Robot obstacle avoidance method and device based on dimension reduction motion parameters, and storage medium

Abstract

The invention relates to the technical field of robot obstacle avoidance, and discloses a robot obstacle avoidance method and device based on dimension reduction motion parameters, and a storage medium, and the method comprises the steps: recording an initial motion track of an obstacle when it is monitored that the obstacle appears in a target working region, and fitting the initial motion track into an initial motion parameter matrix under a robot working coordinate system; acquiring a local base coordinate transformation matrix containing a rotation parameter variable and a non-orthogonal parameter variable; performing multi-round assignment on the rotation parameter variable and the non-orthogonal parameter variable, after each round of assignment, transforming the initial motion parameter matrix into a transformed parameter matrix by using the obtained local base coordinate transformation matrix, performing sparse processing, calculating a sparse parameter matrix, and determining a target rotation parameter and a target non-orthogonal parameter according to the sparse parameter matrix; representing the target rotation parameter, the target non-orthogonal parameter and the corresponding sparse parameter matrix as dimension reduction parameters; and representing a motion track of the predicted obstacle according to the dimension reduction parameter, and generating a robot avoidance path based on the motion track.

Description

Technical Field

[0001] This application relates to the field of robot obstacle avoidance technology, and in particular to a robot obstacle avoidance method, device, and storage medium based on dimensionality-reduced motion parameters. Background Technology

[0002] As industrial robots develop towards higher speed, higher precision, and greater intelligence, their working areas are gradually expanding from structured enclosed spaces to semi-structured open spaces. In such environments, obstacles (such as personnel, mobile tooling, and logistics equipment) often dynamically intrude into the robot's working area in unexpected ways. To achieve safe and reliable human-robot collaboration and dynamic obstacle avoidance, robots need to perceive and predict the trajectory of moving obstacles in real time and plan their own avoidance paths based on the prediction results. However, the motion of dynamic obstacles typically exhibits complex characteristics such as non-stationarity, nonlinearity, multimodality, abrupt changes in direction, and local perturbations, posing a severe challenge to the robot's real-time perception and prediction.

[0003] Existing trajectory modeling and representation methods in robot obstacle avoidance systems mainly fall into two categories: one is to fit the three-dimensional trajectory of obstacles to a high-dimensional function in the robot's working coordinate system, using fixed basis functions such as polynomials, trigonometric functions, and convolution kernels to describe the trajectory shape; the other is a trajectory representation method based on occupying a grid or high-dimensional latent space, which characterizes complex motion by constructing a high-dimensional feature space. The former has high parameter redundancy, unclear geometric meaning, and high-frequency perturbations and directional jumps are difficult to represent with finite parameters, directly affecting the accuracy and reliability of robot obstacle avoidance path planning; the latter, although capable of describing complex motion, has extremely high feature dimensions and heavy computational burden, making it difficult to meet the stringent requirements of real-time response in robot obstacle avoidance.

[0004] Furthermore, existing methods typically pass high-dimensional parameters directly to the robot's motion planning module after trajectory modeling, which leads to computational resource constraints for the robot when dealing with multi-obstacle scenarios, making it difficult to achieve efficient and safe dynamic obstacle avoidance. Summary of the Invention

[0005] In view of this, this application provides a robot obstacle avoidance method, device, and storage medium based on dimensionality-reduced motion parameters. By introducing a dynamic local base coordinate system containing rotation parameters and non-orthogonal parameters, the coordinate system can adaptively align with the main motion direction of the obstacle trajectory. This transforms the complex motion trajectory, which is difficult to align in the fixed robot working coordinate system, into a sparse parameter matrix with highly concentrated energy in the local base coordinate system. This effectively overcomes the problems of high redundancy of fitting parameters and unclear geometric meaning caused by the misalignment of the coordinate system and trajectory direction in traditional methods. At the same time, sparsification processing discards a large number of minor components that contribute little to the trajectory description, significantly reducing the dimensionality of the motion parameters. On this basis, the target rotation parameters, target non-orthogonal parameters, and sparse parameter matrix are used together as dimensionality-reduced parameter representations. This allows the robot motion planning module to avoid directly processing high-dimensional redundant parameters, instead performing trajectory prediction and obstacle avoidance path generation based on the simplified low-dimensional features. This significantly reduces the computational burden in multi-obstacle scenarios, meets the stringent requirements of real-time response for robot obstacle avoidance, and achieves efficient and safe dynamic obstacle avoidance while improving the accuracy and reliability of obstacle avoidance path planning.

[0006] According to one aspect of this application, a robot obstacle avoidance method based on dimensionality-reduced motion parameters is provided, comprising: The robot monitors the target working area in real time using its onboard vision sensors. When an obstacle is detected in the target working area, the robot records the initial motion trajectory of the obstacle in three-dimensional space and fits the initial motion trajectory into the initial motion parameter matrix in the robot's working coordinate system. Obtain a local base coordinate transformation matrix that includes rotational parameter variables and non-orthogonal parameter variables, wherein the rotational parameter variables are used to characterize the rotational attitude from the working coordinate system to the local base coordinate system, and the non-orthogonal parameter variables are used to characterize the angle between two coordinate axes in the local base coordinate system; The rotation parameter variables and the non-orthogonal parameter variables are assigned values ​​in multiple rounds. After each round of assignment, the initial motion parameter matrix is ​​transformed into a transformed parameter matrix in the local base coordinate system using the obtained local base coordinate transformation matrix. The transformed parameter matrix is ​​then sparsified to obtain a sparse parameter matrix. The reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory is calculated. The assignment corresponding to the minimum reconstruction error in the multiple rounds of assignment is taken as the target rotation parameter and the target non-orthogonal parameter. The target rotation parameters, the target non-orthogonal parameters, and the corresponding sparse parameter matrix are used as the dimensionality-reduced parameter representation of the initial motion parameter matrix; Based on the dimensionality reduction parameters, the motion trajectory of the obstacle is predicted, and based on the predicted motion trajectory, an avoidance path for the robot is generated, and the robot is controlled to move along the avoidance path.

[0007] According to another aspect of this application, a robot obstacle avoidance device based on reduced-dimensional motion parameters is provided, comprising: The fitting module is used to monitor the target working area in real time through the vision sensor on the robot. When an obstacle is detected in the target working area, the module records the initial motion trajectory of the obstacle in three-dimensional space and fits the initial motion trajectory into the initial motion parameter matrix in the working coordinate system of the robot. The matrix acquisition module is used to acquire a local base coordinate transformation matrix containing rotation parameter variables and non-orthogonal parameter variables, wherein the rotation parameter variables are used to characterize the rotation attitude from the working coordinate system to the local base coordinate system, and the non-orthogonal parameter variables are used to characterize the angle between two coordinate axes in the local base coordinate system. The sparse processing module is used to assign values ​​to the rotation parameter variables and the non-orthogonal parameter variables in multiple rounds. After each round of assignment, the initial motion parameter matrix is ​​transformed into a transformed parameter matrix in the local base coordinate system using the obtained local base coordinate transformation matrix. The transformed parameter matrix is ​​then sparsified to obtain a sparse parameter matrix. The reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory is calculated. The assignment corresponding to the minimum reconstruction error in the multiple rounds of assignment is used as the target rotation parameter and the target non-orthogonal parameter. The dimension reduction parameter determination module is used to use the target rotation parameters, the target non-orthogonal parameters, and the corresponding sparse parameter matrix as the dimension reduction parameter representation of the initial motion parameter matrix; The obstacle avoidance module is used to predict the motion trajectory of the obstacle based on the dimensionality reduction parameter representation, generate an avoidance path for the robot based on the predicted motion trajectory, and control the robot to move along the avoidance path.

[0008] According to another aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described robot obstacle avoidance method based on dimensionality-reduced motion parameters.

