A robot obstacle-crossing posture adjustment method, system and computer program

Abstract

The present invention discloses a robot obstacle-crossing posture adjustment method, system and computer program. The robot includes a flip arm and a detection device. The method includes: obtaining the obstacle data ahead; according to the obstacle data, using the Bayesian filtering method or the probability evaluation model to obtain the probability that the robot can pass; comparing the probability that the robot can pass with the adaptive probability threshold to determine whether the robot can pass; if it is determined that the robot can pass, calibrating the robot posture according to the obstacle data. The technical solution provided by the present invention obtains the obstacle data ahead and calculates the probability of passing. If it can pass, the robot performs navigation positioning, moves to the obstacle crossing point, and then adjusts the robot's orientation and flip arm angle to achieve the posture adjustment of the robot before crossing the obstacle, thereby improving the reliability and stability of the robot's autonomous obstacle crossing; it can be applied to the robot passing through various types of obstacles such as stairs, slopes, and gullies, and has a wide range of adaptability and strong expansibility.

Description

Technical Field

[0001] The present invention relates to the field of robot obstacle crossing, and in particular to a robot obstacle crossing posture adjustment method, system and computer program. Background Art

[0002] like Figure 1 The mobile robot with a flip arm structure shown has certain abilities of climbing slopes, crossing ditches, overcoming obstacles, and going up and down stairs, as well as motion stability, and can be widely used in complex terrain, anti-terrorism and explosion prevention, or space exploration.

[0003] However, existing robots generally adopt fixed obstacle crossing strategies in obstacle crossing control and lack the ability to adaptively adjust obstacle crossing posture according to obstacle type and environmental conditions, which limits their obstacle crossing performance and application scope and lacks wide adaptability to different environments and tasks. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the current technology, the present invention provides a robot obstacle-crossing posture adjustment method, which obtains the obstacle data ahead and calculates the pass probability. If the robot can pass, the robot posture is calibrated according to the obstacle data.

[0005] To achieve the above object, the embodiments of the present invention adopt the following technical solutions:

[0006] A method for adjusting the posture of a robot overcoming obstacles, wherein the robot comprises a flip arm and a detection device, and comprises the following steps:

[0007] The detection device obtains the point cloud data in front of the robot, performs identification and judgment based on the point cloud data, and obtains the obstacle data in front;

[0008] According to the obstacle data, the probability that the robot can pass is obtained by using the Bayesian filtering method or the probability evaluation model;

[0009] Compare the probability that the robot can pass with the adaptive probability threshold to determine whether the robot can pass;

[0010] If it is determined that the robot can pass, the robot's position and posture are calibrated according to the obstacle data;

[0011] Wherein, calibrating the robot posture according to the obstacle data includes:

[0012] Get the robot coordinate system and camera coordinate system;

[0013] Convert the robot coordinate system to the camera coordinate system;

[0014] The robot performs navigation and positioning according to the coordinate system and moves to the starting point of the obstacle crossing;

[0015] Update obstacle data, and adjust the robot's posture according to the current obstacle data so that the robot's orientation is aligned with the center normal of the obstacle;

[0016] Update obstacle data, and calculate the flip arm adjustment angle ɑ required for the robot to overcome obstacles based on the current obstacle data;

[0017] According to the angle ɑ, adjust the robot flip arm.

[0018] According to one aspect of the present invention, the obstacle data includes: obstacle type, obstacle location and obstacle geometric parameters; the obstacle type includes: stairs, slopes, and gullies.

[0019] According to one aspect of the present invention, obtaining the probability that the robot can pass through the obstacle data includes:

[0020] If the obstacle type is stairs or slopes, the robot is initially judged whether it can pass based on the obstacle geometry parameters and robot parameters;

[0021] The Bayesian filtering method is used to perform dynamic probability updates to obtain the probability P that the robot can pass.

[0022] According to one aspect of the present invention, obtaining the probability that the robot can pass through the obstacle data includes:

[0023] If the obstacle type is a gully, the dynamic safety factor K is obtained based on the obstacle geometry parameters and the robot's maximum traversability.

[0024] Based on the robot parameters and dynamic safety factor K, make a preliminary judgment on whether the robot can pass;

[0025] According to the preset probability evaluation model and dynamic safety factor K, the probability P of the robot being passable is obtained.

[0026] According to one aspect of the present invention, the robot obstacle-crossing posture adjustment method further includes:

[0027] If the robot is judged to be unable to pass, it will suspend its action and wait for the user's subsequent decision.

[0028] According to one aspect of the present invention, the calculation of the flip arm adjustment angle ɑ required for the robot to overcome the obstacle according to the current obstacle data includes:

[0029] Get the angle between the robot body and the horizontal ground;

[0030] According to the current obstacle data, obtain the inclination angle of the obstacle;

[0031] According to the inclination angle of the obstacle and the angle between the robot body and the horizontal ground, the adjustment angle ɑ of the flip arm is calculated.

[0032] According to one aspect of the present invention, the robot obstacle-crossing posture adjustment method further includes:

[0033] When the robot moves to the obstacle crossing starting point, the probability that the robot can pass is updated in real time and compared with the adaptive probability threshold to determine whether the robot can pass.

[0034] According to one aspect of the present invention, the robot obstacle-crossing posture adjustment method further includes:

[0035] After calibrating the robot's position and posture, the robot will traverse obstacles based on the current obstacle data;

[0036] During the robot's obstacle-crossing process, obstacle data is updated in real time and the robot's position and posture are calibrated.

[0037] According to one aspect of the present invention, during the robot's obstacle traversal process, updating obstacle data in real time and calibrating the robot's position and posture includes:

[0038] For stair obstacles, the robot adopts a hierarchical posture adjustment strategy during the obstacle crossing process;

[0039] For slope obstacles, the robot adopts an inclined posture adjustment strategy during the obstacle traversal process.

