Multi-modal fusion perception-based precise positioning method for coal mine auxiliary transportation robot

Abstract

The application relates to a multi-modal fusion perception coal mine auxiliary transport robot precise positioning method, relates to the technical field of coal mine robot positioning, and comprises the following steps: acquiring inertial navigation data of a transport robot, constructing a first inertial navigation track and determining a position; collecting RFID tag information in real time, correcting a follow-up error of the first inertial navigation track according to the RFID tag information, generating a second inertial navigation track and a corresponding position; based on a preset periodic correction constraint, performing multi-anchor point position fitting correction on the second inertial navigation track, generating a third inertial navigation track and determining the position thereof; collecting real-time image information, verifying the third inertial navigation position according to an image recognition result; and if the verification is passed, outputting the third inertial navigation position as a final positioning result. The application solves the problems that in traditional coal mine positioning, an inertial navigation system is prone to error accumulation, positioning precision declines with time due to the fact that real-time correction of RFID anchor points is relied on, and the positioning method is difficult to adapt to complex underground environments.

Description

Technical Field

[0001] This application relates to the field of underground robot localization in coal mines, and in particular to a precise localization method for underground auxiliary transportation robots using multimodal fusion perception. Background Technology

[0002] With the accelerating pace of intelligent transformation in underground coal mines, the precise positioning of auxiliary transportation robots plays an increasingly prominent role in efficient operation and safety assurance. Existing traditional positioning methods mostly rely on single-mode technology. Due to the complex underground environment, single inertial navigation is prone to cumulative errors. Relying solely on RFID anchor point correction lacks a global optimization perspective and an independent accuracy verification mechanism, resulting in a decrease in positioning accuracy over time and insufficient stability. It is difficult to effectively distinguish between the true position and error offset, and thus cannot meet the actual needs of underground auxiliary transportation robots for high-precision and high-reliability positioning. Summary of the Invention

[0003] To address the aforementioned technical issues, this application provides a multimodal fusion perception method for precise positioning of underground auxiliary transportation robots in coal mines. This method solves the problems of large cumulative errors from single inertial navigation, lack of global optimization due to reliance on real-time anchor point correction, and insufficient positioning accuracy and poor stability caused by the lack of independent verification in traditional underground robot positioning. It is adaptable to complex underground environments.

[0004] The embodiments of this application disclose the following technical solutions:

[0005] This application provides a method for precise positioning of an underground auxiliary transport robot in coal mines using multimodal fusion perception, the method comprising:

[0006] Acquire the inertial navigation data of the transport robot, construct a first inertial navigation trajectory based on the inertial navigation data, and determine the first inertial navigation position;

[0007] The tag information of RFID anchor points identified by the transport robot along the task path is updated in real time, and the first inertial navigation trajectory is corrected for follow-up error based on the tag information to generate a second inertial navigation trajectory and the corresponding second inertial navigation position.

[0008] Based on preset periodic correction constraints, the second inertial navigation trajectory is fitted and corrected using multiple anchor points to generate a third inertial navigation trajectory and determine the position of the third inertial navigation system.

[0009] The interactive transport robot collects real-time image information and verifies the position of the third inertial navigation system based on the image recognition results;

[0010] If the location verification result is successful, the third inertial navigation position will be output as the final positioning result of the transport robot.

[0011] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0012] This application proposes a multimodal fusion perception-based method for precise positioning of underground auxiliary transportation robots in coal mines. Through inertial navigation data preprocessing and trajectory construction, RFID anchor point multi-constraint correction, image marker triangulation verification, hierarchical position verification, and global coordinate standardization output, precise positioning of the transportation robot in complex underground environments is achieved. First, the inertial navigation data of the transportation robot is acquired. An initial inertial navigation trajectory is constructed through integral calculation and filtering optimization, and an initial position reference is determined. Then, based on RFID anchor points, follow-up error correction and periodic multi-anchor point fitting are performed to generate second and third inertial navigation trajectories to gradually improve positioning accuracy. Simultaneously, the third inertial navigation position is independently verified using image marker recognition and triangulation technology to ensure the reliability of the positioning result. Finally, through underground boundary constraints and a preset position accuracy threshold verification, the third inertial navigation position that meets the conditions is used as the final positioning result and output, achieving precise positioning based on multimodal fusion.

[0013] This technical solution integrates the continuity advantages of inertial navigation, the absolute positional constraints of RFID anchor points, and the spatial geometric verification of visual positioning. It solves the problems of traditional single-sensor positioning being susceptible to interference, having large cumulative errors, and low reliability in complex underground environments. This effectively improves the adaptability of the positioning process to key scenarios such as roadway intersections and branch areas, providing technical support for the autonomous navigation and intelligent operation of auxiliary transportation robots in coal mines. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0015] Figure 1 A flowchart illustrating the precise positioning method for an underground auxiliary transport robot in coal mines using multimodal fusion perception, as provided in this application embodiment;

[0016] Figure 2 This is a schematic diagram of the process for generating a second trajectory and position by dynamically correcting the first inertial navigation trajectory according to an embodiment of this application. Detailed Implementation

[0017] This application provides a method for precise positioning of underground auxiliary transportation robots in coal mines using multimodal fusion perception, which addresses the technical problems in existing technologies, such as the easy accumulation of errors in single inertial navigation, the lack of global optimization due to reliance on real-time correction via RFID anchor points, and the lack of an independent verification mechanism leading to insufficient positioning accuracy, poor stability, and difficulty in adapting to complex underground environments.

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0020] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0021] Examples, as shown in the appendix Figure 1 As shown, this application provides a method for precise positioning of an underground auxiliary transportation robot in coal mines using multimodal fusion perception. The method includes the following steps:

[0022] S110: Acquire the inertial navigation data of the transport robot, construct a first inertial navigation trajectory based on the inertial navigation data, and determine the first inertial navigation position;

[0023] In this embodiment of the application, in the scenario where the underground environment of a coal mine is complex and it is difficult to rely on a single external positioning reference, in order to provide a continuous initial positioning trajectory, it is necessary to process and correct the inertial navigation data to generate a first inertial navigation trajectory and position with basic accuracy.

[0024] Specifically, inertial navigation data is first acquired from the inertial measurement unit (IMU) of the transport robot. This data includes at least acceleration and angular velocity data, providing the initial measurement basis for constructing the positioning trajectory.

[0025] Furthermore, the acceleration data is integrated for the first time to obtain multi-axis velocity information; then, the multi-axis velocity information is integrated for the second time to generate the preliminary position information of the transport robot, realizing the transformation from motion state information to position information.

[0026] Simultaneously, attitude estimation is performed based on angular velocity data to determine the robot's attitude information in space. This attitude information is used to correct deviations in the initial position information caused by changes in the direction of motion.

[0027] Furthermore, the position and attitude information are optimized using a filtering algorithm, and the complementarity of acceleration, angular velocity and attitude data is integrated to reduce the interference of measurement noise on position calculation. Finally, the first inertial navigation trajectory is generated and the corresponding first inertial navigation position is determined.

