Humanoid robot indoor path planning method and system based on earth surface trafficability estimation, electronic equipment and storage medium

Abstract

The invention belongs to the technical field of path planning, and provides a humanoid robot indoor path planning method and system based on earth surface trafficability estimation, electronic equipment and a storage medium. The method comprises the steps of earth surface type map generation, texture feature extraction, equivalent friction coefficient prediction, flatness index calculation, trafficability probability map prediction, candidate path generation, safety audit screening and comprehensive cost screening, and optimal motion strategy generation. According to the method, the ground surface physical characteristics are introduced into the decision-making process, so that the planning level improvement from geometric obstacle avoidance to physical quality and efficiency is realized; by actively identifying and avoiding specific risks, main accident hidden dangers such as slipping and stumbling are directly eradicated from a decision source, and the reliability is improved.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and in particular to a method, system, electronic device, and storage medium for indoor path planning of humanoid robots based on ground accessibility estimation. Background Technology

[0002] Humanoid robots, due to their bipedal locomotion, face more severe stability challenges in unstructured indoor environments than wheeled robots. Existing path planning methods, whether based on global planning using pre-built 2D grid maps or local real-time obstacle avoidance algorithms combining LiDAR and visual sensors, all fundamentally address the problem of obvious obstacle avoidance.

[0003] However, for humanoid robots, an area without obvious obstacles can become difficult to traverse or high-risk due to the physical characteristics of the surface. For example, low-friction surfaces, such as liquids or scattered paper on floor tiles; soft or uneven surfaces, such as loose mats or carpets, or loose wires, can cause slips, trips, or gait instability. Existing obstacle avoidance techniques in path planning treat all these areas as safe passageways, leading the humanoid robot to plan the shortest path across smooth floor tiles instead of the most stable path around to a non-slip carpet. Therefore, current technology lacks an understanding of the physical semantics of the surface and cannot meet the high requirements of humanoid robots for ground contact texture, which is a major bottleneck for achieving fully autonomous and highly reliable movement in complex indoor environments. CN119915286A provides an indoor path planning method based on multi-location fusion. It mainly relies on multi-sensor technology to acquire indoor location information, and then calculates the optimal path based on the target location, map information, and current location information. This method does not involve obstacle avoidance methods in local path planning. CN112747736A presents a vision-based indoor UAV path planning method. This method primarily utilizes point cloud data to perform global path planning and exploratory local planning in unknown areas, and can quickly respond to sudden obstacles. However, this method does not address the vision-based ground accessibility estimation problem. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method, system, electronic device, and storage medium for indoor path planning of humanoid robots based on ground accessibility estimation, thereby solving the problem that existing methods lack understanding of the physical semantics of the ground.

[0005] To achieve the above objectives, the present invention provides the following solution: An indoor path planning method for humanoid robots based on surface accessibility estimation includes: RGB images and depth point clouds are acquired, and the YOLO11-seg model is used to perform instance segmentation on the RGB images to obtain a land surface type map; The depth point cloud is divided into blocks according to the surface type map to obtain block point clouds. The pre-constructed texture extraction formula is used to extract features from each block point cloud to obtain texture features. The surface type map and the texture features are input into a prefitted regression function for mapping to obtain the equivalent friction coefficient; The local elevation variance mean of the point cloud of the block is calculated to obtain the flatness index; The equivalent friction coefficient and the smoothness index are input into a pre-trained accessibility probability prediction model for prediction, and an accessibility probability map is obtained. Based on the preset start and end points, the planner generates several candidate paths according to the land surface type map. Based on the accessibility probability map, each candidate path is subjected to security audit screening and comprehensive cost screening to obtain the optimal path; The optimal path is transformed into an optimal motion strategy using a gait strategy mapping mechanism based on physical parameters, and the target humanoid robot is controlled to move using the optimal motion strategy.

[0006] Preferably, the texture extraction formula is: ;in, ; ; ; The texture feature; For contrast; Homogeneity; The fusion coefficient; , These represent the i-th row and the j-th column, respectively. The element in the i-th row and j-th column of the gray-level co-occurrence matrix; To offset pixels and The element in the i-th row and j-th column of the gray-level co-occurrence matrix; , These are the horizontal and vertical spatial offsets, respectively; , These are the width and height of the original image pixels, respectively; For indicator functions; The value is set for the pixel.