[0040] According to one aspect of the present invention, during the robot's obstacle traversal process, updating obstacle data in real time and calibrating the robot's position and posture includes:

[0041] During the robot's obstacle crossing process, the Kalman filter is used to estimate and adjust the flip arm angle in real time.

[0042] A robot obstacle-crossing posture adjustment system, based on the robot obstacle-crossing posture adjustment method as described above, comprises:

[0043] The detection module is used to identify and judge based on the point cloud data to obtain the obstacle data ahead;

[0044] A probability calculation module is used to obtain the probability that the robot can pass through the obstacle data using a Bayesian filtering method or a probability evaluation model;

[0045] A judgment module is used to compare the probability that the robot can pass with the adaptive probability threshold to determine whether the robot can pass;

[0046] The calibration module is used to calibrate the robot's posture according to the obstacle data if it is judged that the robot can pass.

[0047] A computer program, which, when executed, implements the steps of the robot obstacle crossing posture adjustment method as described above.

[0048] Advantages of the present invention:

[0049] The present invention provides a robot obstacle crossing posture adjustment method, which obtains the obstacle data ahead and calculates the pass probability. If the obstacle is passable, the robot performs navigation positioning, moves to the obstacle crossing point, and then adjusts the robot's orientation and flip arm angle to achieve the robot's posture adjustment before the obstacle crossing. Through precise posture calibration and flip arm adjustment, the reliability and stability of the robot's autonomous obstacle crossing are improved.

[0050] This method can be applied to robots passing through various types of obstacles such as stairs, slopes, gullies, etc. It has wide adaptability and strong scalability. During the robot's movement or obstacle crossing, this method detects and updates obstacle data in real time, adjusts the flip arm angle, calibrates the robot's posture, and improves the reliability and stability of the robot's autonomous obstacle crossing. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1 It is a structural schematic diagram of the mobile robot with a flip arm structure according to the present invention;

[0053] Figure 2 This is a flow chart of a robot obstacle-crossing posture adjustment method according to Embodiment 1 of the present invention;

[0054] Figure 3 This is a flowchart of the robot's obstacle-crossing posture adjustment according to the present invention;

[0055] Figure 4 This is a flow chart of a robot obstacle-crossing posture adjustment method according to Embodiment 2 of the present invention;

[0056] Figure 5 This is the posture of the robot before crossing the obstacle described in the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] Embodiment 1

[0059] like Figure 1 As shown, a mobile robot with a flip arm structure 1 and a detection device 2 is provided. The robot has a strong obstacle-crossing capability, a stable control system and a flexible flip arm, and can climb and cross obstacles. The robot obstacle-crossing posture adjustment method described in this embodiment is for such a mobile robot with a flip arm structure.

[0060] like Figure 2 and 3 As shown, a method for adjusting the posture of a robot overcoming obstacles, wherein the robot includes a flip arm and a detection device, includes the following steps:

[0061] S1: The detection device obtains the point cloud data in front of the robot, performs identification and judgment based on the point cloud data, and obtains the obstacle data in front.

[0062] Specifically, the obstacle data includes: obstacle type, obstacle location and obstacle geometric parameters.

[0063] In practical applications, the types of obstacles include stairs, slopes, gullies, etc. In this embodiment, the robot posture is adjusted for the three situations of stairs, slopes and gullies.

[0064] Specifically, the obstacle geometric parameters include the slope of stairs or ramps, the length of stair treads, the height of stair risers, the width of gullies, etc.

[0065] In actual applications, the robot will be equipped with detection devices such as cameras or radars. During the movement, the robot will collect real-time visual images, radar images, point clouds and other data in front of it. The pre-built obstacle detection and recognition model can be used to detect and identify the collected data, obtain the obstacle type, location and geometric parameters of the obstacle in front, so as to facilitate the robot to cross the obstacle in the future.

[0066] Specifically, the obstacle detection and recognition model may be a neural network model based on convolutional neural network or semantic segmentation, etc. By pre-collecting obstacle data, building a sample data set, and training the neural network model, the obstacle detection and recognition model is obtained.

[0067] In practical applications, the geometric parameters of obstacles can be obtained by using point cloud data to identify obstacles.

[0068] S2: Based on the obstacle data, the probability of the robot being able to pass is obtained using the Bayesian filtering method or the probability evaluation model.

[0069] In actual applications, different methods are used to calculate the probability that the robot can pass through different types of obstacles. Specifically, the obstacles processed in this embodiment are mainly divided into two situations: stairs or slopes, and gullies.

[0070] 1. If the obstacle type is stairs or slopes, the steps for calculating the probability of passage include:

[0071] A1: Based on the geometric parameters of the obstacle and the robot parameters, make a preliminary judgment on whether the robot can pass.

[0072] The preliminary judgment in step A1 mainly includes:

[0073] (1) If the slope of the stairs or ramp is greater than the maximum traversable slope of the robot, the robot cannot pass through, and the probability P that the robot can pass through is 0.

[0074] The slope of the stairs , is the height of the stairs, is the width of the stairs.

[0075] (2) If the length of the stair tread is less than 1.5 times the length of the robot, the robot cannot pass through, and the probability P that the robot can pass through is 0.

[0076] (3) If the height of the stair riser is greater than the maximum height that the robot can cross, the robot cannot pass through, and the probability P that the robot can pass through is 0.

[0077] A2: Use the Bayesian filtering method to perform dynamic probability updates and obtain the probability P of the robot being able to pass.

[0078] In dynamic systems, it is often necessary to continuously update the probability distribution of the state based on new observation data. The Bayesian filtering method can be used to perform dynamic probability updates. The core idea of ​​Bayesian filtering is to use Bayes' theorem to combine prior probability and likelihood function to calculate the posterior probability. In specific implementation, first set an initial probability distribution, and then continuously adjust this distribution based on new observation data to obtain the latest state probability.