[0009] By employing the aforementioned technical solutions, this application provides a robot obstacle avoidance method, device, and storage medium based on dimensionality-reduced motion parameters. By introducing a dynamic local base coordinate system containing rotation parameters and non-orthogonal parameters, the coordinate system can adaptively align with the main motion direction of the obstacle trajectory. This transforms complex motion trajectories, which are difficult to align in a fixed robot working coordinate system, into a sparse parameter matrix with highly concentrated energy in the local base coordinate system. This effectively overcomes the problems of high redundancy and unclear geometric meaning of fitting parameters caused by the misalignment of the coordinate system and trajectory direction in traditional methods. Simultaneously, sparsification processing discards a large number of minor components that contribute little to the trajectory description, significantly reducing the dimensionality of the motion parameters. Furthermore, by using the target rotation parameters, target non-orthogonal parameters, and the sparse parameter matrix together as dimensionality-reduced parameter representations, the robot motion planning module does not need to directly process high-dimensional redundant parameters. Instead, it performs trajectory prediction and obstacle avoidance path generation based on the simplified low-dimensional features, significantly reducing the computational burden in multi-obstacle scenarios and meeting the stringent requirements of real-time response in robot obstacle avoidance. This improves the accuracy and reliability of obstacle avoidance path planning while achieving efficient and safe dynamic obstacle avoidance.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a robot obstacle avoidance method based on reduced motion parameters provided in an embodiment of this application is shown. Figure 2 A schematic diagram of a robot obstacle avoidance device based on reduced-dimensional motion parameters provided in an embodiment of this application is shown. Figure 3 A schematic diagram of the device structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation

[0012] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0013] This embodiment provides a robot obstacle avoidance method based on dimensionality-reduced motion parameters, such as... Figure 1 As shown, the method includes: Step 101: The target working area is monitored in real time by the vision sensor on the robot. When an obstacle is detected in the target working area, the initial motion trajectory of the obstacle in three-dimensional space is recorded, and the initial motion trajectory is fitted into the initial motion parameter matrix in the working coordinate system of the robot.

[0014] Step 102: Obtain the local base coordinate transformation matrix containing rotation parameter variables and non-orthogonal parameter variables, wherein the rotation parameter variables are used to characterize the rotation attitude from the working coordinate system to the local base coordinate system, and the non-orthogonal parameter variables are used to characterize the angle between two coordinate axes in the local base coordinate system.

[0015] Step 103: Assign values ​​to the rotation parameter variable and the non-orthogonal parameter variable in multiple rounds. After each round of assignment, transform the initial motion parameter matrix into a transformed parameter matrix in the local base coordinate system using the obtained local base coordinate transformation matrix. Perform sparsification on the transformed parameter matrix to obtain a sparse parameter matrix. Calculate the reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory. Use the assignment corresponding to the minimum reconstruction error in the multiple rounds of assignment as the target rotation parameter and the target non-orthogonal parameter.

[0016] Step 104: Use the target rotation parameters, the target non-orthogonal parameters, and the corresponding sparse parameter matrix as the dimensionality-reduced parameter representation of the initial motion parameter matrix; Step 105: Based on the dimensionality reduction parameter representation, predict the motion trajectory of the obstacle, generate an avoidance path for the robot based on the predicted motion trajectory, and control the robot to move along the avoidance path.

[0017] This application provides a robot obstacle avoidance method based on dimensionality-reduced motion parameters. When the robot performs a task, it first continuously monitors the target work area using its onboard visual sensors. These visual sensors can be depth cameras or LiDAR, capable of acquiring real-time depth information of the surrounding environment. When the visual sensors detect an obstacle (such as a person or moving tool) entering the target work area, a recording program is immediately initiated to record the obstacle's continuous position information over a period of time, forming a complete initial motion trajectory. Here, the working coordinate system can be a coordinate system with the robot base as the origin, the X-axis pointing forward, and the Z-axis vertically upward. Then, a predefined set of basis functions is used to fit the initial motion trajectory to obtain an initial motion parameter matrix. The basis function set can include various basic function forms such as polynomials, trigonometric functions, and exponential functions, covering complex behaviors that obstacles may exhibit, such as smooth motion, periodic oscillations, and rapid changes. The initial motion parameter matrix is ​​essentially the coefficient matrix of these basis functions, which completely preserves the geometric information of the obstacle's motion trajectory, but has a high parameter dimension and some redundancy, providing the original data foundation for subsequent dimensionality reduction processing.

[0018] To more concisely represent the initial motion parameter matrix of the obstacle, this embodiment introduces a dynamically adjustable local base coordinate system. This coordinate system is defined by two sets of parameters: rotation parameter variables and non-orthogonal parameter variables. The rotation parameter variables include three rotation angles around the X, Y, and Z axes, which determine how to rotate the robot's working coordinate system to a new orientation, ensuring that the coordinate axes of the new coordinate system are aligned as closely as possible with the main direction of the obstacle's motion. The non-orthogonal parameter variables further break the rigid constraint that coordinate axes in the traditional Cartesian coordinate system must be perpendicular, allowing a variable angle between the two coordinate axes in the local base coordinate system. This enables the coordinate system to more flexibly conform to the actual geometry of the obstacle's trajectory. For example, when the obstacle's trajectory exhibits obvious tilting or skew characteristics, the non-orthogonal parameters allow the coordinate system to "deform" to adapt to this shape. The local base coordinate transformation matrix, constructed by these two sets of parameters, effectively defines a viewing angle that can adaptively adjust according to the obstacle's motion characteristics, laying the foundation for subsequently converting the motion parameters from a redundant original representation to a sparse representation.

[0019] After determining the form of the local base coordinate system, an optimal set of rotation and non-orthogonal parameters can be found through optimization search. Specifically, multiple sets of different parameter assignments can be tried. For each set of assignments, the initial motion parameter matrix is ​​first transformed to the local base coordinate system defined by the corresponding local base coordinate transformation matrix, resulting in a transformed parameter matrix. At this point, since the orientation and shape of the coordinate system are roughly aligned with the main motion direction of the obstacle, the energy in the transformed parameter matrix will be concentrated in a few basis functions. Next, sparsification is performed to obtain a sparse parameter matrix. This sparse parameter matrix actually only retains the motion component with the most concentrated energy in the obstacle trajectory, while filtering out secondary components such as noise and high-frequency disturbances. Then, based on the sparse parameter matrix and the set of basis functions, a motion trajectory can be reconstructed and compared with the initial motion trajectory to calculate the reconstruction error. By traversing multiple sets of parameter assignments, the set of rotation and non-orthogonal parameters that minimizes the reconstruction error is finally selected as the target parameters. This ensures both the accuracy of the trajectory description and achieves a high degree of sparsity in the motion parameters.

[0020] After the optimization search described above, the obtained target rotation parameters, target non-orthogonal parameters, and corresponding sparse parameter matrices together constitute the dimensionality-reduced parameter representation of the initial motion parameter matrix. Compared to the dozens or even hundreds of non-zero coefficients that may be contained in the initial motion parameter matrix, the dimensionality-reduced sparse parameter matrix retains only a few main components, while the rotation parameters and non-orthogonal parameters give these components clear geometric meanings. For example, the main direction of obstacle motion and the degree of trajectory skewness. This representation compresses the originally high-dimensional and redundant motion parameters into low-dimensional, physically meaningful feature vectors, allowing subsequent trajectory prediction and obstacle avoidance planning to process only a very small amount of data, without having to deal with the original high-dimensional parameter matrix.

[0021] After obtaining the dimensionality-reduced parameter representation, these simplified parameters can be used to efficiently predict the future trajectory of obstacles. Since the sparse parameter matrix retains only the main motion components of the obstacle, combined with the basis function set, the position sequence of the obstacle at several time points can be quickly reconstructed. The computational cost of this reconstruction process is far less than that of directly using the original high-dimensional initial motion parameter matrix for prediction. Once the predicted trajectory of the obstacle is obtained, the robot's motion planning module can determine the future spatial occupancy of the obstacle and plan an avoidance path that can safely avoid the obstacle. Finally, the underlying motion controller drives the robot to move along this path, realizing a complete dynamic obstacle avoidance closed loop.