[0007] Preferably, the optimal path is obtained by performing security audit screening and comprehensive cost screening on each candidate path based on the accessibility probability map, including: Extract the minimum passability probability of the grid cells traversed by the candidate path to obtain the minimum passability probability; Candidate paths whose minimum passability probability is less than a preset security threshold are eliminated; The comprehensive cost data is obtained by calculating the comprehensive cost data of the candidate paths whose minimum passability probability is greater than or equal to a preset security threshold using a pre-constructed comprehensive cost function; the expression of the comprehensive cost function is: ;in, For path The overall cost; , These are the first weighting coefficient and the second weighting coefficient, respectively. This is the path length; This represents the average probability of passage. The path with the lowest overall cost is determined as the optimal path based on the comprehensive cost data.

[0008] Preferably, a humanoid robot indoor path planning system based on surface accessibility estimation includes: The instance segmentation module is used to acquire RGB images and depth point clouds, and to perform instance segmentation on the RGB images using the YOLO11-seg model to obtain a land surface type map; The point cloud feature extraction module is used to divide the depth point cloud into blocks according to the land surface type map to obtain block point clouds, and to extract features from each block point cloud using a pre-built texture extraction formula to obtain texture features. The friction coefficient prediction module is used to input the surface type map and the texture features into a prefitted regression function for mapping to obtain the equivalent friction coefficient. The flatness calculation module is used to calculate the mean local elevation variance of the point cloud of the block to obtain the flatness index. The accessibility prediction module is used to input the equivalent friction coefficient and the smoothness index into a pre-trained accessibility probability prediction model for prediction, and obtain an accessibility probability map. The path generation module is used to generate several candidate paths based on the preset start and end points and the land surface type map using a planner. The path filtering module is used to perform security audit filtering and comprehensive cost filtering on each candidate path based on the accessibility probability map to obtain the optimal path; The motion strategy generation module is used to convert the optimal path into an optimal motion strategy using a gait strategy mapping mechanism based on physical parameters, and to control the target humanoid robot to move using the optimal motion strategy.

[0009] Preferably, an electronic device includes: at least one processor and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the aforementioned humanoid robot indoor path planning method based on surface accessibility estimation.

[0010] Preferably, a non-transitory computer-readable storage medium stores computer instructions for causing a computer to execute the aforementioned humanoid robot indoor path planning method based on surface accessibility estimation.

[0011] The present invention discloses the following technical effects: This invention provides a method, system, electronic device, and storage medium for indoor path planning of humanoid robots based on surface accessibility estimation. By incorporating surface physical characteristics into the decision-making process, it solves the problem of existing methods lacking understanding of surface physical semantics, achieving a planning level improvement from geometric obstacle avoidance to physical quality and efficiency. By actively identifying and avoiding specific risks, it solves the problem of existing methods treating all areas with all physical characteristics as safe passage areas, which can easily lead to robot accidents when generating paths, thus eliminating specific accident hazards from the source of decision-making. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A schematic diagram of the indoor path planning process for a humanoid robot based on surface accessibility estimation provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the segmentation of potential areas affecting robot movement on the ground based on YOLO, provided in an embodiment of the present invention. Figure 3 A flowchart of the surface accessibility probability estimation algorithm provided in an embodiment of the present invention; Figure 4 A flowchart of path planning based on surface accessibility estimation is provided for an embodiment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] The purpose of this invention is to provide a method, system, electronic device, and storage medium for indoor path planning of humanoid robots based on ground accessibility estimation, which solves the problem that existing methods lack understanding of the physical semantics of the ground.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] Figure 1 A schematic diagram of the indoor path planning process for a humanoid robot based on surface accessibility estimation provided in this embodiment of the invention is shown below. Figure 1 As shown, this invention provides a method for indoor path planning of humanoid robots based on surface accessibility estimation, including: Step 100: Acquire RGB images and depth point clouds, and use the YOLO11-seg model to perform instance segmentation on the RGB images to obtain a land surface type map; Step 200: Divide the depth point cloud into blocks according to the surface type map to obtain block point clouds, and extract features from each block point cloud using a pre-constructed texture extraction formula to obtain texture features; Step 300: Input the surface type map and the texture features into the prefitted regression function for mapping to obtain the equivalent friction coefficient; Step 400: Calculate the mean local elevation variance of the point cloud of the block to obtain the flatness index; Step 500: Input the equivalent friction coefficient and the smoothness index into the pre-trained passability probability prediction model for prediction to obtain a passability probability map; Step 600: Based on the preset start and end points, generate several candidate paths using the planner according to the surface type map; Step 700: Based on the accessibility probability map, perform security audit screening and comprehensive cost screening on each candidate path to obtain the optimal path; Step 800: The optimal path is transformed into an optimal motion strategy using a gait strategy mapping mechanism based on physical parameters, and the target humanoid robot is controlled to move using the optimal motion strategy.