[0079] Specifically, in this method, the probability update formula is as follows:

[0080]

[0081] P(pass|z t ) indicates that at z t Under the condition, the probability P (i.e., the posterior probability) that the robot can pass; P(z t |pass) means that under the condition that the robot can pass, z tThe probability of occurrence (i.e., likelihood function); P(pass) represents the prior probability that the robot can pass; P(z t ) represents z t Probability of occurrence (usually used as a normalizing factor).

[0082] Dynamic probability update using Bayesian filtering includes the following steps:

[0083] (1) Initialization: Based on the initial conditions or known information of the system, set the prior probability distribution of the state.

[0084] (2) Prediction: Based on the dynamic model of the system, predict the prior probability distribution of the state at the next moment. This usually involves calculating the probability of state transition.

[0085] (3) Update: Using the new observation data, calculate the posterior probability distribution using the above probability update formula.

[0086] (4) Iteration: Use the updated posterior probability as the prior probability at the next moment and repeat the prediction and update steps to achieve continuous estimation of the state.

[0087] 2. If the obstacle type is a gully, the pass probability calculation steps include:

[0088] B1: Obtain the dynamic safety factor K based on the geometric parameters of the obstacle and the maximum traversability of the robot.

[0089] The dynamic safety factor K depends on the gully width, depth, inclination angle and the maximum crossable capacity of the robot, and is calculated using the following formula:

[0090]

[0091] in, for the height of the gully; The maximum height that the robot can cross; is the inclination angle of the gully; The maximum inclination angle that the robot can cross.

[0092] B2: Based on the robot parameters and dynamic safety factor K, make a preliminary judgment on whether the robot can pass.

[0093] The preliminary judgment in step B2 mainly includes:

[0094] (1) Calculate the product A of the dynamic safety factor K and the robot length L;

[0095] (2) Compare the gully widths W and A;

[0096] If W>A, it is considered that the robot cannot pass through the gully, and the probability P of the robot being able to pass is 0;

[0097] If W < A, it is considered that the robot may pass through the gully.

[0098] B3: Obtain the passable probability P of the robot according to the preset probability evaluation model and the dynamic safety factor K.

[0099] The probability evaluation model is usually based on a series of input variables (such as the speed of the robot, the size and shape of the obstacle, the friction coefficient of the ground, etc.) and the influence of these variables on the robot's ability to pass through the obstacle. The dynamic safety factor K may be used as one of the input variables.

[0100] Specifically, the construction of the probability evaluation model includes the following steps:

[0101] (1) Define the input variables:

[0102] Determine all relevant variables that affect the robot's ability to pass through the obstacle, including the physical characteristics of the robot (such as size, weight, speed, acceleration, etc.), the characteristics of the obstacle (such as size, shape, position, material, etc.), the environmental characteristics (such as the friction coefficient of the ground, slope, humidity, etc.), and the dynamic safety factor K.

[0103] (2) Establish the probability evaluation model:

[0104] Use statistical methods or machine learning algorithms (such as logistic regression, decision tree, random forest, neural network, etc.) to establish the relationship between the input variables and the probability of the robot passing through the obstacle.

[0105] (3) Collect data:

[0106] Collect a large amount of data on the robot, obstacle, and environmental characteristics, as well as the results of whether the robot successfully passes through the obstacle. This data will be used to train the probability evaluation model.

[0107] (4) Train the model:

[0108] Use the collected data to train the probability evaluation model.

[0109] During the training process, the model will learn the complex relationship between the input variables and the probability of the robot passing through the obstacle.

[0110] (5) Evaluate the model:

[0111] Use an independent test data set to evaluate the performance of the model.

[0112] Calculate metrics such as the accuracy, recall, and F1 score of the model to ensure that the model has good generalization ability.

[0113] (6) Apply the model:

[0114] Once the model is trained and validated, it can be used to predict the probability P of the robot passing an obstacle given the input variables (including the dynamic safety factor K).

[0115] S3: Compare the probability that the robot can pass with the adaptive probability threshold to determine whether the robot can pass.

[0116] If the probability that the robot can pass is ≥ the adaptive probability threshold, the robot is considered to be passable;

[0117] If the probability that the robot can pass is less than the adaptive probability threshold, it is considered that the robot cannot pass.

[0118] Specifically, an adaptive adjustment mechanism is introduced to dynamically adjust the probability threshold according to different obstacle types and geometric parameters, environmental complexity and historical data to adapt to different environmental conditions and obstacles.

[0119] In practical applications, a multi-factor fusion evaluation framework can be constructed, and the evaluation results of different factors can be integrated using data fusion techniques, such as weighted average and neural network fusion. For example, the robot's power, load, environmental complexity (such as obstacle density, terrain undulation, etc.), and historical passage success rate can be weighted and fused to obtain a comprehensive evaluation value. Then, based on this comprehensive evaluation value, the adaptive probability threshold is dynamically adjusted to achieve precise control of the robot's passage probability. For example, when the environment complexity is high and the robot's power is low, even if the historical passage success rate is high, the probability threshold can be appropriately increased to ensure the robot's safety in the current complex environment and poor conditions.

[0120] Preferably, the robot obstacle-crossing posture adjustment method further comprises:

[0121] If the robot is judged to be unable to pass, it will suspend its action and wait for the user's subsequent decision.

[0122] S4: If it is determined that the robot can pass, the robot posture is calibrated according to the obstacle data.

[0123] The calibrating the robot posture according to the obstacle data includes:

[0124] S41: Obtain the robot coordinate system and the camera coordinate system.

