Charging robot docking path planning method, system, device, medium and product

Abstract

The invention belongs to the technical field of charging robots, and provides a charging robot docking path planning method, system and device, a medium and a product, and the method comprises the steps: fusing multi-modal sensing data, and obtaining a vehicle contour feature vector and a parking space boundary vector; deconstructing a vehicle contour feature vector by using a dynamic space-time modeling algorithm, and constructing a Gaussian process regression model; based on a Gaussian process regression model, parking space geometric constraint conditions are extracted from the enhanced three-dimensional point cloud data, and a pose feasible solution space is constructed; a supporting point set composed of candidate poses is obtained from the pose feasible solution space; constructing a vehicle pose probability distribution model according to the supporting point set; constructing a path optimization objective function according to the working radius of the mechanical arm of the charging robot and the vehicle pose probability distribution model; and solving the trajectory optimization objective function, and determining an optimal docking path of the charging robot. According to the invention, through the multi-modal sensing data and the dynamic space-time modeling algorithm, path planning of accurate docking of the charging robot with the charging port is realized.

Description

Technical Field

[0001] This disclosure belongs to the field of charging robot technology, and in particular relates to a charging robot docking path planning method, system, equipment, medium and product. Background Technology

[0002] Current vehicle position perception and recognition solutions for charging robots suffer from numerous technical bottlenecks, making it difficult to meet the high-precision docking requirements in complex scenarios: 1. The limitations of various sensor types make it difficult to balance environmental adaptability and positioning accuracy. LiDAR is prone to signal scattering and attenuation in adverse weather conditions such as rain, snow, and fog, resulting in distortion of vehicle contour feature extraction; visual cameras are limited by lighting conditions, and the recognition of key features such as parking space boundaries and charging ports is greatly reduced in bright light or dim light environments; millimeter-wave radar can only obtain distance and speed information and cannot provide detailed features of vehicle contours and interfaces.

[0003] 2. The slight displacement of the vehicle when it is parked and the time difference of sensor acquisition are ignored, which causes the constructed vehicle pose model to deviate from the actual state.

[0004] 3. The collision risk along the robotic arm's movement path was not considered.

[0005] The above shortcomings result in problems such as long charging time, high failure rate, and high safety risks when charging robots dock with charging ports, which restricts the promotion and application of automated charging technology. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure provides a path planning method for docking a charging robot. Through multimodal sensing data and dynamic spatiotemporal modeling algorithms, it achieves precise path planning for the charging robot to dock with the charging port. Simultaneously, the path planning takes into account the robot arm's movement radius, mitigating collision risks.

[0007] Firstly, this disclosure provides a method for planning docking paths for a charging robot, including: By fusing multimodal sensor data, vehicle contour feature vectors and parking space boundary vectors are obtained; A Gaussian process regression model is constructed by deconstructing the vehicle contour feature vector using a dynamic spatiotemporal modeling algorithm. Based on the Gaussian process regression model, parking space geometric constraints are extracted from enhanced 3D point cloud data to construct a pose feasible solution space. From the feasible solution space of pose, obtain the set of support points composed of candidate poses; Construct a vehicle pose probability distribution model based on the support point set; Based on the working radius of the charging robot arm and the probability distribution model of the vehicle pose, a path optimization objective function is constructed. Solve the trajectory optimization objective function to determine the optimal docking path for the charging robot.

[0008] Furthermore, By fusing multimodal sensor data, vehicle contour feature vectors and parking space boundary vectors are obtained, including: Acquire multimodal sensing data, including 3D point cloud data and polarization characteristic data; By fusing 3D point cloud data and polarization characteristic data, vehicle contour feature vectors and parking space boundary vectors are obtained.

[0009] Furthermore, By fusing 3D point cloud data and polarization characteristic data, vehicle contour feature vectors and parking space boundary vectors are obtained, including: The three-dimensional point cloud data and polarization characteristic data are preprocessed; after preprocessing, the three-dimensional point cloud data is enhanced according to the polarization characteristic data; the point cloud density gradient features of the enhanced three-dimensional point cloud data are obtained; and the vehicle contour feature vector is obtained based on the point cloud density gradient features.

[0010] Furthermore, Obtain the point cloud density gradient features of the enhanced 3D point cloud data, including: The enhanced 3D point cloud data is divided into voxel grids; within each voxel grid, the number of 3D point cloud data falling into each voxel grid is counted to obtain the point cloud spatial distribution density of each voxel grid; based on the point cloud spatial distribution density, the point cloud density gradient feature is calculated.