[0022] By applying the technical solution of this embodiment, a dynamic local base coordinate system containing rotation parameters and non-orthogonal parameters is introduced, enabling the coordinate system to adaptively align with the main motion direction of the obstacle trajectory. This transforms complex motion trajectories that are difficult to align in the fixed robot working coordinate system into a sparse parameter matrix with highly concentrated energy in the local base coordinate system. This effectively overcomes the problems of high redundancy of fitting parameters and unclear geometric meaning caused by the misalignment of the coordinate system and trajectory direction in traditional methods. At the same time, sparsification processing discards a large number of minor components that contribute little to the trajectory description, significantly reducing the dimensionality of the motion parameters. On this basis, the target rotation parameters, target non-orthogonal parameters, and sparse parameter matrix are used together as dimensionality-reduced parameter representations. This allows the robot motion planning module to avoid directly processing high-dimensional redundant parameters, instead performing trajectory prediction and obstacle avoidance path generation based on the simplified low-dimensional features. This significantly reduces the computational burden in multi-obstacle scenarios, meets the stringent requirements of real-time response for robot obstacle avoidance, and achieves efficient and safe dynamic obstacle avoidance while improving the accuracy and reliability of obstacle avoidance path planning.

[0023] In this embodiment of the application, optionally, the rotation parameter variables include X-axis rotation angle variables, Y-axis rotation angle variables, and Z-axis rotation angle variables; before step 102, the method further includes: constructing corresponding basic rotation matrices based on the X-axis rotation angle variables, Y-axis rotation angle variables, and Z-axis rotation angle variables respectively, and multiplying the basic rotation matrices in sequence to obtain an initial rotation matrix used to describe the attitude adjustment of the robot's working coordinate system; determining two target coordinate axes constrained by the non-orthogonal parameter variables, constructing a non-orthogonal transformation matrix for adapting to the geometric shape of the obstacle's motion trajectory based on the two target coordinate axes and the non-orthogonal parameter variables, and calculating the inverse matrix of the non-orthogonal transformation matrix; multiplying the initial rotation matrix with the inverse matrix to obtain a local base coordinate transformation matrix containing the rotation parameter variables and the non-orthogonal parameter variables.

[0024] In this embodiment, to construct a local base coordinate system that can adaptively align with the main direction of obstacle movement in a robot obstacle avoidance scenario, this embodiment first defines three rotation parameter variables, corresponding to rotation angles around the X, Y, and Z axes of the robot's working coordinate system, respectively. Based on these three rotation angles, respective basic rotation matrices are constructed. These three basic rotation matrices describe independent rotation transformations around the X, Y, and Z axes, respectively. By multiplying the three in a specific order (e.g., ZYX order), an initial rotation matrix can be obtained. The purpose of this initial rotation matrix is ​​to adjust the overall attitude of the robot's working coordinate system, so that the orientation of the coordinate system is roughly aligned with the main direction of obstacle movement. For example, when the obstacle mainly moves diagonally, by selecting an appropriate rotation angle, a certain coordinate axis of the new coordinate system can be made to point to that direction of movement, thereby creating favorable conditions for subsequent sparsification of motion parameters.

[0025] Building upon the coordinate system attitude adjustment, non-orthogonal parameter variables are introduced to break the rigid constraint that coordinate axes in the traditional Cartesian coordinate system must be perpendicular. Specifically, the two target coordinate axes constrained by the non-orthogonal parameter variables are first determined. Then, a non-orthogonal transformation matrix is ​​constructed based on these two target coordinate axes and the non-orthogonal parameter variables. This matrix deviates the standard 90-degree angle between the two target coordinate axes, creating a variable angle. When the non-orthogonal parameter is zero, the matrix degenerates into a standard orthogonal basis; when the non-orthogonal parameter is not zero, a tilt is generated between the two target coordinate axes of the local base coordinate system, allowing the coordinate system to more flexibly conform to the actual geometry of the obstacle trajectory. For example, when the obstacle's trajectory exhibits obvious skewness or asymmetry, this non-orthogonal deformation allows the coordinate system to better encompass the trajectory shape. Since the initial rotation matrix needs to be transformed from the robot's working coordinate system to this non-orthogonal local base coordinate system, the inverse of the non-orthogonal transformation matrix can be obtained to ensure the reversibility and lossless information of the coordinate transformation.

[0026] Subsequently, multiplying the initial rotation matrix by the inverse of the nonorthogonal transformation matrix yields the complete local base coordinate transformation matrix. This composite matrix integrates two transformations: first, the inverse of the nonorthogonal transformation matrix adjusts the angular relationship between the coordinate axes to a nonorthogonal form that adapts to the obstacle's trajectory; second, the initial rotation matrix adjusts the orientation of the entire coordinate system to align with the obstacle's main motion direction. The superposition of these two transformations allows the final local base coordinate system to achieve both arbitrary orientational rotation adjustments and to overcome the limitations of traditional orthogonal coordinate systems, enabling flexible deformation between coordinate axes. This provides rich degrees of freedom for subsequent motion parameter transformations, ensuring that the energy of the initial motion parameter matrix, after transformation to this local base coordinate system, is highly concentrated on a few basis functions, thus laying the foundation for sparsity processing. It is important to note that the local base coordinate transformation matrix here contains rotation parameter variables and nonorthogonal parameter variables, awaiting subsequent assignment.

[0027] This application's embodiments, by constructing and combining rotation parameters and non-orthogonal parameters to form a local base coordinate transformation matrix, offer significant technical advantages compared to traditional methods that only fit motion parameters in a fixed orthogonal coordinate system. On one hand, the rotation parameter variables allow the coordinate system to adaptively adjust its orientation according to the main motion direction of the obstacle, effectively solving the parameter redundancy problem caused by the difficulty in aligning the working coordinate system with the main motion direction of the trajectory in traditional methods. On the other hand, the non-orthogonal parameter variables break the rigid constraint that coordinate axes must be perpendicular, enabling the coordinate system to adapt to various complex geometric shapes of obstacle trajectories, including irregular features such as tilt and skew. This dual-degree-of-freedom design allows the motion parameters to achieve high sparsity after transformation, significantly reducing the computational burden of subsequent trajectory prediction and obstacle avoidance planning. Simultaneously, the entire transformation process maintains reversibility, ensuring that information is not lost during the transformation, providing a solid mathematical foundation for robots to achieve efficient and reliable dynamic obstacle avoidance in complex environments.

[0028] In one specific embodiment, the pose of any rigid body in three-dimensional space can be represented using Euler angles. This embodiment uses a ZYX sequence to rotate the working coordinate system, with the X-axis, Y-axis, and Z-axis corresponding to the following three rotation angle variables: .

[0029] The three basic rotation matrices constructed are as follows: , , ; Initial rotation matrix R Multiplying in order yields: ; This initial rotation matrix performs a rigid body rotation on the initial motion parameter matrix, providing more degrees of freedom for subsequent sparsification.

[0030] In addition to attitude rotation, the embodiments of this application also introduce an additional degree of freedom, namely, non-orthogonal parameter variables. This is used to adjust the angle between two target coordinate axes (here, the X-axis and the Y-axis) in the local base coordinate system, so that they deviate from an orthogonal relationship.

[0031] Define the nonorthogonal transformation matrix of the XY plane as: ; when When, the matrix is ​​transformed into an orthonormal basis; when At that time, the local base coordinate system is non-orthogonal.

[0032] Since the adjusted basis is used throughout the transformation, it is necessary to invert the non-orthogonal transformation matrix: ; The inverse matrix maps the initial motion parameter matrix to the deformed nonorthogonal basis.

[0033] The final local base coordinate transformation matrix consists of two parts: ① the initial rotation matrix and ② the inverse of the non-orthogonal transformation matrix; therefore, the local base coordinate transformation matrix is: .

[0034] This application embodiment, through the separate design and combination of rotational parameter variables and non-orthogonal parameter variables, enables the coordinate system to achieve arbitrary attitude adjustment and breaks through the limitations of traditional orthogonal coordinate systems. This significantly increases the adaptability of the coordinate system to complex trajectory geometry, not only preserving the reversibility of transformation to ensure information integrity, but also providing more adjustable degrees of freedom for subsequent sparsification optimization. Thus, it can achieve a higher parameter compression ratio while ensuring reconstruction accuracy.