[0125] The robot coordinate system is mainly used to describe the position and posture of the robot. Usually, the origin of the coordinate is the reference point of the robot, the coordinate axis indicates the direction of movement of the robot, and the direction of the coordinate axis determines the orientation of the coordinate system. The most common robot coordinate system is the Cartesian coordinate system, which is a three-dimensional rectangular coordinate system used to describe the position and posture of the robot. In the Cartesian coordinate system, three coordinate axes (usually X, Y, and Z axes) are used to define the position of the robot, and three rotation axes (usually rotation around the X, Y, and Z axes) are used to define the posture of the robot.

[0126] The camera coordinate system is a three-dimensional rectangular coordinate system inside the camera, which is used to describe the image information in the camera sensor. The origin of the camera coordinate system is the optical center of the camera, the X-axis and Y-axis are parallel to the X-axis and Y-axis of the image, and the Z-axis is the optical axis of the camera, which is perpendicular to the image plane. Through the camera coordinate system, the relative position relationship between the camera and the object can be accurately described, thereby realizing functions such as object positioning, tracking and recognition.

[0127] Therefore, the coordinate origin and direction of the robot coordinate system and the camera coordinate system are different. In the robot vision system, the robot coordinate system and the camera coordinate system need to be converted and matched to achieve collaborative work between the robot and the camera.

[0128] S42: Convert the robot coordinate system to the camera coordinate system.

[0129] The coordinate transformation relationship is as follows:

[0130]

[0131] in, are the coordinates of the navigation point in the robot coordinate system; are the coordinates of the navigation point in the camera coordinate system; is the coordinate of the obstacle in the camera coordinate system. If it is a staircase, it is the first step. is the unit normal vector of the vertical plane of the obstacle in the camera coordinate system; d is the distance from the navigation point to the front of the stairs; Represents the transformation matrix from the camera coordinate system to the robot coordinate system.

[0132] To ensure the accuracy of the conversion, calibration can be performed through multiple experiments to optimize the conversion matrix. and the unit normal vector Calculation.

[0133] In practical applications, the obstacle information in the camera coordinate system is updated in real time, and the navigation point position in the robot coordinate system is dynamically adjusted according to the above conversion relationship. By continuously optimizing the conversion process, the navigation accuracy is improved, so that the robot can be positioned more accurately, thereby improving the reliability and stability of autonomous obstacle crossing.

[0134] S43: The robot performs navigation positioning according to the coordinate system and moves to the obstacle crossing starting point.

[0135] Based on the coordinate data obtained in the previous step, the robot performs navigation and positioning and moves to the obstacle crossing starting point.

[0136] Preferably, the robot obstacle-crossing posture adjustment method further comprises:

[0137] When the robot moves to the obstacle crossing starting point, the probability that the robot can pass is updated in real time and compared with the adaptive probability threshold to determine whether the robot can pass.

[0138] S44: Update obstacle data, and adjust the robot posture according to the current obstacle data so that the robot is aligned with the center normal of the obstacle.

[0139] Adjust the robot's posture so that it faces the center normal direction of the stairs, slopes, etc. Figure 5 As shown in the figure. The current point cloud features are obtained, and the robot posture is adjusted in real time to ensure that its orientation is aligned with the center normal of the stairs, slopes, etc. Then, the point cloud data is continuously updated, and the detection and calibration steps are repeated until the robot is accurately facing the target direction. Through this process, the robot is ensured to be in the best posture before crossing the obstacle.

[0140] S45: Update obstacle data, and calculate the flip arm adjustment angle ɑ required for the robot to overcome the obstacle according to the current obstacle data.

[0141] The calculation of the flip arm adjustment angle ɑ required for the robot to overcome the obstacle according to the current obstacle data includes:

[0142] Get the angle between the robot body and the horizontal ground;

[0143] According to the current obstacle data, obtain the inclination angle of the obstacle;

[0144] According to the inclination angle of the obstacle and the angle between the robot body and the horizontal ground, the adjustment angle ɑ of the flip arm is calculated.

[0145] Specifically, according to the updated obstacle point cloud feature data, the required flip arm adjustment angle ɑ is calculated, and the calculation formula of the angle ɑ is as follows:

[0146]

[0147] in, is the inclination angle of obstacles such as stairs and slopes, is the angle between the robot body and the horizontal ground. In practical applications, It can be obtained by fitting the point cloud data of obstacles such as stairs and slopes; It is provided by the robot's own sensors.

[0148] S46: Adjust the robot flip arm according to the angle ɑ.

[0149] During the adjustment process, the flip arm will be adjusted accordingly based on the calculated angle ɑ to ensure that it maintains a suitable contact angle with the surface of obstacles such as stairs and slopes, providing stable support for subsequent obstacle crossing. Through this precise posture calibration and flip arm adjustment, the robot can better adapt to different types of obstacles such as stairs and slopes, and improve the success rate of autonomous obstacle crossing.

[0150] The beneficial effects of this embodiment are as follows: this method obtains the obstacle data ahead and calculates the probability of passing. If the obstacle is passable, the robot performs navigation positioning, moves to the obstacle crossing point, and then adjusts the robot's orientation and flip arm angle to achieve the posture adjustment of the robot before crossing the obstacle; through precise posture calibration and flip arm adjustment, the reliability and stability of the robot's autonomous obstacle crossing are improved. This method can be applied to robots passing through various types of obstacles such as stairs, slopes, gullies, etc., and has wide adaptability and strong scalability.

[0151] Embodiment 2

[0152] like Figure 1 As shown, a mobile robot with a flip arm structure 1 and a detection device 2 is provided. The robot has a strong obstacle-crossing capability, a stable control system and a flexible flip arm, and can climb and cross obstacles. The robot obstacle-crossing posture adjustment method described in this embodiment is for such a mobile robot with a flip arm structure.