[0028] This step obtains raw data from the inertial measurement unit, performs integration to obtain position information, and combines attitude estimation and filtering optimization to correct deviations. This provides a continuous and basic trajectory reference for subsequent positioning correction based on RFID anchors and image recognition, solving the problem of positioning continuity when there are no external positioning signals underground, and laying the initial trajectory foundation for the overall positioning process.

[0029] Step S110 in the method provided in this application embodiment includes:

[0030] The inertial navigation data is obtained from the inertial measurement unit of the transport robot, wherein the inertial navigation data includes at least acceleration data and angular velocity data;

[0031] The acceleration data is integrated to obtain multi-axial velocity information, and the multi-axial velocity information is further integrated to generate preliminary position information;

[0032] Attitude estimation is performed based on the angular velocity data to determine the spatial attitude information of the transport robot. The initial position information of the transport robot is then corrected based on the spatial attitude information to obtain the first inertial navigation trajectory and determine the position of the first inertial navigation system.

[0033] In this embodiment of the application, in the complex environment of a coal mine, characterized by darkness, dust, and severe signal obstruction, a basic trajectory needs to be generated through the collection and processing of inertial navigation data to provide continuous initial positioning reference for the transport robot, thus laying the foundation for subsequent fusion correction.

[0034] Specifically, inertial navigation data, including three-dimensional acceleration data and three-dimensional angular velocity data, is first obtained from the inertial measurement unit carried by the transport robot. This data serves as the original basis for positioning.

[0035] Among them, acceleration data can reflect the robot's motion acceleration in each axis of three-dimensional space, while angular velocity data can reflect the robot's rotational motion state.

[0036] For example, if the robot moves along an underground tunnel, the acceleration data might include 0.5 m / s² on the x-axis (direction of movement). 2 0.1 m / s along the y-axis (horizontal and vertical direction of travel). 2 0.05 m / s along the z-axis (vertical direction) 2 Angular velocity data may include 0.02 rad / s around the x-axis, 0.01 rad / s around the y-axis, and 0.03 rad / s around the z-axis.

[0037] Furthermore, the acceleration data undergoes a first integration process to obtain multi-axis velocity information. This integration process requires incorporating a time parameter to convert the cumulative acceleration over time into velocity.

[0038] For example, with an acceleration of 0.5 m / s² on the x-axis 2 Integrating over 10 seconds yields a velocity of 5 m / s along the x-axis. Similarly, the velocity information along the y-axis and z-axis is obtained, forming multi-axis velocity information.

[0039] Furthermore, the multi-axis velocity information is integrated a second time to generate preliminary position information. Again, using time as a variable, the cumulative velocity over time is converted into displacement. For example, if the x-axis velocity is 5 m / s for 10 seconds, the preliminary position change in the x-axis direction is 50 m. The preliminary position coordinates of the robot are obtained by combining the information from all axes.

[0040] Furthermore, attitude estimation is performed based on angular velocity data to determine the spatial attitude information of the transport robot. The angular velocity data allows calculation of the robot's pitch, roll, and yaw angles during movement, thereby determining whether the robot is tilting, turning, etc.

[0041] For example, based on the angular velocity of 0.02 rad / s around the x-axis, a certain pitch change of the robot can be estimated; by combining the angular velocities around the y-axis and z-axis, the current spatial attitude of the robot can be determined.

[0042] Simultaneously, the initial position information is corrected by incorporating spatial attitude information. Since the robot may tilt or turn underground, the initial position information does not take into account the influence of attitude. Therefore, it is necessary to adjust the position components of each axis using attitude information to make the position calculation more closely match the actual motion state.

[0043] For example, if a robot generates a 10° pitch angle while climbing a slope underground, its initial position along the x-axis (direction of travel) is calculated to be 80m, but the actual displacement component along the z-axis (vertical direction) caused by pitch is not considered. When correcting the attitude information, the x-axis displacement is decomposed into 78.78m (80×cos10°) in the horizontal direction and 13.89m (80×sin10°) in the vertical direction through trigonometric function conversion. After correction, the z-axis position is adjusted from the initially calculated 0.5m to 14.39m, making the position coordinates more consistent with the actual climbing state.

[0044] Furthermore, the position and attitude information are optimized using a Kalman filter algorithm. This algorithm can integrate the complementarity of acceleration, angular velocity, and attitude data to reduce errors caused by measurement noise, such as filtering out abnormal acceleration values ​​caused by mechanical vibration, resulting in a smoother and more accurate trajectory.

[0045] For example, if the robot is moving through a bumpy alley, the accelerometer may momentarily detect an x-axis speed of 3.2 m / s. 2 The abnormal value (far exceeding the normal forward acceleration of 0.5~0.8 m / s²) 2 The Kalman filter algorithm predicts the value (based on the preceding motion trend, it is estimated to be 0.6 m / s). 2 Residual analysis between the measured value and the actual value determined that this value was noise, and it was corrected to 0.65 m / s. 2 The corresponding velocity integral result was corrected from the original instantaneous jump of 3.2 m / s to a smooth 0.65 m / s, ultimately preventing the x-axis position trajectory from exhibiting an abnormal offset of 1.5 m and maintaining a continuous and stable path curve.

[0046] Finally, the above processing generates the first inertial navigation trajectory and determines the corresponding first inertial navigation position. This trajectory continuously reflects the robot's motion path, while the position information specifies the robot's exact coordinates at a given moment.

[0047] This step obtains raw data from the inertial measurement unit, performs integration to obtain position information, and combines attitude estimation and filtering optimization to correct deviations. This provides a continuous and basic trajectory reference for subsequent positioning correction based on RFID anchors and image recognition, solving the problem of positioning continuity when there are no external positioning signals underground and ensuring the consistency of the positioning process.

[0048] S120: Real-time update of tag information of RFID anchor points identified by the transport robot along the task path, and correction of follow-up error of the first inertial navigation trajectory based on the tag information, generating a second inertial navigation trajectory and the corresponding second inertial navigation position;

[0049] In this embodiment of the application, in the scenario where inertial navigation in coal mines is prone to trajectory deviation due to accumulated errors, in order to calibrate the positioning deviation in real time, it is necessary to rely on the absolute position reference of RFID anchor points to dynamically correct the first inertial navigation trajectory to ensure the timeliness of positioning.

[0050] Specifically, the tag information of RFID anchor points identified by the transport robot along the task path is updated in real time. This information includes data such as the unique identifier of the anchor point and the signal strength, which serves as a key reference for positioning correction.

[0051] Furthermore, the relative position information between the transport robot and the RFID anchor point is calculated based on the tag information of the RFID anchor point.

[0052] The relative distance is calculated using signal propagation time; the relative azimuth is determined by the phase difference of the signals received by the multi-antenna RFID reader mounted on the robot.

[0053] Simultaneously, the absolute position coordinates of the RFID anchor point are retrieved based on the tag information. These absolute position coordinates are pre-entered during underground deployment and represent its precise position in the underground three-dimensional coordinate system.