[0035] In this embodiment of the application, optionally, step 103, "transforming the initial motion parameter matrix into a transformed parameter matrix in the local base coordinate system using the obtained local base coordinate transformation matrix, and performing sparsification processing on the transformed parameter matrix to obtain a sparse parameter matrix," includes: multiplying the obtained local base coordinate transformation matrix with the initial motion parameter matrix in the working coordinate system, transforming the initial motion parameter matrix to a local base coordinate system aligned with the main motion direction of the obstacle, to obtain a transformed parameter matrix; for each element in the transformed parameter matrix, retaining elements whose absolute value exceeds a preset threshold, and setting the remaining elements to zero, to obtain a sparse parameter matrix, wherein the sparse parameter matrix is ​​used to filter out motion components that significantly contribute to the description of obstacle motion.

[0036] In this embodiment, in the robot obstacle avoidance scenario, after obtaining the local base coordinate transformation matrix, the initial motion parameter matrix can be transformed into the local base coordinate system defined by this matrix. Specifically, the local base coordinate transformation matrix and the initial motion parameter matrix are multiplied. Through this multiplication, the obstacle motion parameters originally described in the robot's working coordinate system are projected as a whole into the local base coordinate system. Since the rotation parameters and non-orthogonal parameters of the local base coordinate system have been optimized to be roughly aligned with the main motion direction of the obstacle, the coefficient distribution in the transformed parameter matrix will change significantly: the values ​​originally scattered on various basis functions will concentrate on a few basis functions related to the main motion direction, while the coefficients on most basis functions that do not match the motion direction will approach zero. This energy concentration process lays an ideal foundation for subsequent sparsity processing.

[0037] After obtaining the transformed parameter matrix, it is sparsified to filter out motion components that significantly contribute to the description of obstacle motion. Specifically, a threshold can be preset, and then each element in the transformed parameter matrix is ​​examined one by one. Elements with absolute values ​​exceeding the preset threshold indicate that the corresponding motion component dominates the description of the obstacle trajectory and are therefore retained; while elements with absolute values ​​less than or equal to the preset threshold are directly set to zero, because the motion components corresponding to these elements contribute weakly to the trajectory description and may originate from sensor noise, high-frequency disturbances, or minor components unrelated to the main motion direction. Through this hard threshold truncation method, the originally dense transformed parameter matrix is ​​transformed into a sparse parameter matrix, which retains only the core parameters that carry the main motion information of the obstacle. The data dimension of this sparse parameter matrix is ​​much lower than that of the initial motion parameter matrix, but it can still accurately describe the macroscopic motion characteristics of the obstacle.

[0038] This application's embodiments achieve adaptive compression of motion parameters through coordinate system transformation and preset threshold truncation. The combination of these two methods significantly reduces the number of effective motion parameters with almost no loss of trajectory reconstruction accuracy. This processing method preserves the macroscopic geometric features of the initial trajectory while effectively filtering out local disturbances and measurement noise, providing a simpler, more stable, and computationally efficient input of motion parameters for subsequent trajectory prediction. Specifically, the sparse parameter matrix retains only a small number of core parameters, significantly reducing the computational load required for subsequent trajectory prediction and obstacle avoidance path generation, enabling the robot to maintain real-time response capabilities in multi-obstacle scenarios. Simultaneously, because the sparsification process effectively filters out sensor noise and high-frequency disturbances in the trajectory, the motion trajectory reconstructed based on the sparse parameter matrix is ​​smoother and more stable, providing a more reliable data foundation for obstacle avoidance path planning, thereby improving the robot's operational efficiency while ensuring obstacle avoidance safety.

[0039] Optionally, in this embodiment, step 101, "fitting the initial motion trajectory into an initial motion parameter matrix in the robot's working coordinate system," includes: obtaining the trajectory position points corresponding to multiple sampling time points of the initial motion trajectory; calculating the basis function value vector corresponding to each sampling time point based on a predefined basis function set, wherein the basis function set includes multiple basis functions used to characterize the motion components of obstacles; constructing a design matrix based on the basis function value vectors corresponding to all sampling time points, and constructing an observation matrix based on the trajectory position points corresponding to all sampling time points; substituting the design matrix and the observation matrix into a least squares fitting model to solve for the initial motion parameter matrix.

[0040] In this embodiment, firstly, multiple trajectory position points corresponding to sampling time points are obtained from the initial motion trajectory. These position points constitute a discrete motion description of the obstacle in continuous time.

[0041] Next, based on a predefined set of basis functions, the basis function value vector (composed of the function values ​​corresponding to each basis function in the set) is calculated for each sampling time point. Here, the set of basis functions can contain various basic functions with different mathematical forms, such as polynomials, trigonometric functions, and exponential functions. Each basis function is used to characterize a specific type of obstacle motion component (such as smoothing trend, periodic oscillation, rapid growth, etc.). The calculated basis function value vector essentially maps the time points to the function space spanned by these basis functions.

[0042] Then, a design matrix is ​​constructed based on the basis function value vectors corresponding to all sampling time points, and an observation matrix is ​​constructed based on the trajectory position points corresponding to all sampling time points. Each row of the design matrix corresponds to the basis function value vector of a time point, and each row of the observation matrix corresponds to the three-dimensional position coordinates of the same time point. The construction of these two matrices organizes the originally scattered time series data into a standardized form suitable for mathematical solution and establishes a linear relationship between basis function weights and trajectory positions.

[0043] Finally, the design matrix and observation matrix are substituted into the least squares fitting model for solution. The essence of the least squares method is to find a set of coefficients that minimizes the sum of squared errors between the fitted trajectory and the initial motion trajectory at all time points. The coefficient matrix obtained is the initial motion parameter matrix, which completely preserves all the information of the initial motion trajectory in the basis function space.

[0044] In one specific embodiment, the trajectory of an obstacle in three-dimensional space typically exhibits highly nonlinear and periodic characteristics. To achieve accurate modeling, this embodiment constructs a 15-dimensional basis function set to fully capture complex dynamic behavior. The basis function set is represented as follows: ; Among them, the polynomial term is used to fit a smooth trend; the trigonometric term is used to capture periodic oscillations; the exponential and logarithmic terms are used to characterize rapid growth or saturation behavior; the hyperbolic tangent term is used to simulate an S-shaped transition; and the modulation term is used to express the oscillation of amplitude evolution over time.

[0045] Given this set of basis functions, any three-dimensional trajectory It can be represented as: ; in, The initial motion parameter matrix is ​​3×15 dimensional. It is a set of 15 basis functions with respect to time t.

[0046] In another embodiment, through four geometric parameters Construct the local basis coordinate transformation matrix ,in , , Euler angles about fixed coordinate axes are used to control the attitude of the local base coordinate system; This is used to adjust the non-orthogonal angle between the local X and Y axes, thereby introducing a certain degree of freedom while maintaining reversibility. After the transformation, the initial motion trajectory can be rewritten as: ; in, This is the transformed parameter matrix in the local base coordinate system.

[0047] This application embodiment discretizes the continuous initial motion trajectory into sampling points and maps them to a predefined basis function space, transforming the trajectory description problem into a standardized linear parameter solving problem. This approach not only ensures fitting accuracy but also provides a unified and operable parameterized input for subsequent dynamic coordinate system transformations. At the same time, the diverse design of the basis functions enables this application to adapt to various complex motion patterns, exhibiting strong versatility and adaptability.

[0048] Optionally, in this embodiment, step 103, "calculating the reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory," includes: determining the reconstructed motion trajectory corresponding to the initial motion trajectory based on the sparse parameter matrix and the basis function set, and calculating the initial reconstruction error between the reconstructed motion trajectory and the initial motion trajectory, wherein the initial reconstruction error is used to measure the fitting accuracy of the obstacle trajectory; identifying the number of non-zero elements in the sparse parameter matrix; and calculating the reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory based on the initial reconstruction error and the number of non-zero elements, through a comprehensive objective function, wherein the comprehensive objective function is used to balance the obstacle avoidance accuracy of the robot and the computational complexity of the robot control system. The comprehensive objective function is as follows: ; in, This represents the reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory. This represents the rotation parameter variable. This represents the non-orthogonal parameter variable. This indicates the number of non-zero elements. Represents the sparse parameter matrix, Represents the parameter matrix after transformation The L1 norm, This represents the initial reconstruction error. This indicates the preset error threshold. This indicates the preset weighting coefficient.