[0153] like Figure 3 and 4 As shown, a method for adjusting the posture of a robot overcoming obstacles, wherein the robot includes a flip arm and a camera, includes the following steps:

[0154] S1: The detection device obtains the point cloud data in front of the robot, performs identification and judgment based on the point cloud data, and obtains the obstacle data in front.

[0155] Specifically, the obstacle data includes: obstacle type, obstacle location and obstacle geometric parameters.

[0156] In practical applications, the types of obstacles include stairs, slopes, gullies, etc. In this embodiment, the robot posture is adjusted for the three situations of stairs, slopes and gullies.

[0157] Specifically, the obstacle geometric parameters include the slope of stairs or ramps, the length of stair treads, the height of stair risers, the width of gullies, etc.

[0158] In actual applications, the robot will be equipped with detection devices such as cameras or radars. During the movement, the robot will collect real-time visual images, radar images, point clouds and other data in front of it. The pre-built obstacle detection and recognition model can be used to detect and identify the collected data, obtain the obstacle type, location and geometric parameters of the obstacle in front, so as to facilitate the robot to cross the obstacle in the future.

[0159] Specifically, the obstacle detection and recognition model may be a neural network model based on convolutional neural network or semantic segmentation, etc. By pre-collecting obstacle data, building a sample data set, and training the neural network model, the obstacle detection and recognition model is obtained.

[0160] In practical applications, the geometric parameters of obstacles can be obtained by using point cloud data to identify obstacles.

[0161] S2: Based on the obstacle data, the probability of the robot being able to pass is obtained using the Bayesian filtering method or the probability evaluation model.

[0162] In actual applications, different methods are used to calculate the probability that the robot can pass through different types of obstacles. Specifically, the obstacles processed in this embodiment are mainly divided into two situations: stairs or slopes, and gullies.

[0163] 1. If the obstacle type is stairs or slopes, the steps for calculating the probability of passage include:

[0164] A1: Based on the geometric parameters of the obstacle and the robot parameters, make a preliminary judgment on whether the robot can pass.

[0165] The preliminary judgment in step A1 mainly includes:

[0166] (1) If the slope of the stairs or ramp is greater than the maximum traversable slope of the robot, the robot cannot pass through, and the probability P that the robot can pass through is 0.

[0167] The slope of the stairs , is the height of the stairs, is the width of the stairs.

[0168] (2) If the length of the stair tread is less than 1.5 times the length of the robot, the robot cannot pass through, and the probability P that the robot can pass through is 0.

[0169] (3) If the height of the stair riser is greater than the maximum height that the robot can cross, the robot cannot pass through, and the probability P that the robot can pass through is 0.

[0170] A2: Use the Bayesian filtering method to perform dynamic probability updates and obtain the probability P of the robot being able to pass.

[0171] In dynamic systems, it is often necessary to continuously update the probability distribution of the state based on new observation data. The Bayesian filtering method can be used to perform dynamic probability updates. The core idea of ​​Bayesian filtering is to use Bayes' theorem to combine prior probability and likelihood function to calculate the posterior probability. In specific implementation, first set an initial probability distribution, and then continuously adjust this distribution based on new observation data to obtain the latest state probability.

[0172] Specifically, in this method, the probability update formula is as follows:

[0173]

[0174] P(pass|z t ) indicates that at z t Under the condition, the probability P (i.e., the posterior probability) that the robot can pass; P(z t |pass) means that under the condition that the robot can pass, z t The probability of occurrence (i.e., likelihood function); P(pass) represents the prior probability that the robot can pass; P(z t ) represents z t Probability of occurrence (usually used as a normalizing factor).

[0175] Dynamic probability update using Bayesian filtering includes the following steps:

[0176] (1) Initialization: Based on the initial conditions or known information of the system, set the prior probability distribution of the state.

[0177] (2) Prediction: Based on the dynamic model of the system, predict the prior probability distribution of the state at the next moment. This usually involves calculating the probability of state transition.

[0178] (3) Update: Using the new observation data, calculate the posterior probability distribution using the above probability update formula.

[0179] (4) Iteration: Use the updated posterior probability as the prior probability at the next moment and repeat the prediction and update steps to achieve continuous estimation of the state.

[0180] 2. If the obstacle type is a gully, the pass probability calculation steps include:

[0181] B1: Obtain the dynamic safety factor K based on the geometric parameters of the obstacle and the maximum traversability of the robot.

[0182] The dynamic safety factor K depends on the gully width, depth, inclination angle and the maximum crossable capacity of the robot, and is calculated using the following formula:

[0183]

[0184] Among them, is the height of the gully; is the maximum height that the robot can cross; is the inclination angle of the gully; is the maximum crossing inclination angle of the robot.

[0185] B2: Based on the robot parameters and the dynamic safety factor K, preliminarily judge whether the robot can pass.

[0186] The content of the preliminary judgment in step B2 mainly includes:

[0187] (1) Calculate the product A of the dynamic safety factor K and the robot length L;

[0188] (2) Compare the gully width W and A;

[0189] If W > A, it is considered that the robot cannot pass the gully, and the probability P that the robot can pass is 0;

[0190] If W < A, it is considered that the robot may pass the gully.

[0191] B3: According to the preset probability evaluation model and the dynamic safety factor K, obtain the probability P that the robot can pass.

[0192] The probability evaluation model is usually based on a series of input variables (such as the speed of the robot, the size and shape of the obstacle, the friction coefficient of the ground, etc.) and the influence of these variables on the robot's ability to pass the obstacle. The dynamic safety factor K may be used as one of the input variables.

[0193] Specifically, the construction of the probability evaluation model includes the following steps:

[0194] (1) Define the input variables:

[0195] Determine all relevant variables that affect the robot's ability to pass the obstacle, including the physical characteristics of the robot (such as size, weight, speed, acceleration, etc.), the characteristics of the obstacle (such as size, shape, position, material, etc.), the environmental characteristics (such as the friction coefficient of the ground, slope, humidity, etc.), and the dynamic safety factor K.

