or local smoothing on the echo intensity sequence in the point cloud data to remove isolated noise points;

[0096] 2. Perform initial effective echo preservation on suspected multiple echo points to suppress false targets caused by specular reflection;

[0097] 3. Mark the sampling points that enter the nonlinear region or saturation region of the receiver, and recover their center positions through neighborhood centroid or local fitting;

[0098] 4. And remove or reduce the weight of outlier points and isolated strong negative points.

[0099] Furthermore, considering the projection compression problem that may occur when observing a rectangular chassis robot at an angle, in an optional embodiment, the system of the present invention also includes, for example, a geometric correction module, which is used to perform perspective distortion compensation on the observed symbol width based on the angle between the normal of the observation surface of the coded surface segment to be decoded and the incident direction of the laser.

[0100] Specifically, for each encoded surface segment to be decoded, the local normal direction of the segment and the angle θ between the normal and the laser incident direction can be estimated. The observed symbol width is then compensated based on the angle θ to obtain a corrected symbol width that more closely approximates the true physical width.

[0101] For example, in a preferred embodiment, the following condition is met:

[0102]

[0103] in, For the observation width, The correct width after compensation is θ, which is the angle between the normal to the observation surface and the incident direction of the laser.

[0104] When in anchor point mode, the local normal direction can be obtained by combining the corner geometry and fitting of adjacent point clouds; when in anchorless degradation mode, the initial value of θ can be estimated based on the phase change trend of the same surface segment in consecutive frames and the contour fitting results, and then iteratively corrected.

[0105] Furthermore, to improve the stability of pose estimation, in an optional embodiment, the system of the present invention also includes, for example, an inter-frame fusion module for performing recursive filtering on pose observations of consecutive frames.

[0106] For example, filtering and fusion can be performed on target heading angle, angular velocity, or phase observations obtained from consecutive frames to improve attitude estimation stability. The filter can be a Kalman filter, an extended Kalman filter, a particle filter, or other recursive estimators suitable for smoothing the attitude of low-speed targets.

[0107] Furthermore, to provide decision-making basis for subsequent obstacle avoidance or path planning systems, the system of the present invention can provide an obstacle avoidance decision interface to directly input the identification results into a path planning or local obstacle avoidance module. The local obstacle avoidance module can construct cost zones of different shapes and weights based on the target identity, size category, and priority obtained in this example, thereby implementing differentiated obstacle avoidance strategies for heavy-duty vehicles, tractor vehicles, or low-priority platforms.

[0108] On the other hand, corresponding to the above system example, such as Figure 6 As shown, the present invention also provides a robot mutual recognition and pose estimation method based on perceptual layer coding, the steps of which include:

[0109] Step S1: Scan the peer robot and extract continuous target contours from the point cloud data;

[0110] Step S2 identifies corner isolation segments based on point cloud echo intensity and divides them into coded surface segments; the geometric turning points adjacent to the corner isolation segments are defined as anchor points;

[0111] Step S3 is based on the angle between the normal of the observation surface of the coded segment to be decoded and the incident direction of the laser, and performs perspective distortion compensation on the observed symbol width to obtain the local bitstream;

[0112] Step S4 binarizes the local bitstream based on the echo intensity, and matches it with the encoding dictionary using a sliding window of preset length to obtain the target sequence identifier, the absolute phase index of the first symbol of the observation window, and the mapped physical parameters.

[0113] Step S5 determines the target's relative position based on the distance data of the point cloud; the anchor point's geometric phase is calculated based on the absolute phase index and global phase offset, and the target's relative heading angle is estimated by combining the anchor point's observed azimuth angle.

[0114] In step S6, when no anchor point is detected and only a single observable plane exists, select at least two effective windows with different observation states on the same coded surface segment and obtain their absolute phase index and ranging value respectively; based on the phase difference of the absolute phase index, the normal constraint of the surface segment, and the distance constraint of the ranging value, solve for the candidate value set of the target heading; introduce at least one of the constraints of upper limit of angular velocity, inter-frame continuity, and minimum reprojection error, and select the initial heading value with the best consistency from the candidate value set, and obtain the target relative heading angle through recursive filtering fusion.

[0115] Furthermore, in an optional implementation, the steps further include:

[0116] Step S7 performs recursive filtering on the pose observations of consecutive frames to improve the stability of pose estimation.

[0117] Furthermore, in an optional implementation, step S1 further includes:

[0118] The echo intensity sequence in the point cloud data is filtered and denoised. The first valid echo is retained for suspected multiple echo points. The sampling points in the saturation area are marked and their center positions are restored by the neighborhood centroid. The distance anomaly points are suppressed.

[0119] In summary, the robot mutual recognition and pose estimation method and system based on perceptual layer coding provided by this invention innovatively utilizes the global uniqueness of local subsequences in cyclic unique window coding. Even if only the local outer surface of the target is observed, the target identity and the absolute phase index of the observation window on the global coding ring can be determined simultaneously, establishing a stable benchmark for relative pose estimation. This achieves global identity recognition and absolute phase index recovery under local visibility conditions, solving the problem in the prior art that the stable heading angle cannot be calculated due to the lack of absolute phase information.

[0120] This scheme effectively suppresses crosstalk and projection distortion during oblique observation by setting boundary isolation sections at the junction of adjacent sides of the multi-plane chassis and combining them with incident angle geometric correction, thus significantly improving the decoding robustness of the rectangular chassis robot under non-orthogonal observation conditions.

[0121] Meanwhile, by designing a synergy between the anchor point mode and the no-anchor point degradation mode, this scheme can accurately calculate the heading angle based on global phase offset when corner points are visible, and iteratively converge the initial heading value based on multi-observation state constraints when only one side is visible, which significantly improves the usability of pose estimation under local occlusion conditions.