[0011] Furthermore, Based on point cloud density gradient features, the vehicle contour feature vector is obtained, including: The non-maximum suppression algorithm is used to scan the point cloud density gradient features of the 3D point cloud data; descriptors are extracted from the 3D point cloud data retained after non-maximum suppression, and the descriptors are encoded into vehicle contour feature vectors.

[0012] Furthermore, The vehicle contour feature vector is deconstructed using a dynamic spatiotemporal modeling algorithm, and a Gaussian process regression model is constructed, including: The vehicle contour feature vector is analyzed into translation and rotation components; Independent Gaussian process regression models are established for the translation and rotation components, respectively.

[0013] Furthermore, From the feasible solution space of poses, construct a set of support points consisting of candidate poses, including: A candidate pose set is generated by sampling within the feasible pose solution space; The Mahalanobis distance is used to measure the degree of matching between each candidate pose and the predicted value of the Gaussian process regression model, and the set of support points is selected.

[0014] Furthermore, Based on the working radius of the charging robot arm and the probability distribution model of the vehicle pose, a path optimization objective function is constructed, including: Calculate the working envelope of the charging robot arm in three-dimensional space; construct a high-pose point cloud subset of the vehicle pose probability distribution model; solve the spatial intersection volume between the working envelope equation of the robot arm and the high-pose point cloud subset of the vehicle pose probability distribution model to obtain the trajectory optimization objective function.

[0015] Secondly, this disclosure provides a charging robot docking path planning system, including a multimodal data fusion module, a regression model construction module, a pose feasible solution space construction module, a support point set acquisition module, a vehicle pose probability distribution model construction module, a path optimization objective function construction module, and a solution module; wherein... The multimodal data fusion module is used to fuse multimodal sensor data to obtain vehicle contour feature vectors and parking space boundary vectors; The regression model building module is used to deconstruct the vehicle contour feature vector using dynamic spatiotemporal modeling algorithms and build a Gaussian process regression model. The pose feasible solution space construction module is used to extract parking space geometric constraints from enhanced 3D point cloud data based on the Gaussian process regression model and construct the pose feasible solution space. The support point set acquisition module is used to acquire a support point set composed of candidate poses from the pose feasible solution space; The vehicle pose probability distribution model construction module is used to construct a vehicle pose probability distribution model based on the support point set. The path optimization objective function construction module is used to construct the path optimization objective function based on the working radius of the charging robot arm and the probability distribution model of the vehicle pose. The solution module is used to solve the trajectory optimization objective function and determine the optimal docking path for the charging robot.

[0016] Thirdly, based on the same inventive concept, this disclosure also provides an electronic device, including at least one processor and at least one storage medium electrically connected; The storage medium is electrically connected to the processor, wherein the storage medium stores instructions executable by at least one of the processors, the instructions being executed by at least one of the processors to enable at least one of the processors to perform the charging robot docking path planning method as described above.

[0017] Fourthly, based on the same inventive concept, this disclosure also provides a computer storage medium storing a computer program. When the computer program is executed by the processor, it implements the charging robot docking path planning method as described above.

[0018] Fifthly, based on the same inventive concept, this disclosure also provides a computer program product, which is stored in at least one storage medium; The computer program product includes several instructions for causing at least one electronic device to execute the charging robot docking path planning method as described above.

[0019] Compared with the prior art, this disclosure provides a docking path planning method for charging robots, which has the following beneficial effects: 1. By incorporating 3D point cloud information, infrared thermal imaging data, and polarization characteristic distribution into the data acquisition range, the vehicle's location can be clearly determined even in low light or inclement weather.

[0020] 2. By using dynamic spatiotemporal modeling algorithms based on the timestamps and spatial coordinates of sensor data, the spatiotemporal alignment of vehicle contour features and parking space geometric constraints is completed. Even if there are slight movements of the vehicle or sensor acquisition delays, the dynamic pose changes of the vehicle can be reflected in real time, ensuring that the two remain consistent.

[0021] 3. By combining the working radius of the robotic arm with the probability distribution model of the vehicle's pose, collision risk prediction can be carried out. It can analyze the spatial overlap area between the robotic arm's motion trajectory and the vehicle when it approaches the vehicle, generate multiple obstacle avoidance trajectory schemes, and then filter the trajectory from dimensions such as path length, motion stability, and docking efficiency through hierarchical optimization to achieve safe and fast positioning and docking of the charging interface.

