CN-121498714-B - Accurate robot foot falling planning method and device based on priori terrain model

CN121498714BCN 121498714 BCN121498714 BCN 121498714BCN-121498714-B

Abstract

The embodiment of the application provides a precise robot foot falling planning method and device based on a priori terrain model, which are characterized in that the method and device are used for digitally modeling preset terrain in a simulation environment, and constructing a digital map containing topographic geometric and semantic information, and definitely calibrating the central coordinate of each ideal foot drop area as a target foot drop point. In the training process, the global coordinates of the foot end of the robot are tracked in real time, the three-dimensional deviation of the foot end of the robot and the nearest target foot drop point is determined through map mapping, and then a reward function taking the deviation as an evaluation index is designed to drive the robot to learn a high-precision foot drop strategy. The method introduces priori terrain target points as guidance, so that the reward signal has definite geometric meaning, can directly and efficiently drive the strategy to converge to accurate foot falling behavior, and greatly improves the motion precision and reliability on the structured terrain.

Inventors

Request for anonymity
Request for anonymity

Assignees

杭州云深处科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260113

Claims (10)

1. The utility model provides a accurate foot planning method that falls of robot based on priori topography model which characterized in that, the method includes: in a simulation environment, carrying out gridding treatment on a preset terrain, and constructing a digital map containing terrain geometry and attributes, wherein the digital map contains a plurality of terrain units; the digital map identifies ideal foot drop areas of each terrain unit, target foot drop points in each ideal foot drop area, and three-dimensional coordinates of each target foot drop point in a predefined world coordinate system; Tracking three-dimensional coordinates of the robot foot end in the world coordinate system in real time, mapping the three-dimensional coordinates to the digital map, and judging whether the robot foot end falls into an ideal foot falling area at the end of the swing phase; setting that when the foot drop point of the foot end of the robot deviates from the target foot drop point, the lower the reward or the heavier the punishment is, and setting a foot drop reward function of the robot according to the setting so as to guide the robot to optimize the foot drop capacity of the robot at the target foot drop point.
2. The accurate robot foot drop planning method based on the prior terrain model according to claim 1, wherein the step of performing gridding processing on the preset terrain to construct a digital map containing terrain geometry and attributes comprises the following steps: Acquiring a three-dimensional model of a preset terrain in a simulation environment, wherein the three-dimensional model comprises geometric information and a preset semantic tag; Defining a fixed world coordinate system as a reference standard of the terrain and the robot pose, and determining a reference level according to the overall trend of the terrain; Performing two-dimensional projection on a three-dimensional model of a preset terrain on the reference horizontal plane to determine a two-dimensional envelope boundary of the terrain, performing grid division in the two-dimensional envelope boundary, and marking each grid unit by a two-dimensional index; Performing attribute assignment on the grid unit, wherein the attribute assignment comprises geometric attribute assignment, semantic attribute assignment and derivative attribute assignment, and the semantic attribute comprises a target foot drop point coordinate; and carrying out structured storage on the generated data containing the geometry and the attributes to obtain the digital map.
3. The accurate robot foot drop planning method based on the prior terrain model according to claim 2, wherein the method comprises the following steps: projecting the grid unit vertically upwards to a three-dimensional model of the terrain, intersecting with the surface of the three-dimensional model, and setting the surface height of the grid unit according to the intersection point; The method for assigning the semantic attributes comprises the steps of defining an ideal foot falling area in advance according to semantic tags of terrains, carrying out special marking on grid cells in the ideal foot falling area, and storing three-dimensional coordinates of target foot falling points in a world coordinate system in attributes of all associated grid cells in the ideal foot falling area; the method for assigning trafficability comprises the steps of assigning a trafficability score to the grid unit according to attribute characteristics of the grid unit, wherein the attribute characteristics comprise gradient, height difference, surface material and friction coefficient; the method for assigning the normal vector comprises the step of calculating an average normal vector of the ground surface where the grid cell is located as the normal vector of the grid cell.