[0049] In this embodiment, in the robot obstacle avoidance scenario, after obtaining the sparse parameter matrix, it can be verified whether the matrix can accurately describe the initial motion trajectory of the obstacle. This verification process is accomplished by calculating the reconstruction error. Specifically, the motion trajectory of the obstacle is first reconstructed based on the sparse parameter matrix and a predefined set of basis functions. The set of basis functions includes various basic function forms such as polynomials and trigonometric functions, which together constitute a function space. Each non-zero coefficient in the sparse parameter matrix corresponds to the weight of a certain basis function. By weighting and superimposing these basis functions according to their corresponding coefficients, a complete reconstructed motion trajectory can be obtained. This reconstructed motion trajectory is then compared point by point with the initial motion trajectory actually recorded by the sensor. The positional deviation between the two is calculated, and the average or integral is taken to obtain the initial reconstruction error. This initial reconstruction error directly reflects the accuracy of the current sparse parameter matrix in describing the obstacle trajectory. The smaller the error, the more accurately the dimensionality-reduced parameters can reproduce the actual motion of the obstacle. This is crucial for robot obstacle avoidance because only by accurately perceiving the motion state of the obstacle can the robot plan a reliable avoidance path and avoid collision risks caused by trajectory description deviations.

[0050] When evaluating the dimensionality reduction effect, in addition to focusing on the accuracy of trajectory fitting, the simplicity of the motion parameters can also be considered, as the number of parameters directly determines the computational load required by the robot control system to handle the obstacle. Therefore, it is crucial to identify the number of non-zero elements in the sparse parameter matrix. During the sparsification process, elements in the transformed parameter matrix with absolute values ​​below a preset threshold are set to zero, while the non-zero elements are the core motion components that significantly contribute to the description of obstacle motion. The fewer the number of non-zero elements, the less data the robot control system needs to process in subsequent trajectory prediction and obstacle avoidance path generation, resulting in a lighter computational burden. Especially in complex scenarios where multiple obstacles appear simultaneously, the parameter dimensions of each obstacle accumulate, and controlling the number of non-zero elements directly affects whether the entire robot system can maintain real-time responsiveness.

[0051] To find the optimal balance between trajectory description accuracy and parameter sparsity, this application proposes a comprehensive objective function that combines the initial reconstruction error and the number of non-zero elements to calculate the final reconstruction error. This comprehensive objective function uses a weighted summation approach. The initial reconstruction error term constrains the accuracy of trajectory fitting, ensuring that the parameters after dimensionality reduction do not lose crucial trajectory information due to excessive sparsity. The number of non-zero elements term constrains the sparsity of the parameters, encouraging the optimization process to choose a simpler representation. Furthermore, the comprehensive objective function includes a threshold penalty term for the initial reconstruction error. When the fitting error exceeds a preset error threshold, this term applies an additional penalty to prevent sacrificing excessive trajectory accuracy in the pursuit of sparsity. By adjusting the weighting coefficients, the emphasis on accuracy and real-time performance can be flexibly adjusted according to the specific needs of the robot's obstacle avoidance task. For example, in high-speed motion scenarios, the weight of the sparsity term can be increased to reduce computational latency, while in confined space operations, the weight of the accuracy term can be increased to ensure obstacle avoidance safety.

[0052] This application's embodiments achieve an intelligent balance between obstacle avoidance accuracy and computational complexity through a comprehensive objective function. Compared to traditional methods that either pursue fitting accuracy, leading to parameter redundancy, or oversimplify, resulting in trajectory distortion, this application's embodiments can adaptively find the optimal compromise between the two: on the one hand, by controlling the number of non-zero elements, the computational load in multi-obstacle scenarios is effectively reduced, enabling the robot to track more obstacles simultaneously and maintain real-time response; on the other hand, through error threshold constraints, it is ensured that the trajectory description accuracy always meets the basic requirements of obstacle avoidance path planning, without sacrificing safety due to dimensionality reduction.

[0053] Optionally, before step 103, the method further includes: constructing a set of candidate heuristic algorithms suitable for robot obstacle avoidance scenarios, wherein the set of candidate heuristic algorithms includes multiple heuristic optimization algorithms; for each heuristic optimization algorithm, independently running the heuristic optimization algorithm multiple times, and calculating the corresponding obstacle motion parameter optimization evaluation index vector based on a set of rotation parameters and non-orthogonal parameters output in each run, wherein the obstacle motion parameter optimization evaluation index vector includes the number of non-zero elements in the corresponding sparse parameter matrix, the average reconstruction error between the reconstructed motion trajectory and the initial motion trajectory, the maximum local error between the reconstructed motion trajectory and the initial motion trajectory, and the optimization time of a single run; based on the optimization of all obstacle motion parameters of each heuristic optimization algorithm... An evaluation index vector is generated to construct an original evaluation dataset. Multiple resampled subsets are generated based on the original evaluation dataset using a sampling with replacement method. For each resampled subset, the comprehensive score of each heuristic optimization algorithm is calculated, where the comprehensive score is calculated based on the obstacle motion parameter optimization evaluation index vector. For each resampled subset, the heuristic optimization algorithm with the highest comprehensive score is determined as the optimal candidate algorithm for that subset. The number of times each heuristic optimization algorithm is determined as the optimal candidate algorithm in all resampled subsets is counted, and the heuristic optimization algorithm with the highest number of occurrences is used as the target optimization algorithm for obstacle motion parameter dimensionality reduction. Multiple rounds of assignment are then performed on the rotation parameter variable and the non-orthogonal parameter variable based on the target optimization algorithm.

[0054] In this embodiment, firstly, a set of candidate heuristic algorithms suitable for robot obstacle avoidance scenarios is constructed, which includes a variety of heuristic optimization algorithms that do not rely on gradient information, such as simulated annealing, particle swarm optimization, and genetic algorithms. These algorithms can effectively handle complex optimization problems where the objective function is non-convex, non-smooth, and difficult to differentiate, providing diverse solution tools for subsequent searching for optimal geometric parameters from different perspectives.

[0055] Next, for each heuristic optimization algorithm in the candidate heuristic algorithm set, it is run independently multiple times. Each run outputs a set of solved rotation parameters and non-orthogonal parameters. Based on these parameters, the corresponding obstacle motion parameter optimization evaluation index vector is calculated. This obstacle motion parameter optimization evaluation index vector can contain four key indicators: the number of non-zero elements in the sparse parameter matrix, which characterizes the number of motion parameters required by the robot control system to handle a single obstacle; the average reconstruction error between the reconstructed motion trajectory and the initial motion trajectory, which characterizes the overall fitting accuracy of the obstacle trajectory, and this accuracy directly affects the reliability of the robot's obstacle avoidance path planning; the maximum local error, which characterizes the deviation in the worst case, and is used to determine the minimum safe distance margin for obstacle avoidance; and the optimization time of a single run reflects the computational efficiency of the algorithm, which characterizes the delay from the robot's perception of the obstacle to the completion of parameter dimensionality reduction, and this delay directly affects the real-time performance of the obstacle avoidance response.

[0056] Subsequently, all the results of running all heuristic optimization algorithms are summarized to construct an original evaluation dataset, which contains all obstacle motion parameter optimization evaluation index vectors for each heuristic optimization algorithm under multiple runs.

[0057] Then, sampling with replacement methods such as Bootstrap can be used to randomly select multiple resampled subsets from the original evaluation dataset. For each resampled subset, the comprehensive score of each heuristic optimization algorithm is calculated. This comprehensive score is obtained by calculating the corresponding obstacle motion parameter optimization evaluation index vector.

