Map building and path planning method of four-footed robot dog

Abstract

The invention discloses a map construction and path planning method of a four-footed robot dog, and belongs to the technical field of intelligent navigation of four-footed robot dogs. The method comprises the following steps: acquiring environment original data through a preset multi-dimensional sensor, and determining a terrain structure, a terrain material, a terrain bearing capacity and a space dynamic obstacle probability of an environment according to the environment original data; constructing a multi-dimensional topological map according to the terrain structure, the terrain material, the terrain bearing capacity and the space dynamic obstacle probability; the method comprises the following steps: acquiring kinematics parameters of a four-footed robot dog, generating a motion path according to the kinematics parameters and a multi-dimensional topological map, and controlling a moving track of the four-footed robot dog according to the motion path, so that parameters such as terrain materials, terrain bearing capacity and space dynamic obstacle probability are supplemented in dimensions of a traditional grid map; the dimension information of the grid map is improved, and the accurate three-dimensional grid map is provided for the robot, so that the task execution capability of the robot in a dynamic complex environment is ensured.

Description

Technical Field

[0001] This application relates to the field of intelligent navigation technology for quadruped robot dogs, and in particular to a map construction and path planning method for quadruped robot dogs. Background Technology

[0002] In recent years, breakthroughs in artificial intelligence, control theory, and materials science have significantly improved the intelligence level and system reliability of robots, driving their deep application in diverse scenarios such as precision agriculture, flexible industrial production, and military reconnaissance. Among various robot systems, platforms with autonomous mobility have become a research hotspot due to their environmental adaptability and scenario generalization capabilities. Legged robots (such as quadrupedal robot dogs) have broken through the limitations of flat surfaces with their biomimetic locomotion mechanisms, enabling them to autonomously traverse unstructured terrains such as forests and grasslands, and stably climb stairs, steps, and steep slopes. Their adaptability in complex environments is significantly better than that of traditional mobile platforms.

[0003] Traditional wheeled or tracked robots typically use two-dimensional grid maps for environmental modeling, employing binary representations of grid occupancy states (obstacles / accessible areas) to make navigation decisions. While this representation method is simple in data structure and highly efficient in storage, it only represents spatial information in a two-dimensional plane and cannot reflect the three-dimensional features of the terrain. For legged robots that rely on precise foot placement control, such two-dimensional maps are insufficient to support their motion planning needs. Furthermore, the lack of dimension in traditional grid maps can lead to obstacle avoidance failures or motion instability, severely limiting the robot's ability to perform tasks in dynamic and complex environments. Summary of the Invention

[0004] The main purpose of this application is to provide a map building and path planning method for a quadruped robot dog, which aims to solve the technical problem that the lack of dimensions in traditional grid maps can lead to obstacle avoidance failure or motion instability, which seriously restricts the robot's ability to perform tasks in dynamic and complex environments.

[0005] To achieve the above objectives, this application provides a map construction and path planning method for a quadruped robot dog, which includes the following steps: By using preset multi-dimensional sensors, raw environmental data is acquired, and based on the raw environmental data, the terrain structure, terrain material, terrain bearing capacity, and probability of spatial dynamic obstacles of the environment are determined. A multi-dimensional topology map is constructed based on the terrain structure, terrain material, terrain bearing capacity, and spatial dynamic obstacle probability. The kinematic parameters of the quadruped robot dog are obtained, and a motion path is generated based on the kinematic parameters and the multi-dimensional topology map, so as to control the running trajectory of the quadruped robot dog according to the motion path.

[0006] In one embodiment, the step of constructing a multi-dimensional topology map based on the terrain structure, the terrain material, the terrain bearing capacity, and the probability of dynamic spatial obstacles includes: Based on the terrain structure, a three-dimensional raster map is generated, which includes the heading angle of a quadruped robot dog walking in the terrain. Based on the terrain material and the terrain bearing capacity, determine the passage cost coefficient when the quadruped robot dog walks in the grid of the three-dimensional grid map; Based on the spatial dynamic obstacle probability, the obstacle risk area in the three-dimensional grid map is determined; A multi-dimensional topology map is constructed based on the three-dimensional grid map, the passage cost coefficient, and the obstacle risk area.

[0007] In one embodiment, the step of generating a three-dimensional raster map with the heading angle of a quadruped robot dog walking in the terrain structure includes: Based on the spatial location information, height structure information, and slope structure information of the terrain structure, as well as the obstacle information in the terrain, a two-dimensional grid map of the terrain is generated, and the position of the quadruped robot dog in the two-dimensional grid map is determined. Generate a simplified cylindrical convex hull of the quadruped robot dog at the specified location, and determine the set of vertices of the cross section of the simplified cylindrical convex hull; Based on the set of vertices of the cross section, the top-view projection shape of the simplified cylindrical convex hull under different heading angles is calculated by using a rotation and torsion matrix; Based on the top-view projection shape and the obstacle information, the passable and impassable areas of the quadruped robot dog when moving in the terrain at different heading angles are determined, and both are added to the two-dimensional grid map to obtain a three-dimensional grid map.

[0008] In one embodiment, the step of determining the passable and impassable areas of the quadruped robot dog when it moves at different heading angles in the terrain based on the top-view projection shape and the obstacle information includes: Obtain the height variation range of the quadruped robot dog while maintaining different postures, and obtain the height threshold of obstacles that the quadruped robot dog can pass through while moving. The obstacle information is filtered based on the height variation range value and the obstacle height threshold; Based on the filtered information and the top-view projection shape, the passable and impassable areas of the quadruped robot dog when moving in the terrain at different heading angles are determined.