[0022] Other features and advantages of this disclosure will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the disclosure. The objects and other advantages of this disclosure may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1A flowchart illustrating a docking path planning method for a charging robot according to an embodiment of the present disclosure is shown. Figure 2 A schematic diagram illustrating the structural principle of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0026] like Figure 1 As shown in the figure, a docking path planning method for a charging robot according to an embodiment of this disclosure includes the following steps: S1 integrates multimodal sensor data to obtain vehicle contour feature vectors and parking space boundary vectors.

[0027] S11, acquire multimodal sensing data including 3D point cloud data and polarization characteristic data.

[0028] A multimodal sensor array is set up in the target parking space (vehicle) and its surrounding environment (scene). The multimodal sensor array includes a 3D LiDAR and a polarization camera.

[0029] The point cloud dataset depicts the spatial geometry of the vehicle surface and is generated by 3D LiDAR scanning. Each point cloud data point in the scene contains 3D spatial coordinate information. .

[0030] The polarization characteristic distribution data reflects the degree of linear polarization and polarization angle of each pixel in the scene. It is generated by a polarization camera. Different materials such as metal parts of the vehicle charging interface, plastic body, and glass have significant differences in their response to polarized light, which is reflected in the polarization characteristic distribution data.

[0031] S12, by fusing 3D point cloud data and polarization characteristic data, obtains the vehicle contour feature vector and parking space boundary vector.

[0032] S121 performs preprocessing on 3D point cloud data and polarization characteristic data.

[0033] The 3D point cloud data and polarization characteristic data are subjected to outlier noise removal and line space downsampling.

[0034] This processing is mainly to reduce data density and decrease the amount of subsequent computation.

[0035] S122. After preprocessing, the 3D point cloud data is enhanced based on the polarization characteristic data.

[0036] (1) Using spatial feature encoding, the two-dimensional coordinates of polarization characteristic data are mapped bidirectionally to the point cloud coordinates of three-dimensional point cloud data.

[0037] This involves establishing a unified coordinate reference system to ensure a one-to-one correspondence between 3D point cloud data and polarization characteristic data.

[0038] (2) Match spatial features and use polarization characteristic data as enhanced labels for three-dimensional point cloud data.

[0039] S123, obtain the point cloud density gradient features of the enhanced 3D point cloud data.

[0040] (1) Divide the enhanced 3D point cloud data into voxel meshes.

[0041] The environmental space of the 3D point cloud data is regularly divided into a series of tiny cubic units, namely voxels, with each voxel being an independent unit.

[0042] (2) Within the divided voxel grid, count the number of 3D point cloud data falling into each voxel grid to obtain the point cloud spatial distribution density of each voxel grid.

[0043] (3) Calculate the density gradient features of point cloud based on the spatial distribution density of point cloud.

[0044] Based on the difference in point cloud spatial distribution density between each voxel grid and its adjacent voxel grids, convolution operations are performed using gradient operators in three-dimensional space to obtain the rate of change of point cloud spatial distribution density in the three coordinate axes, i.e., the point cloud density gradient feature.

[0045] Point cloud density gradient features can effectively reflect the sharp edges and corners of the vehicle surface contour, as well as the boundaries between the vehicle and the ground, and between the vehicle and the surrounding open area, because the point cloud density changes significantly in these areas.

[0046] S124, based on point cloud density gradient features, obtains vehicle contour feature vectors.

[0047] (1) The non-maximum suppression algorithm is used to scan the point cloud density gradient features of the three-dimensional point cloud data.

[0048] The result of the nonmaximum suppression algorithm is that it only retains feature points that are local maxima. These points usually correspond to salient features such as the edges and corners of the vehicle profile.

[0049] For each point cloud data, its feature response value is compared with the feature response values ​​of its neighboring pixels along its gradient direction. If the feature response value of the current data is not a local maximum, its feature response value is suppressed to zero.

[0050] (2) Descriptor extraction is performed on the three-dimensional point cloud data retained after non-maximum suppression, and the descriptors are encoded into vehicle contour feature vectors.

[0051] Commonly used descriptors include the coordinates, scale, and orientation of the feature point, as well as the feature histogram of the surrounding region. These descriptors are encoded into a high-dimensional numerical vector.

[0052] S2 utilizes a dynamic spatiotemporal modeling algorithm to deconstruct the vehicle contour feature vector and construct a Gaussian process regression model.