[0196] (2) Establish a probability evaluation model:

[0197] Use statistical methods or machine learning algorithms (such as logistic regression, decision tree, random forest, neural network, etc.) to establish the relationship between the input variables and the probability that the robot passes the obstacle.

[0198] (3) Collect data:

[0199] Collect a large amount of data about the characteristics of the robot, obstacles, and environment, as well as the corresponding results of whether the robot successfully passes through the obstacles. This data will be used to train the probabilistic evaluation model.

[0200] (4) Training model:

[0201] Use the collected data to train a probability estimation model.

[0202] During training, the model learns the complex relationship between the input variables and the probability of the robot traversing an obstacle.

[0203] (5) Evaluation model:

[0204] An independent test dataset was used to evaluate the performance of the model.

[0205] Calculate the model's accuracy, recall, F1 score and other indicators to ensure that the model has good generalization ability.

[0206] (6) Application model:

[0207] Once the model is trained and validated, it can be used to predict the probability P of the robot passing an obstacle given the input variables (including the dynamic safety factor K).

[0208] S3: Compare the probability that the robot can pass with the adaptive probability threshold to determine whether the robot can pass.

[0209] If the probability that the robot can pass is ≥ the adaptive probability threshold, the robot is considered to be passable;

[0210] If the probability that the robot can pass is less than the adaptive probability threshold, it is considered that the robot cannot pass.

[0211] Specifically, an adaptive adjustment mechanism is introduced to dynamically adjust the probability threshold according to different obstacle types and geometric parameters, environmental complexity and historical data to adapt to different environmental conditions and obstacles.

[0212] In practical applications, a multi-factor fusion evaluation framework can be constructed, and the evaluation results of different factors can be integrated using data fusion techniques, such as weighted average and neural network fusion. For example, the robot's power, load, environmental complexity (such as obstacle density, terrain undulation, etc.), and historical passage success rate can be weighted and fused to obtain a comprehensive evaluation value. Then, based on this comprehensive evaluation value, the adaptive probability threshold is dynamically adjusted to achieve precise control of the robot's passage probability. For example, when the environment complexity is high and the robot's power is low, even if the historical passage success rate is high, the probability threshold can be appropriately increased to ensure the robot's safety in the current complex environment and poor conditions.

[0213] Preferably, the robot obstacle-crossing posture adjustment method further comprises:

[0214] If the robot is judged to be unable to pass, it will suspend its action and wait for the user's subsequent decision.

[0215] S4: If it is determined that the robot can pass, the robot posture is calibrated according to the obstacle data.

[0216] The calibrating the robot posture according to the obstacle data includes:

[0217] S41: Obtain the robot coordinate system and the camera coordinate system.

[0218] The robot coordinate system is mainly used to describe the position and posture of the robot. Usually, the origin of the coordinate is the reference point of the robot, the coordinate axis indicates the direction of movement of the robot, and the direction of the coordinate axis determines the orientation of the coordinate system. The most common robot coordinate system is the Cartesian coordinate system, which is a three-dimensional rectangular coordinate system used to describe the position and posture of the robot. In the Cartesian coordinate system, three coordinate axes (usually X, Y, and Z axes) are used to define the position of the robot, and three rotation axes (usually rotation around the X, Y, and Z axes) are used to define the posture of the robot.

[0219] The camera coordinate system is a three-dimensional rectangular coordinate system inside the camera, which is used to describe the image information in the camera sensor. The origin of the camera coordinate system is the optical center of the camera, the X-axis and Y-axis are parallel to the X-axis and Y-axis of the image, and the Z-axis is the optical axis of the camera, which is perpendicular to the image plane. Through the camera coordinate system, the relative position relationship between the camera and the object can be accurately described, thereby realizing functions such as object positioning, tracking and recognition.

[0220] Therefore, the coordinate origin and direction of the robot coordinate system and the camera coordinate system are different. In the robot vision system, the robot coordinate system and the camera coordinate system need to be converted and matched to achieve collaborative work between the robot and the camera.

[0221] S42: Convert the robot coordinate system to the camera coordinate system.

[0222] The coordinate transformation relationship is as follows:

[0223]

[0224] in, are the coordinates of the navigation point in the robot coordinate system; are the coordinates of the navigation point in the camera coordinate system; is the coordinate of the obstacle in the camera coordinate system. If it is a staircase, it is the first step. is the unit normal vector of the vertical plane of the obstacle in the camera coordinate system; d is the distance from the navigation point to the front of the stairs; Represents the transformation matrix from the camera coordinate system to the robot coordinate system.

[0225] To ensure the accuracy of the conversion, calibration can be performed through multiple experiments to optimize the conversion matrix. and the unit normal vector Calculation.

[0226] In practical applications, the obstacle information in the camera coordinate system is updated in real time, and the navigation point position in the robot coordinate system is dynamically adjusted according to the above conversion relationship. By continuously optimizing the conversion process, the navigation accuracy is improved, so that the robot can be positioned more accurately, thereby improving the reliability and stability of autonomous obstacle crossing.

[0227] S43: The robot performs navigation positioning according to the coordinate system and moves to the obstacle crossing starting point.

[0228] Based on the coordinate data obtained in the previous step, the robot performs navigation and positioning and moves to the obstacle crossing starting point.

[0229] Preferably, the robot obstacle-crossing posture adjustment method further comprises:

[0230] When the robot moves to the obstacle crossing starting point, the probability that the robot can pass is updated in real time and compared with the adaptive probability threshold to determine whether the robot can pass.

[0231] S44: Update obstacle data, and adjust the robot posture according to the current obstacle data so that the robot is aligned with the center normal of the obstacle.