[0009] In one embodiment, the step of determining the obstacle risk area in the three-dimensional grid map based on the spatial dynamic obstacle probability includes: If the probability of the spatial dynamic obstacle is greater than a preset threshold, the corresponding grid in the three-dimensional grid map is designated as the initial risk area. Based on the spatial dynamic obstacle probability, predict the movement trajectory of dynamic obstacles in the three-dimensional grid map over a future period of time; Based on the movement trajectory, the obstacle risk area with a risk of travel is determined in the initial risk area.

[0010] In one embodiment, the step of determining the passage cost coefficient of the quadruped robot dog when walking at the grid in the three-dimensional grid map based on the terrain material and the terrain bearing capacity includes: Based on the terrain material, the passage difficulty coefficient of the quadruped robot dog when walking in the grid of the three-dimensional grid map is preset, and the difficulty fluctuation coefficient between each adjacent grid is preset. Based on the terrain carrying capacity, the passage difficulty coefficient and the difficulty fluctuation coefficient are optimized to determine the passage cost coefficient when the quadruped robot dog walks in the grid of the three-dimensional grid map.

[0011] In one embodiment, the step of optimizing the passage difficulty coefficient and the difficulty fluctuation coefficient based on the terrain carrying capacity includes: Obtain the initial value of the terrain bearing capacity preset for different terrain materials, and determine the deviation value between the initial value and the terrain bearing capacity; Based on the deviation value, the passage difficulty coefficient and the difficulty fluctuation coefficient are optimized.

[0012] In one embodiment, the step of generating a motion path based on the kinematic parameters and the multi-dimensional topology map includes: Based on the multi-dimensional topology map, multiple passable paths are determined from the target starting point to the target ending point; Based on the kinematic parameters, a cost condition function is generated to constrain the quadruped robot dog when adjusting different motion postures. Based on the cost condition function and the preset obstacle avoidance priority condition, the optimal movement path is determined from the multiple passable paths.

[0013] In one embodiment, after the step of generating a motion path based on the kinematic parameters and the multi-dimensional topology map, the method further includes: The multi-dimensional sensors are used to collect changes in motion parameters and movement status of the quadruped robot dog as it moves along the motion path. The motion path is locally optimized based on the changes in the kinematic parameters and the changes in the travel state. Based on the results of local optimization and the multidimensional topology map, global optimization is performed on the parts of the movement path that the quadruped robot dog did not reach.

[0014] In one embodiment, after the step of controlling the running trajectory of the quadruped robot dog according to the motion path, the method further includes: The multi-dimensional sensors are used to collect real-time motion data of the quadruped robot dog as it moves along the motion path. If a ground change occurs in the quadruped robot dog's path based on the motion data, the relevant content in the multi-dimensional topology map is updated in real time, and a walking anti-slip mechanism is triggered to control the quadruped robot dog to move forward steadily.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: Raw environmental data is acquired through preset multi-dimensional sensors, and the terrain structure, terrain material, terrain bearing capacity, and probability of dynamic spatial obstacles are determined based on the raw environmental data; a multi-dimensional topology map is constructed based on the terrain structure, terrain material, terrain bearing capacity, and probability of dynamic spatial obstacles; kinematic parameters of a quadruped robot dog are acquired, and a movement path is generated based on the kinematic parameters and the multi-dimensional topology map, so as to control the running trajectory of the quadruped robot dog according to the movement path, i.e., through multi-dimensional sensors... Raw environmental data is acquired, and based on this data, the terrain structure, terrain material, terrain bearing capacity, and probability of dynamic spatial obstacles in the current environment of the quadruped robot dog are determined. These parameters are then incorporated into the grid map construction process to create a multi-dimensional topological map. This compensates for the insufficient parameters of two-dimensional grid maps, which cannot provide adequate route planning for the quadruped robot dog. Simultaneously, based on the kinematic parameters of the quadruped robot dog, corresponding motion paths are generated in this multi-dimensional topological map, and these paths are used to control the robot dog's trajectory, ensuring the accuracy of route planning and guaranteeing the robot's ability to perform tasks in dynamic and complex environments. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the map construction and path planning method for the quadruped robot dog of this application; Figure 2 This is a schematic diagram of the system architecture corresponding to the map construction and path planning method of the quadruped robot dog in this application.

[0019] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application. In the first embodiment, the map building and path planning method of the quadruped robot dog includes the following steps: S10: Obtain raw environmental data through preset multi-dimensional sensors, and determine the terrain structure, terrain material, terrain bearing capacity and probability of spatial dynamic obstacles based on the raw environmental data. In this embodiment, the preset multi-dimensional sensors refer to the sensors that are pre-configured on the quadruped robot dog according to actual needs. Specifically, they may include LiDAR (Laser Imaging, Detection, and Ranging), RGB-D camera (Red, Green, Blue - Dept), IMU (Inertial Measurement Unit), geomagnetic sensor, and force sensor.

[0021] The process involves using LiDAR to collect data on the terrain of the quadruped robot dog's environment, specifically including the location, elevation, and slope of walkable terrain. An RGB-D camera captures terrain images of the robot dog's environment, and semantic segmentation analysis is used to analyze the terrain texture. Specifically, an edge neural network model configured on the robot dog identifies material types such as grass, sand, and concrete in the RGB-D images and assigns them different coefficients (e.g., grass: 0.8, sand: 1.5, ice: 2.0; the higher the coefficient, the greater the difficulty).

[0022] The robot dog's movement posture is compensated by IMU and geomagnetic sensors, and the heading angle of the robot dog when walking in the environment is determined. The actual heading angle of the robot dog in the corresponding terrain of the environment is dynamically determined by continuously monitoring the two.