[0054] Furthermore, taking the starting point of the transport robot's task as a constant point (i.e., the starting position does not participate in the correction and the initial calibration value is maintained), and taking the minimum relative distance between the robot and the current identification anchor point as the objective, the first inertial navigation trajectory is corrected for follow-up error based on scaling and rotation, so as to reduce the distance deviation between theory and reality.

[0055] Finally, the servo error correction result is output and used as the second inertial navigation trajectory, while the corresponding second inertial navigation position is extracted.

[0056] This step quickly corrects the accumulated errors of inertial navigation by capturing the latest RFID anchor point information in real time, without relying on previous anchor point data. It not only preserves the continuity of the trajectory, but also improves the real-time positioning accuracy through absolute position reference, providing a more reliable basic trajectory for subsequent global optimization and correction.

[0057] As attached Figure 2 As shown, step S120 in the method provided in this application embodiment includes:

[0058] Based on the tag information, the relative position information between the transport robot and the RFID anchor point is calculated, wherein the relative position information includes the relative distance and the relative azimuth angle;

[0059] Based on the tag information, the absolute position coordinates of the RFID anchor point are retrieved accordingly;

[0060] Using the starting point of the transport robot's task as a constant point and minimizing the relative distance as the objective, the first inertial navigation trajectory is subjected to a follow-up error correction based on scaling and rotation.

[0061] The output follow-up error correction result is the second inertial navigation trajectory, and the corresponding second inertial navigation position is extracted.

[0062] In this embodiment of the application, in the scenario where the cumulative error of inertial navigation in coal mines increases over time and relying solely on inertial navigation is prone to trajectory deviation, in order to quickly calibrate the real-time position deviation, it is necessary to rely on the absolute position reference of RFID anchor points and dynamically correct the trajectory based on the latest identified anchor point information to ensure the timeliness and accuracy of positioning and avoid interference caused by the lag of previous anchor point information.

[0063] Specifically, the relative position information between the transport robot and the RFID anchor point is first calculated based on the tag information.

[0064] The relative distance is calculated using the signal propagation time, specifically by measuring the round-trip time of the RFID signal from the anchor tag to the robot reader, combined with the speed of electromagnetic waves in the air (approximately 3 × 10⁻⁶). 8 The relative distance between two objects can be accurately calculated using the speed of light (m / s). The specific formula is "Relative distance = (Signal round-trip time × Speed ​​of light) / 2".

[0065] For example, when the robot travels along a certain section of the alley, it sends an inquiry signal to the anchor tag and records the transmission time t1. The tag receives the signal and immediately returns a response signal. The robot reader receives this signal and records the time t2. Therefore, the round-trip time Δt = t2 - t1. If the measured round-trip time of a certain anchor tag is 30 μs, substituting this into the formula, the relative distance can be obtained as (30 × 10⁻⁶). -6 s×3×10 8 m / s) / 2=4.5m.

[0066] Furthermore, the relative azimuth angle is calculated using the phase difference of the RFID reader array mounted on the robot. For example, if the phase difference between the left and right antennas of the reader receiving the same tag signal is 30°, and combined with the antenna spacing, the anchor point can be determined to be located 25° to the left of the robot's front.

[0067] Furthermore, the system retrieves the corresponding absolute position coordinates based on the tag information. These coordinates are three-dimensional coordinates determined through high-precision mapping and stored in the system before anchor points are deployed underground.

[0068] For example, the database is retrieved by searching the unique ID "RFID-017" of the anchor tag, and its pre-stored absolute position coordinates (892.5, 451.2, -120.8) are quickly retrieved and used as the absolute reference benchmark for correcting the trajectory.

[0069] Meanwhile, the starting point of the transport robot's task is kept constant, meaning the starting position remains fixed and does not participate in scaling or rotation correction, to ensure that the entire trajectory is anchored in the initially calibrated coordinate system.

[0070] Furthermore, with the goal of minimizing the relative distance, a scaling-rotation-based follow-up error correction is performed on the first inertial navigation trajectory.

[0071] Specifically, the theoretical relative distance is first calculated based on the robot's current position on the first inertial navigation system and the absolute position coordinates of the anchor point. If the first inertial navigation system shows the robot's position as (888.0, 450.0, -120.8), the theoretical distance to the anchor point (892.5, 451.2, -120.8) is 4.68m, while the actual calculated relative distance is 4.5m, with a deviation of 0.18m between the two.

[0072] At this point, by scaling the trajectory (that is, scaling the trajectory from the starting point to the current segment by a ratio of 4.5 / 4.68≈0.96) and rotating the trajectory to match the azimuth angle of the robot with that of the anchor point (e.g., if the inertial navigation system shows the anchor point azimuth angle as 10° before correction, but the actual measured azimuth angle is 15°, then the entire trajectory is rotated by 5°), the relative distance between the corrected robot position and the anchor point is minimized.

[0073] For example, if the first inertial navigation trajectory shows that the theoretical distance between the robot's current position and the RFID-017 anchor point is 4.68m and the azimuth angle is 10°, while the actual calculated relative distance is 4.5m and the azimuth angle is 15°.

[0074] First, taking the starting point (800.0, 400.0, -120.0) as a fixed point, we first scale the trajectory segment from the starting point to the current position. That is, we shorten the 88m distance from 800.0 to 888.0 in the x-axis direction by a ratio of 0.96 to 84.48m (88×0.96=84.48). After correction, the current x-coordinate is 800.0+84.48=884.48m.

[0075] Next, the first inertial navigation trajectory is rotated 5° clockwise, and the y-coordinate is adjusted by the coordinate transformation formula so that the relative distance between the corrected robot position (885.1, 450.9, -120.8) and the anchor point (892.5, 451.2, -120.8) is 4.5m and the azimuth angle is 15°, so that it perfectly matches the actual measurement value.

[0076] Finally, the output servo error correction result is used as the second inertial navigation trajectory, and the corresponding second inertial navigation position is extracted. This second inertial navigation trajectory eliminates real-time deviations based on the first inertial navigation trajectory, making the position information more consistent with the actual position of the robot.

[0077] This step utilizes the latest RFID anchor information for dynamic correction, avoiding the cumulative interference of previous anchor data, and quickly responding to real-time changes in the robot's position. While ensuring trajectory continuity, it significantly improves positioning timeliness and provides a high-precision dynamic trajectory foundation for subsequent periodic global optimization, solving the problem of traditional inertial navigation positioning changing over time.

[0078] S130: Based on the preset periodic correction constraints, perform multi-anchor point position fitting correction on the second inertial navigation trajectory to generate the third inertial navigation trajectory and determine the third inertial navigation position;

[0079] In this embodiment of the application, in the complex environment of underground coal mines, although the second inertial navigation trajectory is corrected for follow-up error through RFID anchor points, the correction of a single anchor point is easily affected by local environmental interference and has limited field of view. In order to further improve the global positioning accuracy, it is necessary to optimize the trajectory globally by integrating multiple RFID anchor point information based on preset periodic correction constraints.