[0058] In each resampling subset, the heuristic optimization algorithm with the highest comprehensive score is determined as the optimal candidate algorithm for that resampling subset.

[0059] Finally, the number of times each heuristic optimization algorithm was selected as the optimal candidate algorithm in all resampled subsets was counted. The heuristic optimization algorithm with the highest number of occurrences was selected as the final target optimization algorithm for obstacle motion parameter dimensionality reduction. This voting mechanism effectively avoids the randomness brought about by a single data distribution by aggregating the selection results of multiple resampled subsets, making the selected target optimization algorithm have better robustness and generalization ability. Subsequently, this target optimization algorithm will be used to perform multiple rounds of assignment iterations on rotation parameter variables and non-orthogonal parameter variables to solve for the optimal geometric parameters.

[0060] This application embodiment comprehensively evaluates and votes on the results of various heuristic optimization algorithms, automatically selecting the most suitable target optimization algorithm based on the multi-index requirements of the actual problem. At the same time, the resampling technique enhances the robustness of the selection, enabling the finally selected target optimization algorithm to maintain excellent performance under different data perturbations, thereby providing a reliable and efficient solution basis for subsequent motion parameter optimization, and ensuring that the parameter representation after dimensionality reduction can achieve the best balance between sparsity and reconstruction accuracy.

[0061] Optionally, in this embodiment of the application, the step of "calculating the comprehensive score of each heuristic optimization algorithm under each resampled subset" includes: for each resampled subset, determining each heuristic optimization algorithm contained therein; for each heuristic optimization algorithm contained therein, determining the complete obstacle motion parameter optimization evaluation index vector corresponding to the heuristic optimization algorithm from the resampled subset; and calculating the mean of the heuristic optimization algorithm on each evaluation index based on the complete obstacle motion parameter optimization evaluation index vector, wherein the mean is used to characterize the heuristic optimization algorithm. The average performance of the robot in obstacle avoidance scenarios; for each evaluation index, the evaluation index vector is optimized based on the motion parameters of all obstacles in the resampled subset, and the maximum and minimum values ​​of the evaluation index are determined; for each heuristic optimization algorithm, the normalized score of the heuristic optimization algorithm for each evaluation index in the resampled subset is calculated based on the mean of the heuristic optimization algorithm on each evaluation index and the maximum and minimum values, and the normalized scores of each evaluation index are weighted and summed according to the preset weight coefficients to obtain the comprehensive score of the heuristic optimization algorithm in the resampled subset.

[0062] In this embodiment, during the offline calibration phase of the robot obstacle avoidance system, a comprehensive performance evaluation of each algorithm can be performed to select the most suitable heuristic optimization algorithm from multiple candidate heuristic optimization algorithms for the current application scenario. Specifically, for each resampled subset generated by sampling with replacement, the heuristic optimization algorithms included in the subset are first determined. Then, for each algorithm, all obstacle motion parameter optimization evaluation index vectors generated by the algorithm under multiple independent runs are extracted from the resampled subset. By statistically averaging these obstacle motion parameter optimization evaluation index vectors, the mean value of each evaluation index of the algorithm under the resampled subset can be calculated. This mean value actually represents the average performance of the heuristic optimization algorithm in the robot obstacle avoidance scenario. It can effectively filter out the random fluctuations that may occur in a single run and reflect the stability performance level of the algorithm under typical working conditions.

[0063] After obtaining the mean values ​​of each algorithm across various evaluation metrics, a unified comparison benchmark can be established for these metrics. This is because different evaluation metrics have different dimensions and numerical ranges; for example, optimization time might be measured in milliseconds, while reconstruction error might be measured in millimeters, making direct comparison meaningless. Therefore, for each evaluation metric, the evaluation metric vector is optimized based on all obstacle motion parameters corresponding to all algorithms within the resampled subset, determining the maximum and minimum values ​​of that metric across the entire resampled subset.

[0064] After obtaining the mean and extreme values ​​of each algorithm across all evaluation metrics, the comprehensive score for each algorithm in the resampled subset is calculated. First, for each evaluation metric of each algorithm, normalization is performed using the mean, maximum, and minimum values ​​of that metric, converting the original values ​​into a normalized score between 0 and 1. This normalization process allows evaluation metrics with different dimensions and numerical ranges to be compared on the same scale. Subsequently, the normalized scores of each evaluation metric are weighted and summed according to preset weighting coefficients. These weighting coefficients reflect the emphasis placed on different performance dimensions in the robot obstacle avoidance task. For example, if the robot needs to operate in a high-speed dynamic environment with extremely high real-time requirements, the weight of the optimization time metric can be increased accordingly; if the robot's workspace is narrow and obstacle avoidance accuracy is strictly required, the weight of the maximum local error can be increased. The comprehensive score obtained by weighted summation reflects the overall performance of the heuristic optimization algorithm under the resampled subset. The higher the score, the more suitable the algorithm is for optimizing the motion parameters of the robot obstacle avoidance system in the scenario represented by the current resampled subset.

[0065] This application's embodiments achieve robust screening of multiple heuristic optimization algorithms through resampling technology combined with a normalized weighted evaluation mechanism. Compared to simply comparing algorithm performance based on a single run, this application's embodiments generate multiple resampling subsets through sampling with replacement, independently evaluate and select algorithms on each subset, and finally determine the final target optimization algorithm by counting the number of times each algorithm is selected as optimal. This voting mechanism effectively reduces the randomness caused by a single data distribution, making the selected algorithm more robust to different data perturbations. Simultaneously, the normalized weighted evaluation mechanism quantifies the robot obstacle avoidance task's multi-dimensional requirements for accuracy, real-time performance, and safety into calculable weight coefficients, allowing the algorithm selection process to intuitively reflect the priorities in practical engineering applications. This offline calibration process determines the most suitable optimization algorithm for the robot obstacle avoidance system, ensuring that obstacle motion parameters can be reduced with the highest efficiency and optimal performance during online operation, providing a reliable algorithmic foundation for real-time obstacle avoidance in complex dynamic environments.

[0066] Furthermore, as Figure 1 In terms of specific implementation, this application provides a robot obstacle avoidance device based on reduced-dimensional motion parameters, such as... Figure 2 As shown, the device includes: The fitting module is used to monitor the target working area in real time through the vision sensor on the robot. When an obstacle is detected in the target working area, the module records the initial motion trajectory of the obstacle in three-dimensional space and fits the initial motion trajectory into the initial motion parameter matrix in the working coordinate system of the robot. The matrix acquisition module is used to acquire a local base coordinate transformation matrix containing rotation parameter variables and non-orthogonal parameter variables, wherein the rotation parameter variables are used to characterize the rotation attitude from the working coordinate system to the local base coordinate system, and the non-orthogonal parameter variables are used to characterize the angle between two coordinate axes in the local base coordinate system. The sparse processing module is used to assign values ​​to the rotation parameter variables and the non-orthogonal parameter variables in multiple rounds. After each round of assignment, the initial motion parameter matrix is ​​transformed into a transformed parameter matrix in the local base coordinate system using the obtained local base coordinate transformation matrix. The transformed parameter matrix is ​​then sparsified to obtain a sparse parameter matrix. The reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory is calculated. The assignment corresponding to the minimum reconstruction error in the multiple rounds of assignment is used as the target rotation parameter and the target non-orthogonal parameter. The dimension reduction parameter determination module is used to use the target rotation parameters, the target non-orthogonal parameters, and the corresponding sparse parameter matrix as the dimension reduction parameter representation of the initial motion parameter matrix; The obstacle avoidance module is used to predict the motion trajectory of the obstacle based on the dimensionality reduction parameter representation, generate an avoidance path for the robot based on the predicted motion trajectory, and control the robot to move along the avoidance path.