[0023] Among them, the force sensor detects the ground support force on the feet of the quadruped robot dog, and can combine the results of terrain material analysis to dynamically calculate the terrain bearing capacity, and use this to compensate and control the quadruped robot dog's movement posture and control effect in the terrain in real time.

[0024] The combination of LiDAR and RGB-D can monitor dynamic obstacles in the terrain and predict the probability of dynamic obstacles that may exist in the direction of the quadruped robot dog's movement through predictive analysis, so as to ensure the planning of the quadruped robot dog's movement trajectory. Specifically, based on the prediction of dynamic obstacle trajectory, a spatiotemporal occupancy grid map can be generated, and areas with an obstacle occurrence probability of >30% in the next 5 seconds can be marked as high-risk areas.

[0025] S20, construct a multi-dimensional topology map based on the terrain structure, terrain material, terrain bearing capacity, and spatial dynamic obstacle probability; Understandably, after capturing data from different dimensions through the aforementioned preset sensors, a multi-dimensional topological map of the quadruped robot dog's environment can be constructed by combining the data. This map needs to integrate all the aforementioned terrain structure, terrain material, terrain bearing capacity, spatial obstacle probability, and heading angle into the multi-dimensional topological map to enrich the map information required for the quadruped robot dog to move in the terrain, thereby supplementing the information dimensions required for the quadruped robot dog to move and facilitating better planning of the quadruped robot dog's route.

[0026] In this embodiment, the step of constructing a multi-dimensional topology map based on the terrain structure, terrain material, terrain bearing capacity, and spatial dynamic obstacle probability includes: Based on the terrain structure, a three-dimensional raster map is generated, which includes the heading angle of a quadruped robot dog walking in the terrain. Based on the terrain material and the terrain bearing capacity, determine the passage cost coefficient when the quadruped robot dog walks in the grid of the three-dimensional grid map; Based on the spatial dynamic obstacle probability, the obstacle risk area in the three-dimensional grid map is determined; A multi-dimensional topology map is constructed based on the three-dimensional grid map, the passage cost coefficient, and the obstacle risk area.

[0027] Understandably, traditional wheeled robots plan their routes by generating two-dimensional grid maps and delineating obstacle areas on these maps. However, for quadrupedal robot dogs, due to the structural limitations of their quadrupedal structure, their turning is inherently different from the free turning of wheeled robots. They need to perform multiple coordinated movements of all four legs to generate turning actions. Furthermore, because their quadrupedal structure allows them to climb terrain at certain heights, the simple two-dimensional spatial division of a grid map is insufficient for quadrupedal robot dogs to operate in three-dimensional terrain. To address this issue, this application incorporates the heading angle features of the quadrupedal robot dog walking in the terrain into the traditional two-dimensional grid map, and constructs a corresponding three-dimensional grid map based on these features and the three-dimensional terrain features.

[0028] It needs to be explained that the traditional two-dimensional grid map (XY plane) is expanded into a three-dimensional space, and a new Z-axis is added to represent the heading angle ψ (range 0~360°, resolution Δψ=15°, a total of 24 layers). In this case, each grid state is marked as feasible (0) or infeasible (1), which represents whether the robot is safe at that position and orientation. For example, if a quadruped robot dog is at coordinates (3,5) with a heading angle ψ=30° (corresponding to the 2nd layer of the Z-axis), if the grid is marked as 1, it means that there is a collision risk at this position and orientation.

[0029] Understandably, different terrain materials have different load-bearing capacities, and the difficulty for a quadruped robot dog to traverse these terrains also varies. For example, a quadruped robot dog may slip easily on soft grass, resulting in lower support compared to concrete. Conversely, while ice provides sufficient support due to its smooth surface, it may struggle to gain a foothold and move forward. Therefore, when constructing a 3D grid map, it is necessary to pre-set the terrain material and load-bearing capacity for each grid cell and supplement these parameters into the 3D grid map. Specifically, this supplementary information is a passage cost coefficient, which is calculated by combining the passage difficulty coefficient corresponding to the material and the support capacity coefficient represented by the load-bearing capacity. The difficulty coefficient has a greater weight than the support capacity coefficient.

[0030] The load-bearing capacity calculation includes two methods. One method is to preset a parameter for different materials. However, since the material fluffiness and surface smoothness may vary in actual terrain, it is necessary to combine the sensors on the feet of the quadruped robot dog to collect the corresponding parameters and calculate them during actual walking. Based on the calculation results, the passage cost coefficient in the grid map set according to the preset parameters is corrected.

[0031] Understandably, the above-mentioned process of constructing a 3D grid map mainly involves ground terrain parameters and 3D parameters of the robot dog's walking direction in the environment. In this embodiment, it is also necessary to further supplement the relevant parameter conditions of obstacles. Specifically, obstacles in fixed positions can be determined during the construction of the 3D grid map. Furthermore, it is necessary to supplement the possible running trajectories of dynamic obstacles in the current 3D grid map and the probability of dynamic obstacles appearing in each grid. Specifically, obstacle trajectories can be analyzed through edge neural networks to generate spatiotemporal occupancy grids. ; in, Let be the probability that an obstacle is occupied at position (x, y) at time t. To analyze historical sequences using an edge neural network model ( Predict the future spatiotemporal distribution and output the probability value ∈ [0,1].

[0032] In summary, a multi-dimensional topology map can be constructed based on the three-dimensional grid map, the passage cost coefficient, and the obstacle risk area.

[0033] Specifically, the multi-dimensional topology map refers to a multi-dimensional information fusion map that includes environmental spatial information and the robot dog's heading angle information.