[0053] Dynamic spatiotemporal modeling algorithms transform static features into dynamic probability estimates, with the aim of constructing a probabilistic model that can accurately reflect the distribution of a vehicle's position and attitude in three-dimensional space.

[0054] S21, the vehicle contour feature vector is analyzed into translational and rotational components.

[0055] Translation and rotation components are two physically independent components. The translation component specifically describes the linear movement of the vehicle along each coordinate axis in three-dimensional space, while the rotation component specifically describes the rotational motion of the vehicle around each coordinate axis in three-dimensional space.

[0056] (1) The translation component is usually represented by a three-dimensional vector, whose three elements correspond to the displacement of the vehicle along the X-axis, Y-axis, and Z-axis in a reference coordinate system with the charging robot as the reference. The mathematical expression for the translation component is: , in: The translation component vector (unit: meters); , , These represent the displacements of the vehicle along the X, Y, and Z axes relative to the charging robot.

[0057] (2) The rotation components can be represented using various mathematical tools such as Euler angles, quaternions, or rotation matrices. The core is to uniquely determine the vehicle's yaw, pitch, and roll angles. If Euler angles are used, the mathematical expression for the rotation components is: , in: For rotation component vectors (unit: radians); Yaw angle; The pitch angle; This is the roll angle.

[0058] S22. For the translation and rotation components, separate Gaussian process regression models are established.

[0059] Gaussian process regression is a non-parametric Bayesian statistical method that is particularly suitable for modeling continuous functions and quantifying the uncertainty of predictions.

[0060] (1) Gaussian process regression model of translation component.

[0061] The Gaussian process regression model established for the translation component takes time or observation sequence index as input and outputs the prediction of the vehicle's three-dimensional coordinates and its variance. Its mathematical expression is as follows: , in: For the translation component at time... The predicted value; Represent a Gaussian process; It is a mean function; The kernel function; used to describe time. and Correlation of observed values.

[0062] Set the mean function ( (The initial translation is used), the kernel function employs a radial basis function, and the length scale parameter is 0.5. After training with 10 sets of observation data, the prediction variance of the translation component can be controlled within 0.001. Within.

[0063] (2) Gaussian process regression model of rotational component.

[0064] A Gaussian process regression model established for the rotational component predicts the vehicle's orientation angle and its uncertainty. The mathematical expression is as follows: , in: For the rotational component at time... The predicted value; is the mean function of the rotational component; is the kernel function for the rotation component.

[0065] S3, based on a Gaussian process regression model, extracts parking space geometric constraints from enhanced 3D point cloud data and constructs a pose-feasible solution space.

[0066] Parking space geometric constraints define the physical boundaries of a vehicle's existence. These constraints typically include the length, width, and height limits of the parking space, as well as the orientation of the boundaries.

[0067] The pose feasible solution space is a mathematical set that contains all combinations of vehicle positions and orientations that satisfy the parking space boundary constraints. The pose feasible solution space is constructed by defining constraint inequalities, such as that the circumscribed cuboid of the vehicle profile must not exceed the area defined by the parking space boundary box.

[0068] The constraint inequalities of the pose feasible solution space are expressed as: , in: These are the boundary values ​​of the parking space along each coordinate axis. For rotational components The projection of the corresponding rotation matrix onto each coordinate axis; The length, width, and height of the vehicle (unit: meters).

[0069] S4. Construct a set of support points consisting of candidate poses from the feasible solution space of poses.

[0070] S41, generate a candidate pose set by sampling from the pose feasible solution space.

[0071] Within the constructed feasible solution space for pose, candidate pose sets need to be generated through sampling. The sampling process can employ a systematic sampling strategy, such as uniform grid sampling across the entire solution space, or a random sampling strategy, such as the Monte Carlo method. Each candidate pose generated through sampling contains a complete combination of position coordinates and attitude angles.

[0072] S42, use Mahalanobis distance to measure the matching degree between each candidate pose and the predicted value of the Gaussian process regression model, and select the set of support points.

[0073] The number of candidate pose sets needs to be large enough to cover meaningful regions in the feasible solution space of poses. The quality of the generated candidate pose set directly affects the accuracy of subsequent probability distribution estimation; overly sparse sampling will cause the model to miss the high-probability regions where the true vehicle pose is located.

[0074] (1) Calculate the Mahalanobis distance.

[0075] Mahalanobis distance is a distance metric that takes into account the covariance structure of data, and it reflects statistical proximity better than Euclidean distance.