4. The method for accurately planning foot drop of a robot based on a priori terrain model of claim 1, wherein the method for acquiring the target foot drop point in each ideal foot drop area comprises the following steps: Taking a specific point of the ideal foot falling area on the horizontal plane of the grid unit as a target foot falling point, wherein the specific point comprises a geometric center point and a gravity center point; or distributing a plurality of candidate target foot drop points in an ideal foot drop area, wherein the target foot drop points are points closest to foot drop points at the foot end when the foot drop points at the foot end are foot drop points; the method for acquiring the three-dimensional coordinates of each target foot drop point in the predefined world coordinate system comprises the following steps: the position of the target foot drop point on the horizontal plane of the grid unit is taken as the x coordinate and the y coordinate of the target foot drop point in the world coordinate system, and the surface height of the grid unit corresponding to the target foot drop point is taken as the z coordinate of the target foot drop point in the world coordinate system.
5. The method for planning accurate foot drop of a robot based on a priori terrain model according to claim 1, wherein the step of tracking three-dimensional coordinates of the foot end of the robot in the world coordinate system in real time and mapping the three-dimensional coordinates onto the digital map to determine whether the foot end of the robot falls in an ideal foot drop area at the end of the swing phase comprises the steps of: Calculating three-dimensional coordinates of the foot end of the robot in a world coordinate system in real time through forward kinematics chains of the robot and joint angle sensor data; mapping the three-dimensional coordinates of the foot end of the robot to a digital map to obtain grid cell indexes corresponding to horizontal coordinates of the foot end in the digital map when the foot end is at the end of the swing phase; And querying semantic marks of the grid cells by using the grid cell indexes, and judging whether the foot end of the robot falls in an ideal foot falling area.
6. The method for planning accurate foot drop of a robot based on a priori terrain model according to claim 1, wherein the step of calculating the relative positional relationship between the foot end of the robot and the target foot drop point of the ideal foot drop area if the foot end of the robot drops in the ideal foot drop area comprises: acquiring three-dimensional coordinates of a target foot drop point of an ideal foot drop area of the robot foot drop in a world coordinate system, wherein the three-dimensional coordinates of the target foot drop point are stored in the attribute of each grid cell associated with the ideal foot drop area; According to the three-dimensional coordinates of the foot drop point of the robot foot end in the world coordinate system and the three-dimensional coordinates of the target foot drop point in the world coordinate system, calculating a three-dimensional deviation vector between the three-dimensional coordinates, namely, the relative position relation between the robot foot end and the target foot drop point of the ideal foot drop area.
7. The method for accurately planning foot drop of a robot based on a priori terrain model according to claim 1, wherein the step of setting the foot drop reward function of the robot according to the setting, wherein the lower the reward or the heavier the penalty is when the foot drop point of the foot end of the robot deviates from the target foot drop point is, comprises: Calculating a deviation vector of the foot drop point of the foot end of the robot and the target foot drop point, and inputting the deviation vector into a deviation evaluation function, wherein the deviation evaluation function outputs rewards or punishments according to the setting; The deviation evaluation function comprises a square difference function, an L1 norm penalty term, a saturation type continuous rewarding function, a separation weighting and nonlinear processing function and a piecewise function based on contact point classification.
8. Accurate foot planning device that falls of robot based on priori topography model, its characterized in that, the device includes: The map construction module is used for carrying out gridding treatment on preset terrains in a simulation environment to construct a digital map containing the geometric and attribute of the terrains, wherein the digital map contains a plurality of terrain units; the digital map identifies ideal foot drop areas of each terrain unit, target foot drop points in each ideal foot drop area, and three-dimensional coordinates of each target foot drop point in a predefined world coordinate system; The deviation calculation module is used for tracking three-dimensional coordinates of the robot foot end in the world coordinate system in real time, mapping the three-dimensional coordinates to the digital map, judging whether the robot foot end falls into an ideal foot falling area at the end of the swing phase or not, and calculating the relative position relation between the robot foot end and a target foot falling point of the ideal foot falling area if the robot foot end falls into the ideal foot falling area; The function setting module is used for setting that when the foot drop point of the robot foot end deviates from the target foot drop point, the reward is lower or punishment is heavier, and setting the robot foot drop reward function according to the setting so as to guide the robot to optimize the capacity of the robot for foot drop at the target foot drop point.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method for accurately planning a foot drop for a robot based on a priori terrain model as claimed in any of claims 1 to 7 when said program is executed by said processor.
10. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor implements the steps of the method for accurately planning foot drop of a robot based on a priori terrain model as claimed in any of claims 1 to 7.