[0034] In this embodiment, the step of generating a three-dimensional raster map with the heading angle of a quadruped robot dog walking in the terrain structure includes: Based on the spatial location information, height structure information, and slope structure information of the terrain structure, as well as the obstacle information in the terrain, a two-dimensional grid map of the terrain is generated, and the position of the quadruped robot dog in the two-dimensional grid map is determined. Generate a simplified cylindrical convex hull of the quadruped robot dog at the specified location, and determine the set of vertices of the cross section of the simplified cylindrical convex hull; Based on the set of vertices of the cross section, the top-view projection shape of the simplified cylindrical convex hull under different heading angles is calculated by using a rotation and torsion matrix; Based on the top-view projection shape and the obstacle information, the passable and impassable areas of the quadruped robot dog when moving in the terrain at different heading angles are determined, and both are added to the two-dimensional grid map to obtain a three-dimensional grid map.

[0035] Understandably, when generating a 3D raster map, it is necessary to first combine the spatial location information, height structure information, and slope structure information of the terrain structure, as well as obstacle information (such as trees, weeds, or other debris in the wild) to generate a 2D raster map, and then further generate a 3D raster map based on the 2D raster map.

[0036] It should be noted that when generating the 3D raster map, the main reference is the position information of the quadruped robot dog in the current terrain to determine its own volume in space, as well as the situation of various obstacles in the terrain. It is necessary to determine whether the two will overlap and collide in certain situations.

[0037] Specifically, the quadruped robot dog is first simplified as a convex hull of a cylinder (rectangular from top view, 1.2m long and 0.6m wide). The top-view projection shape at different ψ (heading angles) is calculated using the rotation matrix R(ψ): Secondly, the Minkowski sum of the obstacle point cloud (obstacle information obtained by filtering by longitudinal height) and the robot's projected shape is performed to generate an equivalent obstacle region: in, For the obstacle point set, for The set of projection points of the robot's top-view projection shape is obtained. In summary, the passable and non-passable areas of the quadruped robot dog when it travels in the 3D grid map can be obtained. Specifically, by determining whether the obstacle point set and the projection point set of the top-view projection shape intersect, the passable and non-passable areas of the quadruped robot dog can be determined. When the passable and non-passable areas are determined, their corresponding information is added to the grid information of each grid in the 3D grid map.

[0038] When determining the passable area, the reasonable passable area can be judged according to the above formula, and the passability at each grid can be determined under the condition that the heading angle of the quadruped robot dog in each grid in the terrain is included.

[0039] In this embodiment, the step of determining the passable and impassable areas of the quadruped robot dog when it moves at different heading angles in the terrain based on the top-view projection shape and the obstacle information includes: Obtain the height variation range of the quadruped robot dog while maintaining different postures, and obtain the height threshold of obstacles that the quadruped robot dog can pass through while moving. The obstacle information is filtered based on the height variation range value and the obstacle height threshold; Based on the filtered information and the top-view projection shape, the passable and impassable areas of the quadruped robot dog when moving in the terrain at different heading angles are determined.

[0040] Understandably, the quadruped robot dog also needs to consider its walking posture and passage conditions under the influence of obstacles at different heights when it moves. Obstacles that are too high on the ground and obstacles that are too low in the air will affect the quadruped robot dog's movement. At the same time, the four legs of the quadruped robot dog will change height during the movement, and the robot dog's body will also tilt, lie down or jump as the four legs are bent and straightened.

[0041] Therefore, it is necessary to consider the range of height changes of the quadruped robot dog when maintaining different postures. By using this range of height changes and the height threshold of the obstacles it can pass through, the height range in the three-dimensional grid map can be further filtered out. This allows for the identification of some obstacles in the three-dimensional grid map that pose a movement obstacle to the quadruped robot dog. Based on this, the passable and impassable areas of the quadruped robot dog when traveling at different heading angles can be identified.

[0042] Specifically, the obstacle height threshold refers to the highest ground height threshold that does not obstruct the robot dog's movement and the lowest ground height threshold that does not obstruct the robot dog's movement.

[0043] In this embodiment, the step of determining the obstacle risk area in the three-dimensional grid map based on the spatial dynamic obstacle probability includes: If the probability of the spatial dynamic obstacle is greater than a preset threshold, the corresponding grid in the three-dimensional grid map is designated as the initial risk area. Based on the spatial dynamic obstacle probability, predict the movement trajectory of dynamic obstacles in the three-dimensional grid map over a future period of time; Based on the movement trajectory, the obstacle risk area with a risk of travel is determined in the initial risk area.

[0044] Understandably, there are also some dynamic obstacles in the 3D grid map. The obstacles vary depending on the location where the robot dog is performing its task, such as vehicles, people or other animals or moving objects. These dynamic spatial obstacles may affect the movement of the quadruped robot dog. Therefore, it is necessary to make a comprehensive judgment on the situation of such dynamic obstacles in the 3D grid map.

[0045] Specifically, since dynamic obstacles move over time, it is necessary to predict the probability of the dynamic obstacle passing through each grid position. If the probability is greater than a preset threshold, the likelihood of the current dynamic obstacle hindering the movement of the quadruped robot dog within that grid is greater. Furthermore, due to the continuous movement of the dynamic obstacle, the area where the probability of the spatial dynamic obstacle is greater than the preset threshold needs to be divided into an initial risk area. This initial risk area is a range-based area, mainly defined by the current grid position of the spatial dynamic obstacle in the 3D grid map. The initial risk area covers a certain range with a radius several times that of a single grid area. By predicting the movement trajectory of the dynamic obstacle in the 3D grid map and combining it with its movement trajectory, the obstacle risk area with the highest probability of risk is divided from the initial risk area.