[0232] Adjust the robot's posture so that it faces the center normal direction of the stairs, slopes, etc. Figure 5 As shown in the figure. The current point cloud features are obtained, and the robot posture is adjusted in real time to ensure that its orientation is aligned with the center normal of the stairs, slopes, etc. Then, the point cloud data is continuously updated, and the detection and calibration steps are repeated until the robot is accurately facing the target direction. Through this process, the robot is ensured to be in the best posture before crossing the obstacle.

[0233] S45: Update obstacle data, and calculate the flip arm adjustment angle ɑ required for the robot to overcome the obstacle according to the current obstacle data.

[0234] The calculation of the flip arm adjustment angle ɑ required for the robot to overcome the obstacle according to the current obstacle data includes:

[0235] Get the angle between the robot body and the horizontal ground;

[0236] According to the current obstacle data, obtain the inclination angle of the obstacle;

[0237] According to the inclination angle of the obstacle and the angle between the robot body and the horizontal ground, the adjustment angle ɑ of the flip arm is calculated.

[0238] Specifically, according to the updated obstacle point cloud feature data, the required flip arm adjustment angle ɑ is calculated, and the calculation formula of the angle ɑ is as follows:

[0239]

[0240] in, is the inclination angle of obstacles such as stairs and slopes, is the angle between the robot body and the horizontal ground. In practical applications, It can be obtained by fitting the point cloud data of obstacles such as stairs and slopes; It is provided by the robot's own sensors.

[0241] S45: Adjust the robot flip arm according to the angle ɑ.

[0242] During the adjustment process, the flip arm will be adjusted accordingly based on the calculated angle ɑ to ensure that it maintains a suitable contact angle with the surface of obstacles such as stairs and slopes, providing stable support for subsequent obstacle crossing. Through this precise posture calibration and flip arm adjustment, the robot can better adapt to different types of obstacles such as stairs and slopes, and improve the success rate of autonomous obstacle crossing.

[0243] S5: After calibrating the robot's position and posture, the robot traverses obstacles based on the current obstacle data.

[0244] Specifically, the robot can drive active devices such as flip arms and tracks to perform obstacle crossing operations according to the type and geometric parameters of the obstacle.

[0245] S6: During the robot’s obstacle-crossing process, obstacle data is updated in real time and the robot’s position and posture are calibrated.

[0246] Specifically, the method of adjusting the robot posture in real time in step S6 includes:

[0247] (1) For stair obstacles, a layered posture adjustment strategy is adopted during the robot's obstacle traversal process.

[0248] First, according to the step height of the stairs and the size of the robot, the relative height of the robot and the plane of the stairs is adjusted so that the robot can smoothly place the flip arm on the first step of the stairs.

[0249] Then, according to the inclination angle of the stairs and the position of the center of gravity of the robot, the relative angle between the robot and the plane of the stairs is adjusted to maintain the stability of the robot during the climbing process.

[0250] Next, the angle of the flip arm is adjusted floor by floor to match the plane of the current stair steps. At the same time, the robot's forward direction is adjusted according to the width of the stairs and the size of the robot so that the robot can move along the center line of the stairs and avoid collision with the edge of the stairs.

[0251] On each flight of stairs, repeat the above adjustment process until the robot successfully crosses the entire stair obstacle.

[0252] (2) For slope obstacles, the robot adopts a tilt posture adjustment strategy during the obstacle traversal process.

[0253] According to the slope of the slope and the position of the robot's center of gravity, adjust the robot's inclination angle so that the robot's center of gravity line is as close as possible to the center normal direction of the slope to improve the robot's stability on the slope.

[0254] At the same time, according to the slope of the slope and the support requirements of the robot, the support angle of the flip arm is adjusted so that it can provide sufficient support force for the robot to prevent the robot from sliding or tipping over on the slope.

[0255] During the adjustment process, the contact between the robot and the slope and the movement status of the robot are monitored in real time, and the robot's tilt angle and the support angle of the flip arm are dynamically fine-tuned to adapt to subtle changes in the slope and the robot's movement requirements.

[0256] Preferably, during the process of the robot overcoming obstacles, the Kalman filter can also be used to estimate and adjust the flip arm angle in real time. The obstacle data is updated in real time, and the flip arm angle is adjusted according to the latest point cloud data features of the obstacle. In this way, the estimation effect is continuously optimized, so that the robot can successfully pass the obstacle.

[0257] The specific method of using Kalman filter to estimate the flip arm angle is as follows:

[0258] When adjusting the angle of the flip arm, and the inclination angle of the obstacle Considered as a state variable, predict the next moment angle of the flip arm:

[0259]

[0260] Prediction covariance matrix:

[0261]

[0262] Where P is the state covariance matrix and Q is the process noise covariance.

[0263] Then according to the new observation Update the flip arm angle estimate:

[0264]

[0265]

[0266]

[0267] in, is the Kalman filter, H is the measurement matrix, and R is the measurement noise covariance matrix.

[0268] Through the Kalman filter, the robot can integrate real-time sensor data and dynamically adjust the angle of the flip arm to cope with changes in obstacles such as stairs and slopes.

[0269] The beneficial effect of this embodiment is that: during the movement or obstacle crossing of the robot, this method detects and updates obstacle data in real time, adjusts the angle of the flip arm, calibrates the robot's posture, and improves the reliability and stability of the robot's autonomous obstacle crossing.

[0270] Embodiment 3

[0271] A robot obstacle-crossing posture adjustment system, based on the robot obstacle-crossing posture adjustment method as described in Embodiment 1 or 2, comprises:

[0272] The detection module is used to identify and judge the point cloud data to obtain the obstacle data ahead;

[0273] A probability calculation module is used to obtain the probability that the robot can pass through the obstacle data using a Bayesian filtering method or a probability evaluation model;

[0274] A judgment module is used to compare the probability that the robot can pass with the adaptive probability threshold to determine whether the robot can pass;

[0275] The calibration module is used to calibrate the robot's posture according to the obstacle data if it is judged that the robot can pass.