[0018] Specifically, the texture extraction formula is as follows: ;in, ; ; ; The texture feature; For contrast; Homogeneity; The fusion coefficient; , These represent the i-th row and the j-th column, respectively. The element in the i-th row and j-th column of the gray-level co-occurrence matrix; To offset pixels and The element in the i-th row and j-th column of the gray-level co-occurrence matrix; , These are the horizontal and vertical spatial offsets, respectively; , These are the width and height of the original image pixels, respectively; For indicator functions; The value is set for the pixel.

[0019] Furthermore, based on the accessibility probability map, the optimal path is obtained by security audit screening and comprehensive cost screening for each candidate path, including: Extract the minimum passability probability of the grid cells traversed by the candidate path to obtain the minimum passability probability; Candidate paths whose minimum passability probability is less than a preset security threshold are eliminated; The comprehensive cost data is obtained by calculating the comprehensive cost data of the candidate paths whose minimum passability probability is greater than or equal to a preset security threshold using a pre-constructed comprehensive cost function; the expression of the comprehensive cost function is: ;in, For path The overall cost; , These are the first weighting coefficient and the second weighting coefficient, respectively. This is the path length; This represents the average probability of passage. The path with the lowest overall cost is determined as the optimal path based on the comprehensive cost data.

[0020] Specifically, in this embodiment, an independent and continuously quantified surface accessibility evaluation dimension is introduced in local path planning. This dimension is specifically used to evaluate the access quality of surfaces without visible obstacles and serves as the core basis for path optimization, thereby enabling the robot to actively select the access surface with the highest motion stability and lowest risk.

[0021] refer to Figure 2 This approach focuses on environmental perception and region segmentation to improve accessibility. The robot acquires real-time RGB images and depth point clouds of the surrounding environment through a proprioceptive vision system (preferably an RGB-D depth vision camera). First, to balance real-time performance and accuracy, the YOLO11-seg model is used to perform instance segmentation on the images acquired by the proprioceptive vision system. Its core task is to further identify surface types with different physical properties from "empty" areas, such as smooth tiles, surfaces with scattered small parts, areas with reflective surfaces / liquids, and uneven carpet-like surfaces. The key to this step is refining the traditional "empty / occupied" binary map into a "surface type map" with physical semantics.

[0022] Furthermore, surface parameter quantification estimation is performed through multi-feature fusion. The segmentation results are fused with the depth point cloud to accurately define the 3D extent of each surface block. Subsequently, accessibility quantification estimation is performed for each block: 1) Visual Appearance Feature Analysis: To assess the friction risk of the ground surface, it is necessary to quantify the texture features (such as roughness) of the surface patches. The method involves first calculating the gray-level co-occurrence matrix in the RGB image. Specifically, assume the grayscale image has a set of grayscale levels as follows: and pixel space offset is So, the gray-level co-occurrence matrix... It is A matrix, where the elements are: .

[0023] pixels The grayscale is And offset pixels The grayscale is When the count is incremented by 1, and then normalized by dividing by the number of all pixel pairs, the gray-level co-occurrence matrix can be obtained.