[0080] Specifically, a multi-anchor sampling window is first defined based on a preset periodic correction constraint.

[0081] The window length is dynamically adjusted based on the moving speed of the transport robot and the update frequency of the positioning system; the window position is centered on the current moment and traces back to cover historical trajectory segments to ensure that a sufficient number of anchor point information are included.

[0082] Furthermore, multiple RFID anchor points are selected as a multi-anchor point reference set based on the multi-anchor point sampling window. These RFID anchor points must meet signal quality thresholds and spatial distribution uniformity requirements to avoid redundant calculations caused by anchor point clustering.

[0083] Simultaneously, a multi-anchor-point evaluation function based on the sum of position variances is constructed. This multi-anchor-point evaluation function uses the sum of the variances of the theoretical distances and the actual measured distances from each point on the second inertial navigation trajectory to multiple anchor points as the core indicator.

[0084] Furthermore, using the multi-anchor evaluation function as the objective function, a multi-anchor position fitting correction is performed on the second inertial navigation trajectory. By minimizing the sum of variances, the trajectory parameters are adjusted using an iterative optimization method, making the overall trajectory more closely match the measurement constraints of all anchor points.

[0085] Finally, the fitting and correction results are output, generating the third inertial navigation trajectory and determining the corresponding third inertial navigation position. This trajectory comprehensively utilizes the spatial constraint information of multiple anchor points from a global perspective, effectively avoiding the accumulation of local errors that may be introduced by single anchor point correction, and achieving a positioning accuracy of centimeter level.

[0086] This step achieves globally optimal positioning by periodically performing multi-anchor-point position fitting corrections while maintaining trajectory continuity. Compared to single-anchor-point correction, this method is more robust to complex underground environments and is suitable for precise positioning at key locations such as roadway intersections and branch areas, providing a reliable global position reference for the autonomous navigation of coal mine robots.

[0087] Step S130 in the method provided in this application embodiment includes:

[0088] Based on the aforementioned periodic correction constraint, a multi-anchor sampling window is defined;

[0089] Based on the multi-anchor sampling window, multiple RFID anchor points are selected as a multi-anchor reference set;

[0090] A multi-anchor point evaluation function based on the sum of position variances is constructed, and the multi-anchor point position fitting correction of the second inertial navigation trajectory is performed with the multi-anchor point evaluation function as the objective function.

[0091] In this embodiment of the application, in the underground environment of a coal mine, the trajectory may be locally optimal rather than globally optimal due to local signal interference or anchor point deployment deviation. In order to achieve a global improvement in positioning accuracy, it is necessary to periodically combine information from multiple anchor points to fit and correct the trajectory in order to avoid the limitations of single anchor point correction.

[0092] Specifically, a multi-anchor sampling window is first defined based on a preset periodic correction constraint.

[0093] The periodic correction constraint is set according to the robot's moving speed and the density of downhole anchor points. For example, a correction is triggered every 50 meters or every 3 minutes. The window length is the range of the robot's moving trajectory within the cycle (e.g., 50 meters corresponds to a window length containing 10 anchor points). The window position is the current correction time as the endpoint, and it backtracks to cover the complete cycle trajectory.

[0094] For example, if the robot travels at a speed of 0.8 m / s and the preset period is 60 seconds, the window length is 48 meters (0.8 × 60), and the window position traces back from the current position (coordinate X = 950 m) to X = 902 m, forming a sampling range that includes all identification anchor points within this interval.

[0095] Furthermore, multiple RFID anchor points are selected as a multi-anchor point reference set based on the multi-anchor point sampling window. The screening criteria include anchor point signal quality (e.g., signal round-trip time fluctuation < 0.5 μs) and spatial distribution uniformity (distance between any two anchor points > 5 m) to exclude abnormal anchor points caused by obstruction.

[0096] For example, a total of 12 anchor points were identified within a 48-meter window. After screening, 3 anchor points with excessive signal fluctuations were removed, and finally 9 anchor points were selected to form a multi-anchor point reference set. Their absolute position coordinates are (905.2, 410.5, -120.3), (912.8, 408.7, -120.3)...(949.1, 412.0, -120.3).

[0097] Furthermore, before constructing the multi-anchor-point evaluation function, it is necessary to extract the scaling and rotation scheme based on multiple follow-up error correction results.

[0098] The method provided in this application embodiment, before "constructing a multi-anchor point evaluation function based on the sum of position variances, and performing multi-anchor point position fitting correction on the second inertial navigation trajectory using the multi-anchor point evaluation function as the objective function", further includes:

[0099] Based on the multiple follow-up error correction results, multiple scaling and rotation schemes are extracted accordingly;

[0100] Random cross-mutation is performed on multiple scaling and rotation schemes;

[0101] The parameterized random crossover mutation results are compared with multiple scaling and rotation schemes, and the fitted correction scheme set is initialized based on the parameterized results.

[0102] In this embodiment of the application, in order to ensure that the correction scheme has both historical validity and global optimization potential, it is necessary to extract historical correction experience and generate an initial scheme set through random crossover mutation to provide a variety of optimization starting points for subsequent trajectory fitting, so as to avoid getting trapped in local optima.

[0103] Specifically, multiple scaling and rotation schemes are first extracted based on multiple follow-up error correction results.

[0104] Among them, the follow-up error correction results record the trajectory adjustment parameters of the robot at different positions and different anchor points, the scaling scheme reflects the correction coefficient of the overall trajectory ratio (e.g., 0.96 for anchor point A and 1.02 for anchor point B), and the rotation scheme reflects the correction angle of the trajectory azimuth angle (e.g., +3° for anchor point C and -2° for anchor point D).

[0105] For example, schemes are extracted from the past 10 follow-up correction results: scheme 1 (scale 0.96, rotation +5°), scheme 2 (scale 0.98, rotation -3°), scheme 3 (scale 1.01, rotation +2°) ... scheme 10 (scale 0.97, rotation -1°) to cover the trajectory adjustment features of different road segments.

[0106] Furthermore, random cross-mutation is performed on multiple scaling and rotation schemes.

[0107] The crossover process involves combining the scaling factor and rotation angle of two different schemes. For example, the scaling factor of 0.96 in scheme 1 is crossed with the rotation factor of -3° in scheme 2 to generate a new scheme (0.96, -3°).

[0108] In addition, the mutation process randomly adjusts the parameters of a single scheme within a small range. For example, the scaling of scheme 3 is mutated from 1.01 to 1.005 (within ±0.01), and the rotation of +2° is mutated to +2.3° (within ±0.5°) to introduce new combinations of parameters and enhance the diversity of the scheme set.

[0109] For example, crossover mutation is performed on 10 original schemes: scheme 4 (0.99, +4°) is crossed with scheme 7 (1.03, -1°) to obtain (0.99, -1°) and (1.03, +4°); scheme 5 (0.95, +1°) is mutated to obtain (0.945, +1.2°), and finally 20 mixed candidate sets containing original schemes, crossover schemes and mutated schemes are generated.