[0276] Preferably, the system further comprises:

[0277] The obstacle crossing module is used to calibrate the robot's position and then enable the robot to cross obstacles based on the current obstacle data.

[0278] Embodiment 4

[0279] A computer program, when executed, implements the steps of the robot obstacle crossing posture adjustment method as described in embodiment one or two.

[0280] Embodiment 5

[0281] A readable storage medium having stored thereon a computer program as described in Example 5, wherein when the computer program is executed, the steps of the robot obstacle crossing posture adjustment method as described in Example 1 or 2 are implemented.

[0282] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the art within the technical scope disclosed in the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for adjusting the posture of a robot when overcoming obstacles, the robot comprising a flip arm and a detection device, characterized in that: The following steps are involved: The detection device obtains the point cloud data in front of the robot, performs identification and judgment based on the point cloud data, and obtains the obstacle data in front; According to the obstacle data, the probability that the robot can pass is obtained by using the Bayesian filtering method or the probability evaluation model; Compare the probability that the robot can pass with the adaptive probability threshold to determine whether the robot can pass; If it is determined that the robot can pass, the robot's position and posture are calibrated according to the obstacle data; Wherein, calibrating the robot posture according to the obstacle data includes: Get the robot coordinate system and camera coordinate system; Convert the robot coordinate system to the camera coordinate system; The robot performs navigation and positioning according to the robot coordinate system and moves to the obstacle crossing starting point; Update obstacle data, and adjust the robot's posture according to the current obstacle data so that the robot's orientation is aligned with the center normal of the obstacle; Update obstacle data, and calculate the flip arm adjustment angle ɑ required for the robot to overcome obstacles based on the current obstacle data; According to the angle ɑ, adjust the robot flip arm.

2. The robot obstacle-crossing posture adjustment method according to claim 1, characterized in that: The obstacle data includes: obstacle type, obstacle location and obstacle geometric parameters; the obstacle type includes: stairs and slopes.

3. The robot obstacle-crossing posture adjustment method according to claim 2, characterized in that: The method of obtaining the probability that the robot can pass through the obstacle data includes: If the obstacle type is stairs or slopes, the robot is initially judged whether it can pass based on the obstacle geometry parameters and robot parameters; The Bayesian filtering method is used to perform dynamic probability updates to obtain the probability P that the robot can pass.

4. The robot obstacle-crossing posture adjustment method according to claim 1, characterized in that: The robot obstacle crossing posture adjustment method also includes: If the robot is judged to be unable to pass, it will suspend its action and wait for the user's subsequent decision.

5. The robot obstacle-crossing posture adjustment method according to claim 1, characterized in that: The calculation of the flip arm adjustment angle ɑ required for the robot to overcome the obstacle according to the current obstacle data includes: Get the angle between the robot body and the horizontal ground; Obtain the inclination angle of the obstacle based on the current obstacle data; According to the inclination angle of the obstacle and the angle between the robot body and the horizontal ground, the adjustment angle ɑ of the flip arm is calculated.

6. The robot obstacle-crossing posture adjustment method according to claim 1, characterized in that: The robot obstacle crossing posture adjustment method also includes: When the robot moves to the obstacle crossing starting point, the probability that the robot can pass is updated in real time and compared with the adaptive probability threshold to determine whether the robot can pass.

7. The robot obstacle-crossing posture adjustment method according to claim 1, characterized in that: The robot obstacle crossing posture adjustment method also includes: After calibrating the robot's position and posture, the robot will traverse obstacles based on the current obstacle data; During the robot's obstacle-crossing process, obstacle data is updated in real time and the robot's position and posture are calibrated.

8. The robot obstacle-crossing posture adjustment method according to claim 7, characterized in that: During the robot's obstacle crossing process, updating obstacle data in real time and calibrating the robot's position and posture include: For stair obstacles, the robot adopts a hierarchical posture adjustment strategy during the obstacle crossing process; For slope obstacles, the robot adopts an inclined posture adjustment strategy during the obstacle traversal process.

9. The robot obstacle-crossing posture adjustment method according to claim 7, characterized in that: During the robot's obstacle crossing process, updating obstacle data in real time and calibrating the robot's position and posture include: During the robot's obstacle crossing process, the Kalman filter is used to estimate and adjust the flip arm angle in real time.

10. A robot obstacle-crossing posture adjustment system, characterized in that: The robot obstacle-crossing posture adjustment method according to any one of claims 1 to 9 comprises: The detection module is used to identify and judge based on the point cloud data to obtain the obstacle data ahead; A probability calculation module is used to obtain the probability that the robot can pass through the obstacle data using a Bayesian filtering method or a probability evaluation model; A judgment module is used to compare the probability that the robot can pass with the adaptive probability threshold to determine whether the robot can pass; The calibration module is used to calibrate the robot's position and posture according to the obstacle data if it is determined that the robot can pass; Wherein, calibrating the robot posture according to the obstacle data includes: Get the robot coordinate system and camera coordinate system; Convert the robot coordinate system to the camera coordinate system; The robot performs navigation and positioning according to the robot coordinate system and moves to the obstacle crossing starting point; Update obstacle data, and adjust the robot's posture according to the current obstacle data so that the robot's orientation is aligned with the center normal of the obstacle; Update obstacle data, and calculate the flip arm adjustment angle ɑ required for the robot to overcome obstacles based on the current obstacle data; According to the angle ɑ, adjust the robot flip arm.

11. A computer program, characterized in that When the computer program is executed, the steps of the robot obstacle crossing posture adjustment method as described in any one of claims 1 to 9 are implemented.

Images