[0024] Then from the matrix Contrast calculated Homogeneity The value is used to combine and quantize texture features. .

[0025] Gray-level co-occurrence matrix Contrast directly reflects the clarity of the texture and the depth of the grooves. The higher the contrast, the rougher the surface. The calculation formula is as follows: and These are the grayscale values ​​of two pixels: .

[0026] Gray-level co-occurrence matrix Homogeneity measures how close the gray-level co-occurrence matrix is ​​to the diagonal. In texture features, higher homogeneity results in a smoother surface. The calculation formula is as follows: .

[0027] Finally and Combining them yields texture features. A hyperparameter is introduced here. To adjust the contributions of both, so that the texture features The higher the value, the rougher the surface. The calculation formula is as follows: .

[0028] The above will be discussed next. The value is combined with the ground material identified by the previous YOLO semantic segmentation model, such as asphalt, dry tile, carpet, cement floor, and wet tile, and converted into an equivalent coefficient of friction. By integrating texture features and material category Input a linear function The equivalent friction coefficient is obtained, i.e. The mapping function The following calibration procedure was used to determine the static friction coefficient in a laboratory environment: Images of various standard surface materials (such as dry asphalt, wet tiles, carpet, etc.); calculate the comprehensive texture features of each image. And the corresponding category of the label, thus obtaining a set of calibration data pairs. Finally, the mapping function is determined through linear regression or piecewise linear fitting. The specific parameters.

[0029] 2) Point cloud geometric feature analysis: For geometric abrupt changes that can easily cause tripping, such as carpet edges, loose mats, and steep elevation changes, the average local elevation variance of the point cloud is used to quantify the smoothness index. Local elevation variance describes the dispersion of elevation values ​​among all points in a small neighborhood around a specific point in a point cloud dataset. A larger value indicates greater unevenness. The specific steps are as follows: Select a point from the point cloud of the edge region of loose objects such as carpets identified by the previous segmentation model. .

[0030] Find the distance Recent One point.

[0031] Get all points in this neighborhood The axis coordinate values ​​form an elevation set. .

[0032] calculate Local elevation variance of a point .

[0033] .

[0034] Repeat the above steps to calculate the local elevation variance of all points in the point cloud, and finally calculate the average to obtain the flatness index. .

[0035] refer to Figure 3 A passability cost layer is constructed. A parameterized virtual ground environment is built within the high-fidelity robot simulation platform (NVIDIA-Isaac-Sim). The surface properties of this environment include at least the static friction coefficient. Surface elevation difference By using gridded sampling, a large number (e.g., tens of thousands to hundreds of thousands) of different parameter combinations are generated within the range of the parameter values. The simulation training scenario, in which Indicates the first Each parameter is sampled. In each generated parameterized scene, the reinforcement learning training process for the humanoid robot is initiated, and a fixed training time window is set. .exist Within a given timeframe, if the robot learns a stable walking strategy (e.g., successfully walking a certain distance without falling), then this set of parameters will be... Label it as "Passable" (e.g., label 1); otherwise, label it as "Passable" (e.g., label 0), and record it as... This results in a large-scale dataset of surface accessibility. .

[0036] Using the dataset obtained above, train a classification model capable of probabilistic prediction. Preferably, this model can be a shallow multilayer perceptron (MLP) neural network or a logistic regression model. The input layer of this network is a three-dimensional vector [μ, h, d], and the output layer outputs a two-dimensional probability distribution through a softmax activation function. ,in This is the estimated probability that the robot can successfully traverse the terrain under the current physical parameters. The training objective of this model is to minimize the cross-entropy loss function between its predictions and the simulation labels.

[0037] The previously obtained real-world physics parameter vector This data is input into the pre-trained accessibility probability prediction model. The model outputs... The value represents the final accessibility probability of that surface area. This probability value is a continuous quantity between 0 and 1, and can be directly used to evaluate the surface accessibility of all collision-free paths obtained by conventional path planning algorithms. This then generates a cost map that identifies the quality of passage within all "empty" areas.