[0067] Optionally, the rotation parameter variables include X-axis rotation angle variables, Y-axis rotation angle variables, and Z-axis rotation angle variables; the device further includes a matrix transformation module: the matrix transformation module is used for: Before obtaining the local base coordinate transformation matrix containing rotation parameter variables and non-orthogonal parameter variables, a corresponding basic rotation matrix is ​​constructed based on the X-axis rotation angle variable, Y-axis rotation angle variable and Z-axis rotation angle variable respectively, and the basic rotation matrices are multiplied in order to obtain the initial rotation matrix used to describe the posture adjustment of the robot's working coordinate system. Determine the two target coordinate axes constrained by the non-orthogonal parameter variables, construct a non-orthogonal transformation matrix to adapt to the geometry of the obstacle's motion trajectory based on the two target coordinate axes and the non-orthogonal parameter variables, and calculate the inverse matrix of the non-orthogonal transformation matrix; Multiplying the initial rotation matrix by the inverse matrix yields a local base coordinate transformation matrix containing rotational and non-orthogonal parameter variables.

[0068] Optionally, the sparse processing module is used for: The obtained local base coordinate transformation matrix is ​​multiplied by the initial motion parameter matrix in the working coordinate system, and the initial motion parameter matrix is ​​transformed to the local base coordinate system aligned with the main motion direction of the obstacle to obtain the transformed parameter matrix. For each element in the transformed parameter matrix, elements whose absolute value exceeds a preset threshold are retained, and the remaining elements are set to zero to obtain a sparse parameter matrix. The sparse parameter matrix is ​​used to filter out motion components that contribute significantly to the description of obstacle motion.

[0069] Optionally, the fitting module is used for: Obtain the trajectory position points corresponding to the initial motion trajectory at multiple sampling time points; Based on a predefined set of basis functions, the basis function value vector corresponding to each sampling time point is calculated, wherein the set of basis functions includes multiple basis functions used to characterize the motion components of the obstacle; Based on the basis function value vectors corresponding to all sampling time points, a design matrix is ​​constructed, and based on the trajectory location points corresponding to all sampling time points, an observation matrix is ​​constructed. Substituting the design matrix and the observation matrix into the least squares fitting model, the initial motion parameter matrix is ​​obtained by solving.

[0070] Optionally, the sparse processing module is used for: Based on the sparse parameter matrix and the basis function set, the reconstructed motion trajectory corresponding to the initial motion trajectory is determined, and the initial reconstruction error between the reconstructed motion trajectory and the initial motion trajectory is calculated, wherein the initial reconstruction error is used to measure the fitting accuracy of the obstacle trajectory; Identify the number of non-zero elements in the sparse parameter matrix; Based on the initial reconstruction error and the number of non-zero elements, the reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory is calculated through a comprehensive objective function. The comprehensive objective function is used to balance the obstacle avoidance accuracy of the robot and the computational complexity of the robot control system. The comprehensive objective function is as follows: ; in, This represents the reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory. This represents the rotation parameter variable. This represents the non-orthogonal parameter variable. This indicates the number of non-zero elements. Represents the sparse parameter matrix, Represents the parameter matrix after transformation The L1 norm, This represents the initial reconstruction error. This indicates the preset error threshold. This indicates the preset weighting coefficient.

[0071] Optionally, the device further includes an algorithm filtering module; the algorithm filtering module is used for: Before assigning values ​​to the rotation parameter variable and the non-orthogonal parameter variable in multiple rounds, a set of candidate heuristic algorithms suitable for robot obstacle avoidance scenarios is constructed, wherein the set of candidate heuristic algorithms includes a variety of heuristic optimization algorithms; For each heuristic optimization algorithm, the heuristic optimization algorithm is run independently multiple times. Based on a set of rotation parameters and non-orthogonal parameters output in each run, the corresponding obstacle motion parameter optimization evaluation index vector is calculated. The obstacle motion parameter optimization evaluation index vector includes the number of non-zero elements in the corresponding sparse parameter matrix, the average reconstruction error between the reconstructed motion trajectory and the initial motion trajectory, the maximum local error between the reconstructed motion trajectory and the initial motion trajectory, and the optimization time of a single run. The evaluation index vector is optimized based on all obstacle motion parameters of each heuristic optimization algorithm to construct the original evaluation dataset; Using a sampling with replacement method, multiple resampled subsets are generated based on the original evaluation dataset; For each resampled subset, a comprehensive score for each heuristic optimization algorithm is calculated for the resampled subset, wherein the comprehensive score is calculated based on the obstacle motion parameter optimization evaluation index vector; For each resampled subset, the heuristic optimization algorithm with the highest comprehensive score is determined as the optimal candidate algorithm for that resampled subset; The number of times each heuristic optimization algorithm was identified as the optimal candidate algorithm in all resampled subsets was counted. The heuristic optimization algorithm with the highest number of times was selected as the target optimization algorithm for dimensionality reduction of obstacle motion parameters. The rotation parameter variable and the non-orthogonal parameter variable were then assigned values ​​in multiple rounds based on the target optimization algorithm.

[0072] Optionally, the algorithm filtering module is further configured to: For each resampled subset, determine each heuristic optimization algorithm contained therein. For each heuristic optimization algorithm contained therein, determine the complete obstacle motion parameter optimization evaluation index vector corresponding to the heuristic optimization algorithm from the resampled subset. Based on the complete obstacle motion parameter optimization evaluation index vector, calculate the mean of the heuristic optimization algorithm on each evaluation index. The mean is used to characterize the average performance of the heuristic optimization algorithm in the robot obstacle avoidance scenario. For each evaluation index, the evaluation index vector is optimized based on all obstacle motion parameters in the resampled subset, and the maximum and minimum values ​​of the evaluation index are determined. For each heuristic optimization algorithm, the normalized score of each evaluation index of the heuristic optimization algorithm in the resampled subset is calculated based on the mean, maximum and minimum values ​​of the heuristic optimization algorithm on each evaluation index. Then, the normalized scores of each evaluation index are weighted and summed according to the preset weight coefficients to obtain the comprehensive score of the heuristic optimization algorithm in the resampled subset.

[0073] It should be noted that other corresponding descriptions of the functional units involved in the robot obstacle avoidance device based on dimensionality-reduced motion parameters provided in this application embodiment can be found in the following references. Figure 1 The corresponding descriptions in the method will not be repeated here.

[0074] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, having stored thereon a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0075] This application also provides a computer device, which may specifically be a personal computer, a server, a network device, etc. Figure 3 As shown, the computer device includes a bus, a processor, memory, and a communication interface, and may also include an input / output interface and a display device. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores location information. The network interface allows communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0076] Those skilled in the art will understand that Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0077] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0078] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0079] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0080] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0081] The embodiments described above are merely examples of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application.

Claims

1. A robot obstacle avoidance method based on dimensionality-reduced motion parameters, characterized in that, include: The robot monitors the target working area in real time using its onboard vision sensors. When an obstacle is detected in the target working area, the robot records the initial motion trajectory of the obstacle in three-dimensional space and fits the initial motion trajectory into the initial motion parameter matrix in the robot's working coordinate system. Obtain a local base coordinate transformation matrix that includes rotational parameter variables and non-orthogonal parameter variables, wherein the rotational parameter variables are used to characterize the rotational attitude from the working coordinate system to the local base coordinate system, and the non-orthogonal parameter variables are used to characterize the angle between two coordinate axes in the local base coordinate system; The rotation parameter variables and the non-orthogonal parameter variables are assigned values ​​in multiple rounds. After each round of assignment, the initial motion parameter matrix is ​​transformed into a transformed parameter matrix in the local base coordinate system using the obtained local base coordinate transformation matrix. The transformed parameter matrix is ​​then sparsified to obtain a sparse parameter matrix. The reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory is calculated. The assignment corresponding to the minimum reconstruction error in the multiple rounds of assignment is taken as the target rotation parameter and the target non-orthogonal parameter. The target rotation parameters, the target non-orthogonal parameters, and the corresponding sparse parameter matrix are used as the dimensionality-reduced parameter representation of the initial motion parameter matrix; Based on the dimensionality reduction parameters, the motion trajectory of the obstacle is predicted, and based on the predicted motion trajectory, an avoidance path for the robot is generated, and the robot is controlled to move along the avoidance path.