[0076] The calculation of Mahalanobis distance requires the use of the covariance matrix of the predicted values ​​of the Gaussian process regression model. This covariance matrix represents the uncertainty of the model prediction.

[0077] The Mahalanobis distance is calculated between the translation component of each candidate pose and the current predicted value of the Gaussian process regression model of the translation component, and the Mahalanobis distance is calculated between the rotation component and the current predicted value of the Gaussian process regression model of the rotation component.

[0078] 1) For the translation component of each candidate pose, the Mahalanobis distance between it and the predicted value of the Gaussian process regression model of the translation component is calculated as follows: , in: The Mahalanobis distance for the translation component; The translation component of the candidate pose; The translation component predicted by the Gaussian process regression model; Let be the covariance matrix of the predicted values.

[0079] 2) For the rotational component of the candidate pose, the Mahalanobis distance between it and the predicted value of the rotational component Gaussian process regression model is calculated as follows: , in: The Mahalanobis distance for the rotation component; The rotation component of the candidate pose; The rotational component predicted by the Gaussian process regression model; Let be the covariance matrix of the predicted values ​​of the rotational components.

[0080] (2) Select the set of support points from the candidate poses based on Mahalanobis distance and confidence threshold.

[0081] The smaller the calculated Mahalanobis distance value, the better the candidate pose matches the current model prediction; the larger the distance value, the further it deviates from the prediction.

[0082] A confidence threshold is set, typically based on the quantiles of the chi-square distribution, which defines the acceptable boundary of the matching degree. Candidate poses with a matching degree higher than the confidence threshold are selected by Mahalanobis distance calculation. These selected candidate poses constitute the support point set of the vehicle pose probability distribution model.

[0083] S5. Construct a vehicle pose probability distribution model based on the support point set.

[0084] The support point set represents the most reasonable set of vehicle pose assumptions given the current sensor observations and prior knowledge. Each point in the support point set contributes to the final probability distribution estimate.

[0085] Kernel density estimation is a non-parametric probability density function estimation method that does not require prior assumptions about the shape of the data distribution. It typically chooses a symmetric probability density function as the kernel function, such as the Gaussian kernel function.

[0086] In the embodiments of this disclosure, the core idea of ​​kernel density estimation is to place a kernel function at the location of each support point. Each kernel function is centered at the support point and has a certain bandwidth parameter. The bandwidth parameter controls the smoothness of the kernel function and affects the level of detail of the final probability density estimation. The values ​​of all kernel functions at various points in space are weighted and summed to obtain a continuous vehicle pose probability distribution model. , in: For any point in the pose space The probability density; The number of support points; For bandwidth parameters; The dimensions of the pose space are (3D translation + 3D rotation = 6D). For the first The pose vector of each support point; This is the kernel function.

[0087] The continuous pose probability distribution field generated by kernel density estimation is a scalar field defined on the vehicle pose space, where the value at each point represents the relative probability of the vehicle being in that pose. The vehicle pose probability distribution field can be visually displayed as a probability cloud map or contour map, with the probability peak regions corresponding to the most likely poses of the vehicle.

[0088] The vehicle pose probability distribution model is continuously updated as new sensor data arrives. The update process involves online learning of the Gaussian process regression model and resampling of the support point set, thereby achieving dynamic tracking of vehicle motion.

[0089] S6. Based on the working radius of the charging robot arm and the probability distribution model of vehicle pose, construct the path optimization objective function.

[0090] S61, calculates the working envelope of the charging robot arm in three-dimensional space.

[0091] The collision risk prediction and optimal docking path selection module in the vehicle position perception and recognition method of electric vehicle charging robots is responsible for ensuring the safety and accuracy of the charging robot arm when approaching the vehicle charging socket.

[0092] Collision risk prediction begins with real-time acquisition of kinematic parameters from each joint of the robotic arm from the charging robot's motion controller. These parameters include the current angle and angular velocity values ​​of each joint, as well as inherent parameters such as link length, torsion angle, and offset distance. Based on these complete kinematic parameters, the working envelope of the charging robot's robotic arm in three-dimensional space is calculated. The working envelope describes the set of all spatial points reachable by the end effector within the joint's range of motion. The mathematical description of the working envelope is typically represented by an implicit equation: , in: The implicit equation for the working envelope surface; Let be the coordinates of any point in three-dimensional space; This refers to the number of joints in the robotic arm. For joint and link serial numbers; From the base coordinate system to the 1st Homogeneous transformation matrix of the joint coordinate system; For the first The angle of joint rotation; For the first The position vector of the end of the link; This is the boundary correction factor for the envelope surface.