Description

Accurate robot foot falling planning method and device based on priori terrain model Technical Field The application relates to the field of data processing, in particular to a precise robot foot drop planning method and device based on a priori topographic model. Background In the technical field of robot control, motion control of foot-type and wheel-foot-type robots, in particular self-adaptive crossing under complex unknown terrains, is always the key point and difficulty of research. In recent years, reinforcement learning-based methods have shown great potential in the field, and by letting robots learn autonomously in a simulation environment, strategies thereof can handle high-dimensional state inputs and output complex motion instructions. However, existing reinforcement learning methods still have significant limitations. First, many approaches rely on simplified terrain representation or ontology-aware information, lacking accurate geometric awareness of the surrounding environment, on coupling of perception and action, resulting in a robot that cannot achieve accurate adaptive footfall. Secondly, in the adaptability of complex scenes, when the complex environments such as discrete footfalls (e.g. quincuncial piles) or gaps, ravines and the like exist, the existing strategy is difficult to generate safe and efficient motion sequences, and the problems of collision, instability, planning failure and the like are easy to occur. This is mainly because sparse reward signals are difficult to capture fine terrain geometry, and end-to-end strategies also lack explicit constraints on drop-foot security and long Cheng Guihua capabilities. The invention aims to break through the limitation of insufficient adaptability of the existing foot type/wheel foot type robot in complex terrains, and particularly realizes accurate, stable and self-adaptive traversing capability on discrete foot falling points (such as quincuncial piles). Disclosure of Invention Aiming at the problems in the prior art, the application provides a precise foot drop planning method and device for a robot based on a priori terrain model, so as to guide the robot to conduct precise foot drop planning, and enable the robot to realize precise, stable and self-adaptive traversing capability on discrete foot drop points. In order to solve at least one of the problems, the application provides the following technical scheme: In a first aspect, the present application provides a method for accurately planning foot drop of a robot based on a priori terrain model, comprising: in a simulation environment, carrying out gridding treatment on a preset terrain, and constructing a digital map containing terrain geometry and attributes, wherein the digital map contains a plurality of terrain units; the digital map identifies ideal foot drop areas of each terrain unit, target foot drop points in each ideal foot drop area, and three-dimensional coordinates of each target foot drop point in a predefined world coordinate system; Tracking three-dimensional coordinates of the robot foot end in the world coordinate system in real time, mapping the three-dimensional coordinates to the digital map, and judging whether the robot foot end falls into an ideal foot falling area at the end of the swing phase; setting that when the foot drop point of the foot end of the robot deviates from the target foot drop point, the lower the reward or the heavier the punishment is, and setting a foot drop reward function of the robot according to the setting so as to guide the robot to optimize the foot drop capacity of the robot at the target foot drop point. Further, the step of performing gridding processing on the preset terrain to construct a digital map including geometry and attributes of the terrain includes: Acquiring a three-dimensional model of a preset terrain in a simulation environment, wherein the three-dimensional model comprises geometric information and a preset semantic tag; Defining a fixed world coordinate system as a reference standard of the terrain and the robot pose, and determining a reference level according to the overall trend of the terrain; Performing two-dimensional projection on a three-dimensional model of a preset terrain on the reference horizontal plane to determine a two-dimensional envelope boundary of the terrain, performing grid division in the two-dimensional envelope boundary, and marking each grid unit by a two-dimensional index; Performing attribute assignment on the grid unit, wherein the attribute assignment comprises geometric attribute assignment, semantic attribute assignment and derivative attribute assignment, and the semantic attribute comprises a target foot drop point coordinate; and carrying out structured storage on the generated data containing the geometry and the attributes to obtain the digital map. Further, the method comprises the steps of, Projecting the grid unit vertically upwards to a three-dimensional model of th