[0046] By capturing the movement of dynamic obstacles over a period of time, the movement tendency of the dynamic obstacle in the future can be predicted. Based on the current position of the obstacle in the grid, multiple movement trajectories with different probability values ​​may be predicted at the same time. For example, trajectory 1 has a probability of 90% and trajectory 2 has a probability of 70%. Combining these probabilities, areas on multiple paths can be designated as risk areas at the same time.

[0047] The probability here mainly refers to the probability of possible movement trajectories, and the above-mentioned spatial dynamic obstacle probability refers to the probability of a dynamic obstacle appearing at a certain grid.

[0048] In this embodiment, the step of determining the passage cost coefficient of the quadruped robot dog when walking at the grid in the three-dimensional grid map based on the terrain material and the terrain bearing capacity includes: Based on the terrain material, a passage difficulty coefficient is preset for the quadruped robot dog when it walks in the grid of the three-dimensional grid map, and a difficulty fluctuation coefficient between adjacent grids is preset; based on the terrain carrying capacity, the passage difficulty coefficient and the difficulty fluctuation coefficient are optimized to determine the passage cost coefficient for the quadruped robot dog when it walks in the grid of the three-dimensional grid map.

[0049] It should be noted that when constructing a 3D raster map, various parameters of the environmental terrain need to be included to improve the completeness and richness of the 3D raster map through a multi-dimensional information architecture. Compared with a general 2D raster, in this embodiment, the raster at different locations additionally introduces terrain material information. Specifically, since the terrain in which the quadruped robot dog performs its tasks is usually quite varied, there may be multiple terrain materials on a single path, such as sandy slopes, asphalt roads, and grassy ground. Therefore, each raster may have different terrain materials, and there may be abrupt changes in terrain materials between adjacent rasteres. In order to enrich the information richness of the 3D raster map, this information is supplemented.

[0050] The main supplementary parameter is the passage cost coefficient, which represents the difficulty of the quadruped robot dog's movement in the corresponding terrain and serves as a weighting factor in the passage cost calculation. This coefficient primarily ranges from [0, 2]. When the passage difficulty is low, its value is less than 1; if the passage difficulty is high, corresponding to a higher cost for the quadruped robot dog's movement, its value is greater than 1.

[0051] The passage cost coefficient is mainly calculated using terrain carrying capacity, passage difficulty coefficient, and passage difficulty fluctuation coefficient.

[0052] Specifically, by analyzing the terrain material and its carrying capacity, the expected carrying capacity of the terrain corresponding to each grid in the 3D raster map can be determined. Based on the magnitude of the expected carrying capacity, a passage difficulty coefficient is set. The greater the carrying capacity and the less slippage occurs on the terrain, the lower the corresponding passage difficulty coefficient. Based on this, the passage difficulty fluctuation coefficient between multiple adjacent grids mainly considers the changes in terrain material between grids and the changes in the power required by the quadruped robot dog under these changes, which is represented by the difficulty fluctuation coefficient.

[0053] The passage difficulty coefficient can be preset according to different terrain materials, while the passage difficulty fluctuation coefficient is calculated based on the difference between the coefficients corresponding to each grid.

[0054] In this process, the terrain bearing capacity of the ground on which the quadruped robot dog walks is determined during its movement. The two coefficients are then optimized based on the terrain bearing capacity, and the passage cost coefficient is calculated by combining the two optimized coefficients.

[0055] In this embodiment, the step of optimizing the passage difficulty coefficient and the difficulty fluctuation coefficient based on the terrain bearing capacity includes: obtaining an initial value of the terrain bearing capacity preset for different terrain materials, and determining the deviation value between the initial value and the terrain bearing capacity; Based on the deviation value, the passage difficulty coefficient and the difficulty fluctuation coefficient are optimized.

[0056] Understandably, when analyzing different terrain materials through image information, an initial value for bearing capacity can be preset for each material. For example, an initial value for bearing capacity corresponding to the terrain can be preset for materials such as sand, grass, cement, asphalt, and ice. However, in reality, the bearing capacity of specific terrains such as grass, sand, and ice may vary depending on the actual situation. For example, grass with sufficient moisture has a lower bearing capacity, ice with a thinner surface has a lower bearing capacity, and sandy surfaces with slight quicksand have a lower bearing capacity.

[0057] Therefore, the initial value may not meet the mechanical support requirements for the quadruped robot dog to walk on it. Therefore, it is necessary to use the corresponding mechanical sensors to calculate the support force provided by the ground in real time as the quadruped robot dog walks forward. Based on the difference between the calculation result and the initial value, the deviation value is calculated, and the passage difficulty coefficient and difficulty fluctuation coefficient are optimized based on the deviation value.

[0058] The specific optimization process mainly includes calculating the deviation value when the quadruped robot dog steps into the new material terrain. If the deviation value is too large (exceeding the corresponding deviation threshold), the corresponding coefficient needs to be optimized.

[0059] Specifically, regarding the passage difficulty coefficient, the main modification is to increase the grid information of the same material in the multi-dimensional topology map that has an excessively large deviation value, so as to indicate that the passage difficulty has increased and the required terrain carrying capacity has increased.

[0060] Specifically, the difficulty fluctuation coefficient is mainly optimized and adjusted based on the difficulty fluctuation of the ground material changes between adjacent grids. For example, if the load-bearing capacity of the area the robot dog is currently stepping into is less than the initial value, the difficulty fluctuation coefficient of the adjacent ground of the same material should increase synchronously, indicating that the passage difficulty of the area corresponding to multiple consecutive grids has increased.

[0061] S30: Obtain the kinematic parameters of the quadruped robot dog, and generate a motion path based on the kinematic parameters and the multi-dimensional topology map, so as to control the running trajectory of the quadruped robot dog according to the motion path.