[0093] This equation defines a surface in space, and the points inside the surface represent the area that the robotic arm can reach.

[0094] S62, construct a high-pose point cloud subset of the vehicle pose probability distribution model.

[0095] In the vehicle pose probability distribution model, a probability threshold is set. This threshold is used to distinguish between high-probability regions and low-probability regions. Pose points with probability values ​​higher than the probability threshold are selected to form a subset of pose point cloud.

[0096] The pose point cloud subset represents the spatial volume that the vehicle is most likely to occupy. This subset is a discrete set of points, and each point has probability information in addition to its three-dimensional coordinates.

[0097] S63, solve for the spatial intersection volume between the working envelope equation of the robotic arm and the high pose point cloud subset of the vehicle pose probability distribution model, and obtain the trajectory optimization objective function.

[0098] The intersection volume is calculated using a spatial discretization method. The robotic arm's workspace is discretized into small voxels. Which voxels are simultaneously located within the robotic arm's working envelope and covered by a subset of the vehicle's pose point cloud are detected. The total volume of these voxels is then calculated as an estimate of the intersection volume. When the calculated intersection volume exceeds a preset safety threshold according to safety standards, it indicates that the risk of the robotic arm colliding with the vehicle along its current path is unacceptable, and a new safe trajectory needs to be generated.

[0099] Based on the spatial intersection volume, a trajectory optimization objective function incorporating obstacle avoidance constraints is generated. This trajectory optimization objective function is a mathematical expression that integrates multiple optimization objectives: , in: The total cost of the trajectory optimization objective function is a dimensionless scalar. The path length of the candidate trajectory is expressed in meters. A normalized reference value representing the path length, measured in meters; The smoothness cost of the trajectory is calculated by integrating the trajectory curvature, and its dimension is per meter ( / m). The normalized reference value representing the smoothness cost is expressed in units of per meter. This represents the minimum distance between the trajectory and the nearest obstacle, measured in meters. This represents the preset safe distance threshold, measured in meters. These are dimensionless weighting coefficients used to adjust the relative importance of path length, smoothness, and safety in the total cost function.

[0100] S7, solve the trajectory optimization objective function to determine the optimal docking path for the charging robot.

[0101] Solving the objective function for trajectory optimization is a high-dimensional nonlinear optimization problem, and direct solution involves a large amount of computation.

[0102] Sparsification is employed to reduce the dimensionality of the objective function for trajectory optimization. Sparsification methods include using a small number of critical path points instead of continuous trajectories (trajectory points) during trajectory parameterization, or using a linear combination of basis functions to represent the trajectory.

[0103] After trajectory parameterization using a linear combination of basis functions, the critical path points can be represented as: , Where: p The position vector of the trajectory point; For path parameters; The number of basis functions; j is the index of the basis function; are basis functions; The weight coefficients are those of the j-th basis function.

[0104] The trajectory smoothness cost is calculated from the trajectory curvature integral: , in: For curvature; and These are the first and second derivatives of the trajectory, respectively. For the trajectory points; Step size; This represents the number of critical path points.

[0105] The trajectory path length, calculated using weighting coefficients, is as follows: .

[0106] Through this parameterization, the original continuous trajectory optimization problem is transformed into optimizing the weight coefficients. The optimization problem. The mathematical expression of the objective function after sparsification is: , in: The total cost of the objective function is optimized for the sparsified trajectory; This is a vector of weight coefficients; The length of the trajectory path; This serves as a reference value for the normalized path length. Cost for trajectory smoothness; This serves as a reference value for the normalization of smoothness cost; The minimum distance between the trajectory and the nearest obstacle; To preset a safe distance threshold; This is a dimensionless weighting coefficient.

[0107] Solve the sparsified trajectory optimization objective function and output a set of multiple collision-free safe approach trajectories that satisfy kinematic constraints. Each trajectory in the safe approach trajectory set is a spatial path from the current position of the robotic arm to a pre-dock point near the vehicle's charging socket.