[0110] Meanwhile, the random crossover mutation results and multiple scaling and rotation schemes are parameterized. The scaling factor is constrained to the range of [0.9, 1.1] (to avoid overscaling that could cause trajectory distortion), the rotation angle is limited to the range of [-10°, 10°] (to conform to the physical limits of turning in underground roadways), and each scheme is assigned a unique parameter identifier (such as a binary tuple of "scaling factor - rotation angle" (s, θ)).

[0111] For example, the scheme (0.96, -3°) is parameterized as (s=0.96, θ=-3°) and the scheme (1.03, +4°) is parameterized as (s=1.03, θ=+4°) so that all schemes satisfy the constraints 0.9≤s≤1.1 and -10°≤θ≤10°.

[0112] Furthermore, the fitting correction scheme set is initialized based on the parameterization results. This involves selecting non-repeating and evenly distributed schemes from the parameterized schemes, ultimately retaining 30 schemes to form the initial fitting correction scheme set. This set includes both historically valid schemes that have been verified in practice and novel schemes generated through crossover mutation, balancing stability and exploratory potential.

[0113] For example, the initial fitting correction scheme set includes the original effective scheme (0.98, -3°), the cross scheme (0.96, +2°), and the variant scheme (1.005, -1.5°), covering the scaling factor range of 0.92 to 1.08 and the rotation angle range of -8° to +7°, to ensure full coverage of possible optimal solutions.

[0114] The method provided in this application embodiment includes the step of "constructing a multi-anchor point evaluation function based on the sum of position variances, and performing multi-anchor point position fitting correction on the second inertial navigation trajectory using the multi-anchor point evaluation function as the objective function" as follows:

[0115] Using the set of fitted correction schemes as the initial solution set, the second inertial navigation trajectory is updated accordingly;

[0116] The objective function value of the updated result is updated by traversing the evaluation trajectory according to the multi-anchor point evaluation function;

[0117] Based on preset iterative constraints, the fitting correction scheme set is iteratively updated in combination with the objective function value;

[0118] Select the optimal objective function value from all trajectory update results and output it as the second inertial navigation trajectory.

[0119] In this embodiment, to avoid the accumulation of local errors from single-anchor-point correction, multi-anchor-point collaborative optimization is required to achieve global optimality of the trajectory. This process iteratively updates the set of fitted correction schemes, gradually converging to the global optimal solution to ensure the consistency and accuracy of the trajectory under multiple anchor-point constraints.

[0120] First, the second inertial navigation trajectory is updated using the fitted correction scheme set as the initial solution set. The fitted correction scheme set contains 30 sets of parameterized scaling and rotation schemes (e.g., (s=0.98, θ=-3°)), and each scheme corresponds to a trajectory adjustment strategy.

[0121] Specifically, for each fitting correction scheme, the second inertial navigation trajectory is scaled and rotated with the task starting point as the constant point. That is, the coordinates of each point on the trajectory are first adjusted proportionally by the scaling factor s, and then rotated around the starting point by an angle θ to generate new trajectory point coordinates.

[0122] For example, for the fitted correction scheme (s=0.96, θ=+2°), the point P1 (820.0, 410.0, -120.0) on the trajectory is scaled to (819.2, 409.6, -120.0), and then rotated by 2° to become (818.9, 410.3, -120.0). Similarly, all points on the trajectory are processed to obtain the updated trajectory based on this scheme. That is, this operation is repeated for 30 schemes to generate 30 candidate trajectories.

[0123] Furthermore, the objective function value of the result is updated by traversing the evaluation trajectory according to the multi-anchor evaluation function.

[0124] The multi-anchor evaluation function is defined as the sum of the variances of the actual measured distances from each point on the trajectory to all reference anchor points and the theoretically calculated distances. The specific calculation formula is as follows:

[0125]

[0126] Where n is the number of trajectory sampling points (e.g., 100), representing 100 feature points uniformly selected on the second inertial navigation trajectory segment covered by the multi-anchor sampling window; m is the number of reference anchor points (e.g., 5), representing 5 RFID anchor points with qualified signal quality and uniform spatial distribution selected from the sampling window, serving as the absolute position reference for trajectory correction.

[0127] In addition, the signal propagation time (e.g., 4.5m) is used to calculate the actual relative distance between the transport robot at the i-th trajectory sampling point and the j-th reference anchor point, based on the round-trip time of the RFID signal. This distance is used to update the Euclidean distance from the i-th point to the j-th anchor point on the trajectory, and to measure the degree of agreement between the corrected trajectory and the absolute position of the anchor point.

[0128] For example, for point P1 (818.9, 410.3, -120.0) on trajectory T1, the theoretical distance to the reference anchor point RFID-017 (892.5, 451.2, -120.8) is calculated to be 4.65m, while the actual measured distance is 4.5m, with a deviation of 0.15m. By traversing all combinations of points and anchor points on the trajectory, the summation yields JT1 = 3.2m² for this trajectory. Similarly, the J value for all 30 trajectories is calculated.

[0129] Meanwhile, based on the preset iterative constraints, the fitting correction scheme set is iteratively updated in combination with the objective function value.

[0130] The iterative constraints include the maximum number of iterations (e.g., 10 times) and the convergence threshold (e.g., ΔJ < 0.05m2).

[0131] Specifically, in each iteration, the top 10% of schemes with the best objective function values ​​(e.g., 3 groups) are retained, and cross-mutation is performed on them to generate new schemes. That is, two groups are randomly selected from the 3 retained optimal schemes for cross-mutation.

[0132] For example, select scheme A (s=0.98, θ=-3°) and scheme B (s=1.02, θ=+2°), combine the scaling factor of scheme A with the rotation angle of scheme B to generate a new scheme C (s=0.98, θ=+2°), and combine the scaling factor of scheme B with the rotation angle of scheme A to generate a new scheme D (s=1.02, θ=-3°).

[0133] In addition, the parameters of each retained scheme were mutated within a small range. For example, the scaling factor of scheme E (s=1.00, θ=+1°) was adjusted to 1.005 within the range of ±0.01, and the rotation angle was adjusted to +1.3° within the range of ±0.5°, resulting in the mutated scheme F (s=1.005, θ=+1.3°).

[0134] By cross-introducing advantageous parameter combinations from different schemes and expanding the parameter search range through mutation, the generated new schemes will replace the corresponding number of schemes with the worst objective function values ​​in the original solution set, thus ensuring the stability and continuous optimization of the solution set size.

[0135] For example, after the first iteration, the optimal solution (s=0.975, θ=+1.2°) has J=2.8m2. Five new solutions are generated by cross-mutation of the optimal solution (s=0.975, θ=+1.2°) to replace the five solutions with the largest J values ​​in the original solution set (e.g., J=4.5m2).

[0136] Furthermore, the optimal objective function value among all trajectory update results is selected and output as the second inertial navigation trajectory. After multiple iterations (e.g., 10 times), the solution set converges to the scheme with the minimum J value (e.g., s=0.982, θ=+0.8°, J=1.5m2), and the corresponding trajectory is the corrected optimal trajectory (i.e., the second inertial navigation trajectory).