[0062] Understandably, quadruped robot dogs have certain kinematic limitations during movement, specifically due to their four-legged structure. The rotation angle of the four legs has a maximum value, the maximum power of the four legs has an upper limit, and the terrain that the four legs can traverse in a coordinated state is limited. Therefore, in this embodiment, the kinematic parameters of the quadruped robot dog are introduced as constraint functions for generating the motion path, and after the motion path is generated, the quadruped robot dog is controlled according to the motion path.

[0063] It should be noted that when generating motion paths using kinematic parameters and multi-dimensional topology maps, the main approach is to use the dynamic parameters and motion structure constraints (heading angle, maximum overturning height, quadrupedal support force, and quadrupedal structural coordination, etc.) in the kinematic parameters as cost functions. This involves analyzing each path to identify the optimal path that minimizes the power consumption and motion complexity required by the quadruped robot dog, rather than simply selecting based on the number of obstacles or the flatness of the path. Instead, the kinematic parameters are used to further filter for motion paths with low travel difficulty and low motion consumption.

[0064] In this embodiment, the step of generating a motion path based on the kinematic parameters and the multi-dimensional topology map includes: Based on the multi-dimensional topology map, multiple passable paths are determined from the target starting point to the target ending point; Based on the kinematic parameters, a cost condition function is generated to constrain the quadruped robot dog when adjusting different motion postures. Based on the cost condition function and the preset obstacle avoidance priority condition, the optimal movement path is determined from the multiple passable paths.

[0065] Understandably, a multi-dimensional topology map contains multi-dimensional information, from which passable and impassable areas can be identified, as well as the movement trajectories of dynamic obstacles. Therefore, by simply determining the target start point and target end point in the multi-dimensional topology map, a suitable passable path can be determined by integrating the multi-dimensional information in the map.

[0066] Furthermore, after determining the passable paths, it is necessary to evaluate the costs of different paths. For example, calculating the path with the lowest energy consumption and the simplest movement for the quadruped robot dog (low difficulty ratio in traversing complex terrain) among all passable paths. Specifically, based on kinematic parameters, a cost condition function is needed for the quadruped robot dog to generate different movement postures when traversing different passable paths. In summary, the optimal movement path can be determined from multiple passable paths based on the cost condition function and the preset obstacle avoidance priority condition.

[0067] The preset obstacle avoidance priority condition refers to prioritizing the avoidance of dynamic obstacles in the terrain. in, The actual cost (such as distance traveled or time) from the target starting point to the current node of the quadruped robot dog. d is the heuristic cost (e.g., Euclidean distance) from the current node to the target endpoint; d is the distance from the current node to the nearest obstacle; D is the safe radius of the quadruped robot dog; This is an adjustment factor (usually taken as 0.5-1.0). Let be the real-time torque of the i-th joint. Let λ be the maximum joint torque (e.g., 35 Nm), and λ be the energy consumption weight. It should be noted that, according to the above formula, the specific process involves guiding the search towards the target direction to ensure path optimality, and increasing the heuristic term when the quadruped robot dog approaches an obstacle. Weighting is used to prioritize obstacle avoidance, avoiding prioritizing obstacle avoidance over path length. Joint physical limits are also introduced as path costs to prevent planning paths beyond the robot's capabilities (e.g., avoiding sharp turns that could overload the quadruped robot's motors). Regarding kinematic parameter fusion, this primarily involves linking the quadruped robot's kinematic model with joint motor performance constraints (e.g., maximum output torque and moment of inertia limits of joint modules), adding a joint energy consumption term to the algorithm's cost function. in, Let be the instantaneous torque of the i-th joint. This represents the time taken per step.

[0068] The kinematic model of this quadruped robot dog is represented by the state-space equations: in, ( () represents world coordinates, ,) represents the velocity in the body coordinate system. angular velocity Among these, the constraints set by the kinematic parameter conditions also include velocity ellipse constraints: ; Where a = 1.0 m / s, b = 0.5 m / s; In addition, it is also necessary to limit the characteristic that reflects forward velocity > lateral velocity, and implement anti-tipping constraints: ; Where d=2.0 / ; Among them, the feasible speed calculation for limiting centripetal acceleration during high-speed cornering involves optimizing the cost function: Feasible speeds for travel direction and heading angle φ:

[0069] in, ; The movement cost of the quadruped robot dog is as follows: Where l is the grid spacing and λ is the energy consumption weight.

[0070] In this embodiment, after the step of generating a motion path based on the kinematic parameters and the multi-dimensional topology map, the method further includes: The multi-dimensional sensors are used to collect changes in motion parameters and movement status of the quadruped robot dog as it moves along the motion path. The motion path is locally optimized based on the changes in the kinematic parameters and the changes in the travel state. Based on the results of local optimization and the multidimensional topology map, global optimization is performed on the parts of the movement path that the quadruped robot dog did not reach.

[0071] Understandably, multi-dimensional sensors can collect relevant data in real time during the movement of a quadruped robot dog, especially the kinematic parameters of the quadruped robot dog, such as the terrain bearing capacity of the feet, the friction coefficient of the feet when moving, and the movement state of the quadruped robot dog (running, crawling). Based on this data, the movement path can be locally optimized.

[0072] It should be noted that the local optimization mainly refers to the fact that the movement of the quadruped robot dog will always change due to terrain characteristics when moving along the main movement path according to the predetermined route. For example, when the quadruped robot dog is climbing a landslide, improper force may cause it to slip, resulting in its heading angle deviating from the predetermined direction, or the quadruped robot dog may be affected by sudden intervention of dynamic obstacles during its movement. Therefore, it is necessary to optimize the local route for the above situations, and after optimizing the local route, to optimize the subsequent untraveled path globally.