[0108] A hierarchical optimization strategy is employed to screen the optimal docking path from a set of safe approach trajectories. This strategy refines the screening process into multiple stages. The first layer of the hierarchical optimization strategy applies a kinematic constraint filter to the set of safe approach trajectories, prioritizing candidate trajectories that meet the maximum speed and acceleration limits of each joint of the charging robot's arm. The second layer of the hierarchical optimization strategy calculates quantitative evaluation metrics for each candidate trajectory. Key metrics include the expected alignment error between the charging interface and the vehicle socket, which is obtained by calculating the positional and directional deviations between the robot arm interface and the center of the vehicle socket at the end of the trajectory.

[0109] The path smoothness index is evaluated by calculating the rate of change of curvature or acceleration of the trajectory. A smooth trajectory helps reduce vibration and wear during the movement of the robotic arm.

[0110] A multi-objective optimization problem is constructed, where the optimization objectives simultaneously include minimizing the charging interface alignment error and maximizing the path smoothness index. These two objectives often conflict and require trade-offs. The mathematical model formula for the multi-objective optimization is as follows: , in: This is the sparsed weight coefficient vector; The number of basis functions; This is due to misalignment of the charging port. This is a path smoothness index; The position deviation weighting coefficient (dimensionless) ); These are the maximum permissible speed and acceleration at the end effector of the robotic arm, respectively. The minimum distance between the trajectory and the nearest obstacle; To preset a safe distance threshold; This is an implicit function for the working envelope of the robotic arm.

[0111] A multi-objective optimization algorithm is used to solve the Pareto front solution set. The solution on the Pareto front indicates that there is no solution that is better than it in all objectives. A compromise solution that achieves a balance between the two objectives is selected from the Pareto front as the optimal docking path.

[0112] Based on the same inventive concept as the above-disclosed content, this disclosure also provides a charging robot path planning system, including a multimodal data fusion module, a regression model construction module, a pose feasible solution space construction module, a support point set acquisition module, a vehicle pose probability distribution model construction module, a path optimization objective function construction module, and a solution module; wherein... The multimodal data fusion module is used to fuse multimodal sensor data to obtain vehicle contour feature vectors and parking space boundary vectors; The regression model building module is used to deconstruct the vehicle contour feature vector using dynamic spatiotemporal modeling algorithms and build a Gaussian process regression model. The pose feasible solution space construction module is used to extract parking space geometric constraints from enhanced 3D point cloud data based on the Gaussian process regression model and construct the pose feasible solution space. The support point set acquisition module is used to acquire a support point set composed of candidate poses from the pose feasible solution space; The vehicle pose probability distribution model construction module is used to construct a vehicle pose probability distribution model based on the support point set. The path optimization objective function construction module is used to construct the path optimization objective function based on the working radius of the charging robot arm and the probability distribution model of the vehicle pose. The solution module is used to solve the trajectory optimization objective function and determine the optimal docking path for the charging robot.

[0113] Based on the same inventive concept as the above-disclosed content, this disclosure also provides an electronic device. For example... Figure 2 As shown, the electronic device of this disclosure includes at least one processor and at least one storage medium electrically connected to the processor. The storage medium is electrically connected to the processor, wherein the storage medium stores instructions executable by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the charging robot docking path planning method as described above.

[0114] It should be noted that the electrical connection between the above-mentioned units does not necessarily mean the connection between lines. The indirect connection method can be applied to the embodiments of this disclosure as long as it achieves the purpose of this disclosure.

[0115] Based on the same inventive concept, this disclosure also provides a computer storage medium storing a computer program, which, when executed by a processor, implements the charging robot docking path planning method as described above.

[0116] Based on the same inventive concept, this disclosure also provides a computer program product stored in at least one storage medium; the computer program product includes several instructions to cause at least one computer device to execute the charging robot docking path planning method as described above.

[0117] Although the present disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A method for planning docking paths for a charging robot, characterized in that, The method includes: By fusing multimodal sensor data, vehicle contour feature vectors and parking space boundary vectors are obtained; A Gaussian process regression model is constructed by deconstructing the vehicle contour feature vector using a dynamic spatiotemporal modeling algorithm. Based on the Gaussian process regression model, parking space geometric constraints are extracted from enhanced 3D point cloud data to construct a pose feasible solution space. From the feasible solution space of pose, obtain the set of support points composed of candidate poses; Construct a vehicle pose probability distribution model based on the support point set; Based on the working radius of the charging robot arm and the probability distribution model of the vehicle pose, a path optimization objective function is constructed. Solve the trajectory optimization objective function to determine the optimal docking path for the charging robot.