[0137] This step refines the trajectory through iterative optimization under global constraints of multiple anchor points, avoiding local deviations inherent in single-anchor-point corrections. The final output second inertial navigation trajectory achieves globally optimal positioning while maintaining continuity, providing a high-precision position reference for coal mining robots in complex environments.

[0138] S140: The interactive transport robot collects real-time image information and verifies the position of the third inertial navigation system based on the image recognition results;

[0139] In this embodiment of the application, since there may be anchor point signal interference and deployment deviations underground, in order to ensure the final reliability of the positioning results, the image recognition results need to be verified. By cross-verifying the image information and the inertial navigation information, the reliability of the positioning accuracy can be further improved.

[0140] Specifically, the transport robot first collects real-time image information and performs image marker recognition to obtain M image markers.

[0141] These image markers are pre-deployed underground markers with unique visual characteristics. After images are captured by cameras mounted on the robot, they are identified and extracted using deep learning algorithms to obtain the number of markers and their coordinate information in the images.

[0142] Furthermore, the number of M identified image markers is judged: if M is less than 3, it means that the current visual reference is insufficient and triangulation verification cannot be performed. The robot needs to be controlled to update the real-time image information and re-identify the image markers until a sufficient number of markers are obtained.

[0143] Conversely, if M is greater than or equal to 3, then the analysis calculates whether there exists a combination of three image markers among the M image markers that satisfies the preset azimuth constraint.

[0144] Among them, the three-image marker combination refers to the selection of 3 from M image markers. The azimuth constraint means that the relative azimuth angle between the 3 image markers must be consistent with the pre-stored three-dimensional spatial orientation relationship, so as to ensure that the selected combination is a real spatially associated image marker, rather than a randomly identified interference target.

[0145] Specifically, if there is no combination of three image markers that satisfies the azimuth constraint, the real-time image information is continuously updated, the image markers are re-identified, and different combinations of three image markers are iteratively selected for azimuth constraint discrimination until a combination that meets the conditions is found.

[0146] Furthermore, if a combination of three image markers exists that satisfies the azimuth constraint, triangulation is performed based on this combination. Using the pre-stored absolute position coordinates of the three markers, combined with the visual azimuth angles between the robot's camera and each marker, the robot's visual positioning coordinates are calculated through trigonometric operations.

[0147] Furthermore, the triangulation result is compared with the third inertial navigation position to perform position verification. If the difference between the two three-dimensional coordinates is within the preset accuracy threshold range, the verification is deemed successful.

[0148] This step introduces an image recognition verification process, utilizing the spatial uniqueness of image information to cross-verify the inertial navigation position after correction by multiple anchor points. This effectively compensates for potential errors in a single anchor point or inertial navigation system, providing double assurance for the final positioning result of the transport robot. It is suitable for high-precision positioning scenarios in key underground coal mine operation areas.

[0149] Step S140 in the method provided in this application embodiment includes:

[0150] Acquire images and perform image marker recognition to obtain M image markers;

[0151] If M is less than 3, then update the real-time image information and update the image markers;

[0152] If M is greater than or equal to 3, then analyze and calculate whether there is a combination of three image markers among the M image markers that satisfies the preset azimuth constraint, wherein the combination of three image markers includes 3 image markers;

[0153] If not satisfied, the real-time image information is updated, the image markers are updated and identified, and the combination of the three image markers is iteratively selected to make a judgment facing the azimuth constraint.

[0154] If the conditions are met, triangulation is performed based on the combination of the three image markers, and position verification is performed by comparing the triangulation result with the third inertial navigation position.

[0155] In this embodiment of the application, in order to ensure the reliability of the third inertial navigation position, it is necessary to form a position verification independent of the RFID anchor point through image recognition and triangulation, use multimodal information cross-verification to eliminate the limitations of a single positioning method, and at the same time, use multiple verifications to ensure the accuracy and rationality of the final positioning result.

[0156] First, real-time environmental images of the surrounding environment along the transport robot's path are collected to obtain M image markers.

[0157] Specifically, the transport robot uses its onboard high-definition camera to capture real-time images of the underground environment during its journey. These images include pre-deployed image markers (such as QR code labels affixed to the tunnel walls, geometric patterns coated with fluorescent paint, etc.). These markers have unique visual characteristics and their three-dimensional coordinates have been pre-recorded in the system.

[0158] Furthermore, feature extraction and matching of the acquired images are performed using deep learning-based image recognition algorithms (such as YOLOv5) to identify all image markers in the images, and the number of image markers is counted to obtain the M value. At the same time, the pixel coordinates of each marker in the image and its confidence level are recorded.

[0159] For example, the robot collects an image frame in a certain alleyway and identifies four image markers: marker A (confidence 0.95), marker B (confidence 0.91), marker C (confidence 0.87), and marker D (confidence 0.79). Since the confidence of marker D is lower than 0.8, M is ultimately determined to be 3.

[0160] Further, the value of M is evaluated. Specifically, if M is less than 3, it indicates that there are insufficient visual references available for triangulation, making it impossible to construct an effective spatial positioning relationship. In this case, the robot needs to adjust the camera angle (e.g., increase the shooting range) or shorten the shooting interval (e.g., from 0.5 seconds / frame to 0.2 seconds / frame) to update the real-time image information and re-execute the image marker recognition process until M is greater than or equal to 3.

[0161] For example, if only 2 valid image markers are obtained in the first recognition (M=2), the robot automatically controls the camera focal length to be adjusted from 50mm to 35mm to expand the field of view, and after re-acquiring the image, it recognizes it again, and finally obtains 5 valid image markers (M=5).

[0162] Furthermore, if M is greater than or equal to 3, then the analysis calculates whether there exists a combination of three image markers among the M image markers that satisfies the preset azimuth constraint.

[0163] Among them, the three-image marker combination refers to any three images selected from M images. The azimuth constraint means that the difference in the azimuth angle between any two images in the combination and the robot must be consistent with the difference in the spatial azimuth angle of the three images in the three-dimensional coordinate system in the well, so as to exclude false image marker combinations caused by image noise or misidentification.

[0164] For example, a combination (A, B, C) is selected from the image markers with M=5. The azimuth difference between A and B is calculated to be 29°, the difference between B and C is 46°, and the difference between A and C is 75°. The pre-stored azimuth differences of the three are 30°, 45°, and 75°, respectively, all within the allowable error range of ±2°. Therefore, it is determined that the combination satisfies the azimuth constraint.

[0165] If no three-image marker combination satisfies the azimuth constraint, the real-time image information is updated (e.g., the robot moves to a more well-lit area and takes a new picture), the image markers are re-identified, and different three-image marker combinations are iteratively selected from the updated markers. The azimuth constraint discrimination is repeated until a combination that meets the conditions is found.

[0166] For example, the three sets of image markers initially selected did not meet the constraints. After the robot moved 1 meter, it re-acquired images and identified 6 markers. The combination (B, C, E) was selected again for discrimination, and finally the azimuth constraint was met.