[0073] This means optimizing the detailed features of the quadruped robot dog during its movement to ensure that it can smoothly pass through the planned movement path. For example, it can perform emergency obstacle avoidance or urgently increase the support of its legs to ensure the quadruped robot dog moves forward steadily.

[0074] In optimizing the movement parameters of the quadruped robot dog, its heading angle and expected route may deviate to some extent. In order to avoid the increased cost when the quadruped robot dog returns to the old route after deviating from the predetermined route, the route that has not yet been traversed in the subsequent movement path is optimized and adjusted starting from the local optimization point to ensure the availability of the entire path.

[0075] In this embodiment, after the step of controlling the running trajectory of the quadruped robot dog according to the motion path, the method further includes: The multi-dimensional sensors are used to collect real-time motion data of the quadruped robot dog as it moves along the motion path. If a ground change occurs in the quadruped robot dog's path based on the motion data, the relevant content in the multi-dimensional topology map is updated in real time, and a walking anti-slip mechanism is triggered to control the quadruped robot dog to move forward steadily.

[0076] Understandably, when constructing the multi-dimensional topology map, some of the terrain bearing capacity is calculated by collecting data from the quadruped robot dog's sensors, and some of the terrain bearing capacity is based on fixed parameter values ​​preset according to the terrain material. However, these fixed parameter values ​​cannot fully represent the actual bearing capacity of the terrain. Therefore, during the quadruped robot dog's movement, it is still necessary to calculate the bearing capacity in real time through the sensors configured on its legs and optimize and correct the multi-dimensional topology map based on the real-time calculation results.

[0077] During the quadruped robot dog's movement, it may encounter sudden changes in ground conditions (specifically, changes in terrain material or load-bearing capacity). To ensure the robot dog's stability, an anti-slip mechanism is needed to control the increase or decrease of the output torque of the four legs, ensuring steady progress. Specifically, this involves calculating the terrain load-bearing capacity, adjusting the foot positions, and using the ground friction coefficient as a triggering condition for the mechanism (wherein...). > 0.3 triggers):

[0078] in, This is the adjustment amount for the foot landing depth. The real-time normal force of the foot tribe point, Let k be the expected normal force at the foot landing point; where k is the ground stiffness. ; in, The real-time normal force of the foot tribe point, The depth of the subsidence.

[0079] Where μ is the coefficient of friction: in, The lateral friction force at the foot landing point. The longitudinal friction force at the foot landing point, When the real-time normal friction coefficient μ at the foot landing point is less than 0.3, the foot pad texture enhancement (such as increasing the texture depth at the foot tip) is triggered.

[0080] In summary, referring to Figure 2 This embodiment provides a system architecture corresponding to the map building and path planning method of a quadruped robot dog, which specifically includes four main layers: a multi-sensor fusion layer, a map building layer, a path planning layer, and a control execution layer.

[0081] The multi-sensor fusion layer integrates parameters such as terrain structure, terrain material, terrain bearing capacity, spatial dynamic obstacle probability, and the heading angle of the quadruped robot dog through preset multi-dimensional sensors. The map building layer first constructs a two-dimensional grid map, and after introducing other parameters, generates a three-dimensional grid map and a multi-dimensional topology map. Combining the multi-dimensional topology map, the path planning layer plans the route that the quadruped robot dog needs to walk within it. Through adaptive optimization, the control execution layer controls various motion parameters of the quadruped robot dog. When constructing a multi-dimensional topological map using pre-set multi-dimensional sensors, the system primarily uses LiDAR to collect data on the terrain structure of the quadruped robot dog's environment. This data includes the location, height, and slope of walkable surfaces. An RGB-D camera captures terrain images of the robot dog's environment, and semantic segmentation analysis is used to analyze the terrain material. Specifically, an edge neural network model configured on the robot dog identifies material types such as grass, sand, and concrete in the RGB-D images, and semantic segmentation assigns different coefficients to different materials. An IMU and geomagnetic sensors compensate for the robot dog's movement posture and determine its heading angle in the environment. These sensors are continuously monitored to dynamically determine the robot dog's actual heading angle within the corresponding terrain. Force sensors measure the ground support force on the robot dog's feet, and combined with terrain material analysis results, dynamically calculate the terrain bearing capacity, using this to compensate for and control the robot dog's movement posture and control effectiveness in the terrain in real time.

[0082] This embodiment acquires raw environmental data through pre-set multi-dimensional sensors and determines the terrain structure, terrain material, terrain bearing capacity, and probability of dynamic spatial obstacles based on the raw environmental data. A multi-dimensional topology map is constructed based on the terrain structure, terrain material, terrain bearing capacity, and probability of dynamic spatial obstacles. The kinematic parameters of the quadruped robot dog are acquired, and a motion path is generated based on the kinematic parameters and the multi-dimensional topology map. The robot dog's trajectory is controlled according to this motion path. Specifically, by acquiring raw environmental data through multi-dimensional sensors and determining the terrain structure, terrain material, terrain bearing capacity, and probability of dynamic spatial obstacles in the current environment of the quadruped robot dog, these parameters are incorporated into the grid map construction process to create a multi-dimensional topology map. This compensates for the insufficient parameters of two-dimensional grid maps, which cannot reasonably plan routes for the quadruped robot dog. Furthermore, based on the quadruped robot dog's kinematic parameters, a corresponding motion path is generated in this multi-dimensional topology map, and this path is used to control the robot dog's trajectory, ensuring the accuracy of route planning and guaranteeing the robot's task execution capability in dynamic and complex environments.