2. The method according to claim 1, characterized in that, By fusing multimodal sensor data, vehicle contour feature vectors and parking space boundary vectors are obtained, including: Acquire multimodal sensing data, including 3D point cloud data and polarization characteristic data; By fusing 3D point cloud data and polarization characteristic data, vehicle contour feature vectors and parking space boundary vectors are obtained.

3. The method according to claim 2, characterized in that, By fusing 3D point cloud data and polarization characteristic data, vehicle contour feature vectors and parking space boundary vectors are obtained, including: Preprocessing of 3D point cloud data and polarization characteristic data; After preprocessing, the 3D point cloud data is enhanced based on the polarization characteristic data; Obtain the point cloud density gradient features of the enhanced 3D point cloud data; Vehicle contour feature vectors are obtained based on point cloud density gradient features.

4. The method according to claim 3, characterized in that, Obtain the point cloud density gradient features of the enhanced 3D point cloud data, including: The enhanced 3D point cloud data is divided into voxel meshes; Within the predefined voxel grid, the number of 3D point cloud data falling into each voxel grid is counted to obtain the point cloud spatial distribution density of each voxel grid. Calculate the density gradient features of point clouds based on their spatial distribution density.

5. The method according to claim 4, characterized in that, Based on point cloud density gradient features, the vehicle contour feature vector is obtained, including: The non-maximum suppression algorithm is used to scan the point cloud density gradient features of 3D point cloud data; Descriptors are extracted from the 3D point cloud data retained after nonmaximum suppression, and the descriptors are encoded into vehicle contour feature vectors.

6. The method according to any one of claims 1-5, characterized in that, The vehicle contour feature vector is deconstructed using a dynamic spatiotemporal modeling algorithm, and a Gaussian process regression model is constructed, including: The vehicle contour feature vector is analyzed into translation and rotation components; Independent Gaussian process regression models were established for the translation and rotation components, respectively.

7. The method according to claim 6, characterized in that, From the feasible solution space of poses, construct a set of support points consisting of candidate poses, including: A candidate pose set is generated by sampling within the feasible pose solution space; The Mahalanobis distance is used to measure the degree of matching between each candidate pose and the predicted value of the Gaussian process regression model, and the set of support points is selected.

8. The method according to any one of claims 1-5 and 7, characterized in that, Based on the working radius of the charging robot arm and the probability distribution model of the vehicle pose, a path optimization objective function is constructed, including: Calculate the working envelope of the charging robot arm in three-dimensional space; Construct a high-pose point cloud subset for the vehicle pose probability distribution model; The spatial intersection volume between the working envelope equation of the robotic arm and the high pose point cloud subset of the vehicle pose probability distribution model is solved to obtain the trajectory optimization objective function.

9. A docking path planning system for a charging robot, characterized in that, The system includes a multimodal data fusion module, a regression model construction module, a pose feasible solution space construction module, a support point set acquisition module, a vehicle pose probability distribution model construction module, a path optimization objective function construction module, and a solution module; wherein... The multimodal data fusion module is used to fuse multimodal sensor data to obtain vehicle contour feature vectors and parking space boundary vectors; The regression model building module is used to deconstruct the vehicle contour feature vector using dynamic spatiotemporal modeling algorithms and build a Gaussian process regression model. The pose feasible solution space construction module is used to extract parking space geometric constraints from enhanced 3D point cloud data based on the Gaussian process regression model and construct the pose feasible solution space. The support point set acquisition module is used to acquire a support point set composed of candidate poses from the pose feasible solution space; The vehicle pose probability distribution model construction module is used to construct a vehicle pose probability distribution model based on the support point set. The path optimization objective function construction module is used to construct the path optimization objective function based on the working radius of the charging robot arm and the probability distribution model of the vehicle pose. The solution module is used to solve the trajectory optimization objective function and determine the optimal docking path for the charging robot.

10. An electronic device, characterized in that, Includes at least one processor and at least one storage medium electrically connected; The storage medium is electrically connected to the processor, wherein the storage medium stores instructions executable by at least one of the processors, the instructions being executed by at least one of the processors to enable at least one of the processors to perform the charging robot docking path planning method as described in any one of claims 1-8.

11. A computer storage medium, characterized in that, The computer storage medium stores a computer program. When the computer program is executed by the processor, it implements the docking path planning method for the charging robot according to any one of claims 1-8.

12. A computer program product, characterized in that, The computer program product is stored in at least one storage medium; The computer program product includes several instructions for causing at least one electronic device to execute the charging robot docking path planning method according to any one of claims 1-8.

Images