[0167] Furthermore, if there exists a combination of three image markers that satisfies the azimuth constraint, then triangulation is performed based on that combination.

[0168] Specifically, using the pre-stored absolute coordinates of three markers (such as A (850.2, 420.5, -120.3), B (855.1, 422.3, -120.3), and C (852.7, 425.6, -120.3)), and combining the pixel coordinates of each marker in the image, the azimuth angle and distance ratio relative to the robot's camera are calculated. The robot's triangulation result is then calculated using the principle of triangulation (i.e., solving for the coordinates of the spatial points where the three spheres intersect).

[0169] Furthermore, the triangulation results are compared with the third inertial navigation position to perform position verification.

[0170] In the method provided in this application embodiment, the step of "comparing the triangulation result with the third inertial navigation position to perform position verification" includes:

[0171] Based on the preset downhole boundary constraints, the first position verification is performed on the third inertial navigation position;

[0172] Based on preset downhole boundary constraints, a second position verification is performed on the triangulation results;

[0173] If both the first and second position verification results are passed, the third inertial navigation position is compared with the triangulation result to calculate the relative distance verification value, and the position verification is determined in conjunction with the preset position accuracy constraints.

[0174] In this embodiment of the application, in order to ensure the rationality and accuracy of the final positioning result, multiple verifications are required to gradually screen the valid results and avoid affecting the robot operation due to incorrect positioning information.

[0175] Specifically, the first position verification is performed on the third inertial navigation position based on the preset downhole boundary constraints.

[0176] Among them, the underground boundary constraints are three-dimensional coordinate thresholds that are pre-set based on the actual size and spatial range of the roadway, including x-axis boundaries (e.g., 800.0m~950.0m), y-axis boundaries (e.g., 400.0m~450.0m), and z-axis boundaries (e.g., -130.0m~-110.0m). These thresholds cover all legal operating areas of the robot, and positions outside these ranges are considered physically impossible outliers.

[0177] For example, if the third inertial navigation position is (852.8m, 422.9m, -120.3m), and it is compared with the boundaries of the x-axis (800.0≤852.8≤950.0), y-axis (400.0≤422.9≤450.0), and z-axis (-130.0≤-120.3≤-110.0), it is within the boundary constraint range. Therefore, the verification result of the first position is passed.

[0178] Furthermore, a second position verification is performed on the triangulation results based on the same downhole boundary constraints.

[0179] The verification logic of the second position verification is the same as that of the first position verification, which determines whether the triangulation result falls within the preset alleyway space range, thereby eliminating the situation where the positioning point goes out of bounds due to misidentification of the image.

[0180] For example, the triangulation result is (853.0m, 423.0m, -120.3m). After boundary comparison, the x-axis (800.0≤853.0≤950.0), y-axis (400.0≤423.0≤450.0), and z-axis (-130.0≤-120.3≤-110.0) are all within the boundary constraints. Therefore, the second position verification result is passed.

[0181] Furthermore, if both the first and second position verification results are passed, the next step of accuracy verification is performed, which involves comparing the third inertial navigation position with the triangulation positioning result to calculate the relative distance verification value between the two.

[0182] The relative distance verification value is calculated using a three-dimensional spatial distance formula, specifically the formula: ", (x1, y1, z1) is the third inertial navigation position, and (x2, y2, z2) is the triangulation result.

[0183] For example, if the third inertial navigation position is (852.8m, 422.9m, -120.3m), the corresponding triangulation result is (853.0m, 423.0m, -120.3m). Substituting these values ​​into the formula, the relative distance verification value between the two is approximately 0.22m.

[0184] Simultaneously, position verification and discrimination are performed in conjunction with preset position accuracy constraints.

[0185] Among them, the position accuracy constraint is the maximum allowable deviation threshold (such as 0.5m) set according to the robot's operation requirements. If the relative distance verification value is less than or equal to the preset position accuracy constraint, it is determined that the two are in good agreement and the position verification is passed; otherwise, it is considered that there is a significant deviation and the trajectory recalibration process needs to be triggered.

[0186] For example, the relative distance verification value of 0.22m is less than the preset position accuracy constraint of 0.5m, so the final position verification result is passed, and the third inertial navigation position (852.8m, 422.9m, -120.3m) is confirmed as a valid positioning result.

[0187] Conversely, if the first or second position verification fails (e.g., the third inertial navigation position is (790.0m, 420.0m, -120.0m), exceeding the lower limit of the x-axis), the corresponding position is determined to be an outlier, the result must be discarded, and the repositioning process must be triggered, such as reacquiring inertial navigation data or image information.

[0188] In addition, if the first or second location verification passes but the relative distance verification value exceeds the limit (e.g., 1.2m > 0.5m), it indicates that there is an unacceptable deviation between the two positioning methods. It is necessary to further investigate the source of error by combining historical trajectory trends, such as checking whether the RFID anchor point is offset or whether the image marker is blurry.

[0189] This step first verifies the physical rationality of the location, then checks the consistency of the accuracy of both, gradually filtering out reliable positioning results. This not only avoids interference from outliers, but also improves the reliability of positioning accuracy through cross-validation of multimodal information, providing the final guarantee for the precise operation of underground coal mine transport robots.

[0190] S150: If the position verification result is successful, the third inertial navigation position is output as the final positioning result of the transport robot.

[0191] In this embodiment of the application, in the complex environment of underground coal mines, the dual guarantee of multi-anchor point fitting correction and image recognition verification is used. When the position verification result is passed, it indicates that the third inertial navigation position not only conforms to the underground physical space constraints, but also highly matches the visual positioning result. At this time, it is used as the final positioning result output, which can ensure that the transport robot obtains a high-precision and reliable position reference.

[0192] Specifically, when the position verification result is determined to be passed, that is, both the third inertial navigation position and the triangulation positioning result pass the boundary constraint verification, and the relative distance verification value between the two is ≤ the preset position accuracy constraint threshold.

[0193] First, the coordinate format of the third inertial navigation system position is standardized and converted into a unified expression in the downhole global coordinate system to ensure data format consistency with other modules (such as the dispatch center and path planning module).

[0194] Meanwhile, to improve the reliability of the data, the final positioning results are continuously compared with historical trajectories to check for any abnormalities such as positional jumps or non-kinematic patterns.

[0195] For example, the theoretical position at the current moment is predicted by a Kalman filter. If the deviation between the third inertial navigation position and the predicted position is within a reasonable range (e.g., ≤0.3m), it is confirmed that it meets the requirements of motion continuity.

[0196] Furthermore, the third inertial navigation position is converted into a standard data frame, which includes information such as timestamp, coordinate values, and data source. This data is then transmitted in real time to the navigation control module of the transport robot via the underground industrial wireless communication network, and simultaneously synchronized to the ground dispatch center, providing precise location information for the robot's path planning and task execution.

[0197] This step, through rigorous result verification, format standardization, continuity verification, and synchronous output, ensures the accuracy, reliability, and usability of the final positioning results.