[0038] refer to Figure 4 This system focuses on local path optimization with accessibility as its core principle. First, it employs traditional path planning to obtain multiple collision-free candidate paths. Then, based on real-time data from LiDAR or depth cameras, the system constructs a binary obstacle map representing the geometric space occupancy. Subsequently, a planner (e.g., a lightweight...) The algorithm (or Fast Random Tree Search, RRT) starts at the robot's current position and ends at a given local target point, performing a multi-path search on the obstacle map. The core objective of this stage is to generate a map containing... strip( The set of geometrically collision-free candidate paths These paths exhibit diversity in length and smoothness, providing a basis for the next stage of optimization. After obtaining a set of geometrically feasible candidate paths, surface accessibility assessment is introduced in this embodiment as a basis for higher-level decision-making. This step first utilizes a pre-constructed accessibility probability map. For each candidate path Perform a passability security audit. Specifically, calculate the minimum passability probability of all grids traversed by the path. and compare it with a preset security threshold. Comparison. Any existence Paths deemed to traverse unacceptably high-risk terrain will be immediately flagged as unqualified and eliminated. For all qualified paths that pass the security audit, the system further calculates a comprehensive cost function, which is typically a weighted combination of path length and average passability probability. Ultimately, the system selects the most cost-effective path from the qualified paths. The path with the shortest length is the optimal path for final execution. .

[0039] Preferably, gait adaptation is based on accessibility prediction. To achieve end-to-end adaptation from environmental perception to underlying motion, this embodiment introduces a gait strategy mapping mechanism based on physical parameters after path planning. Specifically, for each surface area traversed by the planned path, the system matches the equivalent friction coefficient and smoothness parameters calculated in real time with the closest parameter sample in the simulation training database. By retrieving and calling the optimal motion strategy trained by reinforcement learning under this set of optimal matching parameters, the system can perform feedforward online adjustments to the humanoid robot's underlying gait (such as foot trajectory, foot impact, torso posture, and zero-moment point ZMP stability boundary). This mechanism ensures that the robot's motion mode can accurately match the physical characteristics of the current surface, thereby actively maintaining dynamic balance and walking efficiency under complex ground conditions, realizing an intelligent closed loop from environmental understanding to behavior generation.

[0040] As an optional implementation, this embodiment also provides a humanoid robot indoor path planning system based on surface accessibility estimation, including: The instance segmentation module is used to acquire RGB images and depth point clouds, and to perform instance segmentation on the RGB images using the YOLO11-seg model to obtain a land surface type map; The point cloud feature extraction module is used to divide the depth point cloud into blocks according to the land surface type map to obtain block point clouds, and to extract features from each block point cloud using a pre-built texture extraction formula to obtain texture features. The friction coefficient prediction module is used to input the surface type map and the texture features into a prefitted regression function for mapping to obtain the equivalent friction coefficient. The flatness calculation module is used to calculate the mean local elevation variance of the point cloud of the block to obtain the flatness index. The accessibility prediction module is used to input the equivalent friction coefficient and the smoothness index into a pre-trained accessibility probability prediction model for prediction, and obtain an accessibility probability map. The path generation module is used to generate several candidate paths based on the preset start and end points and the land surface type map using a planner. The path filtering module is used to perform security audit filtering and comprehensive cost filtering on each candidate path based on the accessibility probability map to obtain the optimal path; The motion strategy generation module is used to convert the optimal path into an optimal motion strategy using a gait strategy mapping mechanism based on physical parameters, and to control the target humanoid robot to move using the optimal motion strategy.

[0041] As an optional implementation, this embodiment also provides an electronic device, including: at least one processor and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the aforementioned humanoid robot indoor path planning method based on surface accessibility estimation.

[0042] As an optional implementation, this embodiment also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the aforementioned humanoid robot indoor path planning method based on surface accessibility estimation.

[0043] The beneficial effects of this invention are as follows: This invention improves planning efficiency by incorporating surface physical characteristics into the decision-making process, moving from geometric obstacle avoidance to physical quality and effectiveness. By proactively identifying and mitigating specific risks, it directly eliminates major accident hazards such as slipping and tripping at the source of decision-making, thereby enhancing reliability.