[0122] Furthermore, by embedding size, priority, and security boundary parameters into the encoding sequence, this scheme also achieves integrated encoding of physical parameters and identity recognition, as well as structured decision output. This allows the recognition results to be directly output as structured avoidance decision inputs, reducing system complexity and decision delay.

[0123] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The present invention is limited only by the claims and their full scope and equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.

[0124] Those skilled in the art will understand that, besides implementing the system, apparatus, unit, and its modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and its modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0125] Furthermore, all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0126] Furthermore, various different implementations of the present invention can be combined arbitrarily, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed in the present invention.

Claims

1. A robot mutual recognition and pose estimation system based on perceptual layer coding, comprising: Encoded reflector: It consists of high and low reflectivity symbols, uses cyclic unique window encoding, is arranged circumferentially along the outer contour of the robot, and has corner isolation segments at the junction of adjacent sides; wherein the cyclic unique window encoding and encoding dictionary are constructed using de Bruin sequence, so that any consecutive symbol window on the same target robot that does not cross the corner isolation segment and is within a preset length has a unique index; the encoding content includes: target sequence identification information, absolute phase index information, and associated physical parameter information used to map at least some of the target size category, priority, or safety boundary parameters; the corner isolation segment consists of a continuous low reflectivity area, the width of which is greater than the minimum symbol width, and no effective observation window crosses the corner isolation segment; Laser scanning device: Installed on the robot, used to scan and acquire point cloud data containing the target robot's coded reflective components; Segmentation detection module: used to extract continuous target contours from point cloud data, identify corner isolation segments based on point cloud echo intensity, divide coded surface segments, and define the geometric turning points adjacent to corner isolation segments as anchor points; Sequence matching module: It is used to binarize the corresponding symbols of the coded surface segment according to the echo intensity, and match them with the coded dictionary using a sliding window of preset length to obtain the target sequence identifier, the absolute phase index of the first symbol of the observation window, and the mapped physical parameters. Pose estimation module: used to determine the relative position of the target based on the distance data of the point cloud; and when an anchor point is detected, to calculate the geometric phase of the anchor point based on the absolute phase index and global phase offset, and to estimate the relative heading angle of the target by combining the azimuth angle of the anchor point observation.

2. The system according to claim 1, further comprising: The geometric correction module is used to perform perspective distortion compensation on the observed symbol width based on the angle between the normal of the observed surface of the encoded surface segment to be decoded and the incident direction of the laser.

3. The system according to claim 1, wherein the pose estimation module is further configured to, when no anchor point is detected and only a single observable plane exists, select at least two effective windows with different observation states on the same coded surface segment, solve for the initial value of the target heading by combining geometric constraints, and obtain the target relative heading angle by recursive filtering fusion.

4. The system according to claim 3, wherein the geometric constraints include: The constraints for solving the initial value of the target heading include at least one of the following: phase difference of the absolute phase index, normal constraint of the surface segment, and distance constraint of the ranging value. The constraints for solving the initial value of the target heading also include at least one of the following: upper limit constraint of angular velocity, inter-frame continuity constraint, and minimum constraint of reprojection error.

5. The system according to claim 1, further comprising: The preprocessing module is used to perform any of the following processes: filtering and denoising the echo intensity sequence in the point cloud data, preserving the first valid echo for suspected multiple echo points, marking the sampling points in the saturation region and restoring their center position through the neighborhood centroid, and suppressing distance outliers.

6. The system according to any one of claims 1 to 5, further comprising: The inter-frame fusion module is used to perform recursive filtering on the pose observations of consecutive frames.

7. A robot mutual recognition and pose estimation method based on perceptual layer coding, used to control the system as described in any one of claims 1 to 6, comprising the following steps: Step S1: Scan the peer robot and extract continuous target contours from the point cloud data; The cyclic unique window encoding and encoding dictionary are constructed using the de Bruin sequence, ensuring that any consecutive symbol window on the same target robot that does not cross a corner isolation segment and is within a preset length has a unique index. The encoding content includes: target sequence identification information, absolute phase index information, and associated physical parameter information used to map at least some of the target size category, priority, or safety boundary parameters. The corner isolation segment is composed of a continuous low-reflection area, the width of which is greater than the minimum symbol width, and no effective observation window crosses the corner isolation segment. Step S2 identifies corner isolation segments based on point cloud echo intensity and divides them into coded surface segments; the geometric turning points adjacent to the corner isolation segments are defined as anchor points; Step S3 is based on the angle between the normal of the observation surface of the coded segment to be decoded and the incident direction of the laser, and performs perspective distortion compensation on the observed symbol width to obtain the local bitstream; Step S4 binarizes the local bitstream based on the echo intensity, and matches it with the encoding dictionary using a sliding window of preset length to obtain the target sequence identifier, the absolute phase index of the first symbol of the observation window, and the mapped physical parameters. Step S5 determines the target's relative position based on the distance data of the point cloud; the anchor point's geometric phase is calculated based on the absolute phase index and global phase offset, and the target's relative heading angle is estimated by combining the anchor point's observed azimuth angle.

8. The method of claim 7, wherein the step further comprises: Step S6: When no anchor point is detected and only a single observable plane exists, select at least two valid windows with different observation states on the same coded surface segment and obtain their absolute phase index and distance measurement value respectively. Based on the phase difference of the absolute phase index, the normal constraint of the surface segment, and the distance constraint of the range value, the set of candidate values ​​for the target heading is solved. Introducing at least one of the following constraints: upper limit of angular velocity, inter-frame continuity, and minimum reprojection error, the initial heading value with the best consistency is selected from the candidate value set, and the target relative heading angle is obtained by recursive filtering and fusion.

Images