[0198] Ultimately, in the complex environment of underground coal mines, the output multimodal fusion positioning results can effectively avoid the impact of single sensor failure or environmental interference, providing a key guarantee for the safe operation of transport robots. It is suitable for scenarios with high positioning accuracy requirements, such as intersections of underground coal mine roadways and equipment docking areas.

[0199] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:

[0200] This application proposes a multimodal fusion perception method for precise positioning of an underground auxiliary transportation robot in coal mines. First, the inertial navigation data of the transportation robot is acquired. Through integration and filtering optimization, a first inertial navigation trajectory is constructed, and the first inertial navigation position is determined, achieving continuous trajectory construction even in the absence of external positioning signals underground. Next, RFID tag information is collected in real time, relative positions are calculated, and absolute coordinates are retrieved. Using the task starting point as a constant point, follow-up error correction is performed to generate a second inertial navigation trajectory and position, quickly eliminating accumulated errors in inertial navigation. Based on periodic correction constraints, a multi-anchor sampling window is defined, an anchor reference set is selected, and position variance and multi-anchor evaluation functions are constructed. Through scaling and rotation scheme cross-variation and iterative optimization, a third inertial navigation trajectory and position are generated, achieving globally optimal positioning. Image recognition of markers is collected, and the third inertial navigation position is verified through triangulation. Finally, the verified third inertial navigation position is output. Cross-validation of multimodal information improves the reliability of positioning accuracy.

[0201] The method provided in this application, through the technical solution of "inertial navigation trajectory construction - RFID follow-up correction - multi-anchor point global optimization - visual positioning verification - result output", integrates the advantages of inertial navigation continuity, RFID absolute constraint and visual geometric verification, and solves the problem of traditional single sensor being susceptible to interference and large cumulative error in underground, thus realizing accurate positioning of auxiliary transportation robots in the complex environment of underground coal mines.

[0202] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0203] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0204] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for precise positioning of an underground auxiliary transport robot in coal mines using multimodal fusion perception, characterized in that, include: Acquire the inertial navigation data of the transport robot, construct a first inertial navigation trajectory based on the inertial navigation data, and determine the first inertial navigation position; The tag information of RFID anchor points identified by the transport robot along the task path is updated in real time, and the first inertial navigation trajectory is corrected for follow-up error based on the tag information to generate a second inertial navigation trajectory and the corresponding second inertial navigation position. Based on preset periodic correction constraints, the second inertial navigation trajectory is fitted and corrected using multiple anchor points to generate a third inertial navigation trajectory and determine the position of the third inertial navigation system. The interactive transport robot collects real-time image information and verifies the position of the third inertial navigation system based on the image recognition results; If the location verification result is successful, the third inertial navigation position will be output as the final positioning result of the transport robot. Specifically, based on preset periodic correction constraints, multi-anchor point position fitting correction is performed on the second inertial navigation trajectory to generate a third inertial navigation trajectory and determine the third inertial navigation position, including: Based on the aforementioned periodic correction constraint, a multi-anchor sampling window is defined; Based on the multi-anchor sampling window, multiple RFID anchor points are selected as a multi-anchor reference set; Construct a multi-anchor point evaluation function based on the sum of position variances, and use the multi-anchor point evaluation function as the objective function to perform multi-anchor point position fitting correction on the second inertial navigation trajectory; The process includes, prior to, constructing a multi-anchor-point evaluation function based on the sum of position variances, and using the multi-anchor-point evaluation function as the objective function to perform multi-anchor-point position fitting correction on the second inertial navigation trajectory: Based on the multiple follow-up error correction results, multiple scaling and rotation schemes are extracted accordingly; Random cross-mutation is performed on multiple scaling and rotation schemes; The parameterized random crossover mutation results are compared with multiple scaling and rotation schemes, and the fitting correction scheme set is initialized based on the parameterized results; The process includes constructing a multi-anchor-point evaluation function based on the sum of position variances, and using this multi-anchor-point evaluation function as the objective function to perform multi-anchor-point position fitting correction on the second inertial navigation trajectory, including: Using the set of fitted correction schemes as the initial solution set, the second inertial navigation trajectory is updated accordingly; The objective function value of the updated result is updated by traversing the evaluation trajectory according to the multi-anchor point evaluation function; Based on preset iterative constraints, the fitting correction scheme set is iteratively updated in combination with the objective function value; Select the optimal objective function value from all trajectory update results and output it as the second inertial navigation trajectory.

2. The method as described in claim 1, characterized in that, Acquiring inertial navigation data of the transport robot, constructing a first inertial navigation trajectory based on the inertial navigation data, and determining the first inertial navigation position, includes: The inertial navigation data is obtained from the inertial measurement unit of the transport robot, wherein the inertial navigation data includes at least acceleration data and angular velocity data; The acceleration data is integrated to obtain multi-axial velocity information, and the multi-axial velocity information is further integrated to generate preliminary position information; Attitude estimation is performed based on the angular velocity data to determine the spatial attitude information of the transport robot. The initial position information of the transport robot is then corrected based on the spatial attitude information to obtain the first inertial navigation trajectory and determine the position of the first inertial navigation system.

3. The method as described in claim 2, characterized in that, The system continuously updates the tag information of RFID anchor points identified by the transport robot along the task path, and corrects the follow-up error of the first inertial navigation trajectory based on the tag information to generate a second inertial navigation trajectory and the corresponding second inertial navigation position, including: Based on the tag information, the relative position information between the transport robot and the RFID anchor point is calculated, wherein the relative position information includes the relative distance and the relative azimuth angle; Based on the tag information, the absolute position coordinates of the RFID anchor point are retrieved accordingly; Using the starting point of the transport robot's task as a constant point and minimizing the relative distance as the objective, the first inertial navigation trajectory is subjected to a follow-up error correction based on scaling and rotation. The output follow-up error correction result is the second inertial navigation trajectory, and the corresponding second inertial navigation position is extracted.

4. The method as described in claim 3, characterized in that, The interactive transport robot collects real-time image information and verifies the position of the third inertial navigation system based on the image recognition results, including: Acquire images and perform image marker recognition to obtain M image markers; If M is less than 3, then update the real-time image information and update the image markers; If M is greater than or equal to 3, then analyze and calculate whether there is a combination of three image markers among the M image markers that satisfies the preset azimuth constraint, wherein the combination of three image markers includes 3 image markers; If not satisfied, the real-time image information is updated, the image markers are updated and identified, and the combination of the three image markers is iteratively selected to make a judgment facing the azimuth constraint. If the conditions are met, triangulation is performed based on the combination of the three image markers, and the triangulation result is compared with the third inertial navigation position to perform position verification.

5. The method as described in claim 4, characterized in that, Comparing the triangulation results with the third inertial navigation position to perform position verification includes: Based on the preset downhole boundary constraints, the first position verification is performed on the third inertial navigation position; Based on preset downhole boundary constraints, a second position verification is performed on the triangulation results; If both the first and second position verification results are passed, the third inertial navigation position is compared with the triangulation result to calculate the relative distance verification value, and the position verification is determined in conjunction with the preset position accuracy constraints.

Images