[0044] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0045] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for indoor path planning of humanoid robots based on surface accessibility estimation, characterized in that, include: RGB images and depth point clouds are acquired, and the YOLO11-seg model is used to perform instance segmentation on the RGB images to obtain a land surface type map; The depth point cloud is divided into blocks according to the surface type map to obtain block point clouds. The pre-constructed texture extraction formula is used to extract features from each block point cloud to obtain texture features. The surface type map and the texture features are input into a prefitted regression function for mapping to obtain the equivalent friction coefficient; The local elevation variance mean of the point cloud of the block is calculated to obtain the flatness index; The equivalent friction coefficient and the smoothness index are input into a pre-trained accessibility probability prediction model for prediction, and an accessibility probability map is obtained. Based on the preset start and end points, the planner generates several candidate paths according to the land surface type map. Based on the accessibility probability map, each candidate path is subjected to security audit screening and comprehensive cost screening to obtain the optimal path; The optimal path is transformed into an optimal motion strategy using a gait strategy mapping mechanism based on physical parameters, and the target humanoid robot is controlled to move using the optimal motion strategy.

2. The indoor path planning method for humanoid robots based on surface accessibility estimation according to claim 1, characterized in that, The texture extraction formula is: ;in, ; ; ; The texture feature; For contrast; Homogeneity; The fusion coefficient; , These represent the i-th row and the j-th column, respectively. The element in the i-th row and j-th column of the gray-level co-occurrence matrix; To offset pixels and The element in the i-th row and j-th column of the gray-level co-occurrence matrix; , These are the horizontal and vertical spatial offsets, respectively; , These are the width and height of the original image pixels, respectively; For indicator functions; The value is set for the pixel.

3. The indoor path planning method for humanoid robots based on surface accessibility estimation according to claim 1, characterized in that, Based on the accessibility probability map, the optimal path is obtained by security audit screening and comprehensive cost screening for each candidate path, including: Extract the minimum passability probability of the grid cells traversed by the candidate path to obtain the minimum passability probability; Candidate paths whose minimum passability probability is less than a preset security threshold are eliminated; The comprehensive cost data is obtained by calculating the comprehensive cost data of the candidate paths whose minimum passability probability is greater than or equal to a preset security threshold using a pre-constructed comprehensive cost function; the expression of the comprehensive cost function is: ;in, For path The overall cost; , These are the first weighting coefficient and the second weighting coefficient, respectively. This represents the path length. This represents the average probability of passage. The path with the lowest overall cost is determined as the optimal path based on the comprehensive cost data.

4. A humanoid robot indoor path planning system based on surface accessibility estimation, characterized in that, The system for implementing the humanoid robot indoor path planning method based on surface accessibility estimation as described in claim 1 includes: The instance segmentation module is used to acquire RGB images and depth point clouds, and to perform instance segmentation on the RGB images using the YOLO11-seg model to obtain a land surface type map; The point cloud feature extraction module is used to divide the depth point cloud into blocks according to the land surface type map to obtain block point clouds, and to extract features from each block point cloud using a pre-built texture extraction formula to obtain texture features. The friction coefficient prediction module is used to input the surface type map and the texture features into a prefitted regression function for mapping to obtain the equivalent friction coefficient. The flatness calculation module is used to calculate the mean local elevation variance of the point cloud of the block to obtain the flatness index. The accessibility prediction module is used to input the equivalent friction coefficient and the smoothness index into a pre-trained accessibility probability prediction model for prediction, and obtain an accessibility probability map. The path generation module is used to generate several candidate paths based on the preset start and end points and the land surface type map using a planner. The path filtering module is used to perform security audit filtering and comprehensive cost filtering on each candidate path based on the accessibility probability map to obtain the optimal path; The motion strategy generation module is used to convert the optimal path into an optimal motion strategy using a gait strategy mapping mechanism based on physical parameters, and to control the target humanoid robot to move using the optimal motion strategy.

5. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform a humanoid robot indoor path planning method based on ground accessibility estimation as described in any one of claims 1 to 3.

6. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute any one of claims 1 to 3, a humanoid robot indoor path planning method based on surface accessibility estimation.

Images