2. The method according to claim 1, characterized in that, The rotation parameter variables include X-axis rotation angle variables, Y-axis rotation angle variables, and Z-axis rotation angle variables; before obtaining the local base coordinate transformation matrix containing the rotation parameter variables and non-orthogonal parameter variables, the method further includes: Based on the X-axis rotation angle variable, the Y-axis rotation angle variable, and the Z-axis rotation angle variable, respectively, a basic rotation matrix is ​​constructed, and the basic rotation matrices are multiplied in sequence to obtain an initial rotation matrix used to describe the posture adjustment of the robot's working coordinate system. Determine the two target coordinate axes constrained by the non-orthogonal parameter variables, construct a non-orthogonal transformation matrix to adapt to the geometry of the obstacle's motion trajectory based on the two target coordinate axes and the non-orthogonal parameter variables, and calculate the inverse matrix of the non-orthogonal transformation matrix; Multiplying the initial rotation matrix by the inverse matrix yields a local base coordinate transformation matrix containing rotational and non-orthogonal parameter variables.

3. The method according to claim 1, characterized in that, The process of transforming the initial motion parameter matrix into a transformed parameter matrix in the local base coordinate system using the obtained local base coordinate transformation matrix, and then sparsifying the transformed parameter matrix to obtain a sparse parameter matrix includes: The obtained local base coordinate transformation matrix is ​​multiplied by the initial motion parameter matrix in the working coordinate system, and the initial motion parameter matrix is ​​transformed to the local base coordinate system aligned with the main motion direction of the obstacle to obtain the transformed parameter matrix. For each element in the transformed parameter matrix, elements whose absolute value exceeds a preset threshold are retained, and the remaining elements are set to zero to obtain a sparse parameter matrix. The sparse parameter matrix is ​​used to filter out motion components that contribute significantly to the description of obstacle motion.

4. The method according to claim 1, characterized in that, The step of fitting the initial motion trajectory into an initial motion parameter matrix in the robot's working coordinate system includes: Obtain the trajectory position points corresponding to the initial motion trajectory at multiple sampling time points; Based on a predefined set of basis functions, the basis function value vector corresponding to each sampling time point is calculated, wherein the set of basis functions includes multiple basis functions used to characterize the motion components of the obstacle; Based on the basis function value vectors corresponding to all sampling time points, a design matrix is ​​constructed, and based on the trajectory location points corresponding to all sampling time points, an observation matrix is ​​constructed. Substituting the design matrix and the observation matrix into the least squares fitting model, the initial motion parameter matrix is ​​obtained by solving.

5. The method according to claim 4, characterized in that, The calculation of the reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory includes: Based on the sparse parameter matrix and the basis function set, the reconstructed motion trajectory corresponding to the initial motion trajectory is determined, and the initial reconstruction error between the reconstructed motion trajectory and the initial motion trajectory is calculated, wherein the initial reconstruction error is used to measure the fitting accuracy of the obstacle trajectory; Identify the number of non-zero elements in the sparse parameter matrix; Based on the initial reconstruction error and the number of non-zero elements, the reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory is calculated through a comprehensive objective function. The comprehensive objective function is used to balance the obstacle avoidance accuracy of the robot and the computational complexity of the robot control system. The comprehensive objective function is as follows: ； in, This represents the reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory. This represents the rotation parameter variable. This represents the non-orthogonal parameter variable. This indicates the number of non-zero elements. Represents the sparse parameter matrix, Represents the parameter matrix after transformation The L1 norm, This represents the initial reconstruction error. This indicates the preset error threshold. This indicates the preset weighting coefficient.

6. The method according to claim 1, characterized in that, Before performing multiple rounds of assignment to the rotation parameter variable and the non-orthogonal parameter variable, the method further includes: Construct a set of candidate heuristic algorithms suitable for robot obstacle avoidance scenarios, wherein the set of candidate heuristic algorithms includes a variety of heuristic optimization algorithms; For each heuristic optimization algorithm, the heuristic optimization algorithm is run independently multiple times. Based on a set of rotation parameters and non-orthogonal parameters output in each run, the corresponding obstacle motion parameter optimization evaluation index vector is calculated. The obstacle motion parameter optimization evaluation index vector includes the number of non-zero elements in the corresponding sparse parameter matrix, the average reconstruction error between the reconstructed motion trajectory and the initial motion trajectory, the maximum local error between the reconstructed motion trajectory and the initial motion trajectory, and the optimization time of a single run. The evaluation index vector is optimized based on all obstacle motion parameters of each heuristic optimization algorithm to construct the original evaluation dataset; Using a sampling with replacement method, multiple resampled subsets are generated based on the original evaluation dataset; For each resampled subset, a comprehensive score for each heuristic optimization algorithm is calculated for the resampled subset, wherein the comprehensive score is calculated based on the obstacle motion parameter optimization evaluation index vector; For each resampled subset, the heuristic optimization algorithm with the highest comprehensive score is determined as the optimal candidate algorithm for that resampled subset; The number of times each heuristic optimization algorithm was identified as the optimal candidate algorithm in all resampled subsets was counted. The heuristic optimization algorithm with the highest number of times was selected as the target optimization algorithm for dimensionality reduction of obstacle motion parameters. The rotation parameter variable and the non-orthogonal parameter variable were then assigned values ​​in multiple rounds based on the target optimization algorithm.

7. The method according to claim 6, characterized in that, For each resampled subset, the comprehensive score of each heuristic optimization algorithm under that resampled subset is calculated, including: For each resampled subset, determine each heuristic optimization algorithm contained therein. For each heuristic optimization algorithm contained therein, determine the complete obstacle motion parameter optimization evaluation index vector corresponding to the heuristic optimization algorithm from the resampled subset. Based on the complete obstacle motion parameter optimization evaluation index vector, calculate the mean of the heuristic optimization algorithm on each evaluation index. The mean is used to characterize the average performance of the heuristic optimization algorithm in the robot obstacle avoidance scenario. For each evaluation index, the evaluation index vector is optimized based on all obstacle motion parameters in the resampled subset, and the maximum and minimum values ​​of the evaluation index are determined. For each heuristic optimization algorithm, the normalized score of each evaluation index of the heuristic optimization algorithm in the resampled subset is calculated based on the mean, maximum and minimum values ​​of the heuristic optimization algorithm on each evaluation index. Then, the normalized scores of each evaluation index are weighted and summed according to the preset weight coefficients to obtain the comprehensive score of the heuristic optimization algorithm in the resampled subset.

8. A robot obstacle avoidance device based on reduced-dimensional motion parameters, characterized in that, include: The fitting module is used to monitor the target working area in real time through the vision sensor on the robot. When an obstacle is detected in the target working area, the module records the initial motion trajectory of the obstacle in three-dimensional space and fits the initial motion trajectory into the initial motion parameter matrix in the working coordinate system of the robot. The matrix acquisition module is used to acquire a local base coordinate transformation matrix containing rotation parameter variables and non-orthogonal parameter variables, wherein the rotation parameter variables are used to characterize the rotation attitude from the working coordinate system to the local base coordinate system, and the non-orthogonal parameter variables are used to characterize the angle between two coordinate axes in the local base coordinate system. The sparse processing module is used to assign values ​​to the rotation parameter variables and the non-orthogonal parameter variables in multiple rounds. After each round of assignment, the initial motion parameter matrix is ​​transformed into a transformed parameter matrix in the local base coordinate system using the obtained local base coordinate transformation matrix. The transformed parameter matrix is ​​then sparsified to obtain a sparse parameter matrix. The reconstruction error between the reconstructed motion trajectory corresponding to the sparse parameter matrix and the initial motion trajectory is calculated. The assignment corresponding to the minimum reconstruction error in the multiple rounds of assignment is used as the target rotation parameter and the target non-orthogonal parameter. The dimension reduction parameter determination module is used to use the target rotation parameters, the target non-orthogonal parameters, and the corresponding sparse parameter matrix as the dimension reduction parameter representation of the initial motion parameter matrix; The obstacle avoidance module is used to predict the motion trajectory of the obstacle based on the dimensionality reduction parameter representation, generate an avoidance path for the robot based on the predicted motion trajectory, and control the robot to move along the avoidance path.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images