[0083] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for map construction and path planning for a quadruped robot dog, characterized in that, The map construction and path planning method for the quadruped robot dog includes the following steps: acquiring raw environmental data through preset multi-dimensional sensors, and determining the terrain structure, terrain material, terrain bearing capacity, and spatial dynamic obstacle probability of the environment based on the raw environmental data; constructing a multi-dimensional topology map based on the terrain structure, terrain material, terrain bearing capacity, and spatial dynamic obstacle probability; acquiring the kinematic parameters of the quadruped robot dog, and generating a movement path based on the kinematic parameters and the multi-dimensional topology map, so as to control the running trajectory of the quadruped robot dog according to the movement path.

2. The method as described in claim 1, characterized in that, The step of constructing a multi-dimensional topology map based on the terrain structure, terrain material, terrain bearing capacity, and spatial dynamic obstacle probability includes: generating a three-dimensional grid map with the heading angle of a quadruped robot dog walking in the terrain based on the terrain structure; determining the passage cost coefficient of the quadruped robot dog walking at the grid in the three-dimensional grid map based on the terrain material and terrain bearing capacity; determining the obstacle risk area in the three-dimensional grid map based on the spatial dynamic obstacle probability; and constructing a multi-dimensional topology map based on the three-dimensional grid map, the passage cost coefficient, and the obstacle risk area.

3. The method as described in claim 2, characterized in that, The step of generating a three-dimensional grid map with the heading angle of a quadruped robot dog walking in the terrain structure includes: generating a two-dimensional grid map of the terrain based on the spatial location information, height structure information, and slope structure information of the terrain structure, as well as obstacle information in the terrain, and determining the position of the quadruped robot dog in the two-dimensional grid map; generating a simplified cylindrical convex hull of the quadruped robot dog at the position, and determining the set of vertices of the cross section of the simplified cylindrical convex hull; calculating the top-view projection shape of the simplified cylindrical convex hull under different heading angles based on the set of vertices of the cross section by using a rotation and torsion matrix; determining the passable and impassable areas of the quadruped robot dog moving in the terrain at different heading angles based on the top-view projection shape and the obstacle information, and adding both to the two-dimensional grid map to obtain a three-dimensional grid map.

4. The method as described in claim 3, characterized in that, The step of determining the passable and impassable areas of the quadruped robot dog when moving at different heading angles in the terrain based on the top-view projection shape and the obstacle information includes: obtaining the height variation range value of the quadruped robot dog while maintaining different postures, and obtaining the obstacle height threshold that the quadruped robot dog can pass through while moving; filtering the obstacle information based on the height variation range value and the obstacle height threshold; and determining the passable and impassable areas of the quadruped robot dog when moving at different heading angles in the terrain based on the filtered information and the top-view projection shape.

5. The method as described in claim 3, characterized in that, The step of determining the obstacle risk area in the three-dimensional grid map based on the spatial dynamic obstacle probability includes: if the spatial dynamic obstacle probability is greater than a preset threshold, then defining the corresponding grid in the three-dimensional grid map as an initial risk area; predicting the movement trajectory of dynamic obstacles in the three-dimensional grid map in the future period based on the spatial dynamic obstacle probability; and determining the obstacle risk area with a travel risk in the initial risk area based on the movement trajectory.

6. The method as described in claim 2, characterized in that, The step of determining the passage cost coefficient of the quadruped robot dog when walking in the grid of the three-dimensional grid map based on the terrain material and the terrain bearing capacity includes: presetting the passage difficulty coefficient of the quadruped robot dog when walking in the grid of the three-dimensional grid map based on the terrain material, and presetting the difficulty fluctuation coefficient between each adjacent grid; optimizing the passage difficulty coefficient and the difficulty fluctuation coefficient based on the terrain bearing capacity, and determining the passage cost coefficient of the quadruped robot dog when walking in the grid of the three-dimensional grid map.

7. The method as described in claim 6, characterized in that, The step of optimizing the passage difficulty coefficient and the difficulty fluctuation coefficient based on the terrain bearing capacity includes: obtaining an initial value of the terrain bearing capacity preset for different terrain materials, and determining the deviation value between the initial value and the terrain bearing capacity; and optimizing the passage difficulty coefficient and the difficulty fluctuation coefficient based on the deviation value.

8. The method as described in claim 1, characterized in that, The step of generating a motion path based on the kinematic parameters and the multi-dimensional topology map includes: determining multiple passable paths from the target starting point to the target ending point based on the multi-dimensional topology map; generating a cost condition function to constrain the quadruped robot dog when adjusting different motion postures based on the kinematic parameters; and determining the optimal motion path from the multiple passable paths based on the cost condition function and a preset obstacle avoidance priority condition.

9. The method as described in claim 8, characterized in that, After the step of generating a motion path based on the kinematic parameters and the multi-dimensional topology map, the method further includes: acquiring changes in motion parameters and travel status of the quadruped robot dog as it moves along the motion path using the multi-dimensional sensor; performing local optimization on the motion path based on the changes in kinematic parameters and travel status; and performing global optimization on the portion of the motion path that the quadruped robot dog has not reached based on the results of the local optimization and the multi-dimensional topology map.

10. The method as described in claim 8, characterized in that, After the step of controlling the running trajectory of the quadruped robot dog according to the movement path, the method further includes: collecting real-time motion data of the quadruped robot dog as it moves forward according to the movement path through the multi-dimensional sensor; if a ground change is determined based on the motion data, the relevant content in the multi-dimensional topology map is updated in real time, and a walking anti-slip mechanism is triggered to control the quadruped robot dog to move forward steadily.

Images