Search

CN-121979267-A - Dynamic obstacle avoidance method and system for humanoid robot based on image segmentation

CN121979267ACN 121979267 ACN121979267 ACN 121979267ACN-121979267-A

Abstract

The invention relates to the technical field of intelligent obstacle avoidance, and particularly discloses a dynamic obstacle avoidance method and a dynamic obstacle avoidance system for a humanoid robot based on image segmentation, wherein the method comprises the steps of performing pixel-level semantic segmentation on environment original data in the robot advancing direction to obtain an environment mask map; the method comprises the steps of carrying out contour centroid tracking on an obstacle in a travelling direction, mapping the tracked centroid position into a three-dimensional coordinate system to obtain a motion state vector of the obstacle, carrying out grid terrain analysis on an environment mask map to construct a two-dimensional passable area map, carrying out incremental path searching on the two-dimensional passable area map to obtain a local obstacle avoidance path, combining the motion state vector to adjust a safe distance threshold between the local obstacle avoidance path and the obstacle, carrying out gait phase planning on the robot to generate a joint space track instruction of the robot, and improving the dynamic obstacle avoidance efficiency of the humanoid robot based on image segmentation.

Inventors

  • CAI ZHIQUAN

Assignees

  • 天津天星科技发展有限公司

Dates

Publication Date
20260505
Application Date
20260407

Claims (10)

  1. 1. The dynamic obstacle avoidance method for the humanoid robot based on image segmentation is characterized by comprising the following steps of: Performing pixel-level semantic segmentation on the environment original data in the robot advancing direction to obtain an environment mask map of the environment original data; Based on the environment mask map, tracking the outline centroid of the obstacle in the travelling direction, and mapping the tracked centroid position into a three-dimensional coordinate system to obtain a motion state vector of the obstacle; Carrying out grid terrain analysis on the environment mask map, and constructing a two-dimensional passable area map in the advancing direction; performing incremental path search on the two-dimensional passable region map by taking the current pose of the robot as a starting point to obtain a local obstacle avoidance path of the robot, and adjusting a safety distance threshold between the local obstacle avoidance path and an obstacle by combining a motion state vector; and planning gait phases of the robot based on the local obstacle avoidance path and the safety distance threshold value, and generating a joint space track instruction of the robot.
  2. 2. The dynamic obstacle avoidance method of a humanoid robot based on image segmentation as set forth in claim 1, wherein the performing pixel-level semantic segmentation on the environmental raw data in the robot travelling direction to obtain an environmental mask map of the environmental raw data includes: Acquiring an environment original image in the moving direction of the robot in real time through a binocular vision sensor of the head of the robot; contrast enhancement is carried out on the environment original image, scene semantic analysis is carried out on the enhanced environment image, and an initial segmentation map of the enhanced environment image is obtained; And (3) carrying out conditional random field optimization on the initial segmentation map to obtain an environment mask map of the environment original data.
  3. 3. The image segmentation-based humanoid robot dynamic obstacle avoidance method as claimed in claim 1, wherein the contour centroid tracking of the obstacle in the traveling direction based on the environment mask map comprises: based on connectivity of the environment mask map, performing neighborhood attribution division on the environment mask map to obtain obstacle pixel clusters of the environment mask map; Analyzing the center point of the obstacle pixel cluster to obtain the two-dimensional centroid coordinates of the obstacle pixel cluster; Performing multi-frame association matching on the two-dimensional centroid coordinates to obtain continuous centroid coordinates of the obstacle; and carrying out recursive average filtering on the continuous centroid coordinates to obtain the centroid position of the obstacle.
  4. 4. The method for dynamically avoiding obstacles in a humanoid robot based on image segmentation according to claim 1, wherein mapping the tracked centroid position into a three-dimensional coordinate system to obtain a motion state vector of the obstacle comprises: performing bilinear interpolation reading on the centroid position to obtain a sub-pixel depth sequence of the centroid position; converting the mass center position and the sub-pixel depth sequence into a three-dimensional coordinate system with the hip joint center of the robot as an origin to obtain a three-dimensional space position sequence of the obstacle; Carrying out time sequence smooth estimation on the three-dimensional space position sequence to obtain an instantaneous speed vector and an instantaneous acceleration vector of the obstacle; The instantaneous velocity vector and the instantaneous acceleration vector are integrated into a motion state vector of the obstacle.
  5. 5. The method for dynamically avoiding the obstacle for the humanoid robot based on image segmentation according to claim 1, wherein the step of carrying out grid terrain analysis on the environment mask map to construct a two-dimensional passable area map in the travelling direction comprises the following steps: Performing space discrete projection on the environment mask map by taking the current position of the robot as an origin to obtain an initial semantic grid map in the advancing direction; carrying out neighborhood elevation difference analysis on the ground area grids in the initial semantic grid map to obtain the topographic gradient amplitude of the initial semantic grid map; performing Euclidean distance transformation on the non-ground obstacle region grids in the initial semantic grid map to obtain an obstacle distance field of the initial semantic grid map; calculating the passing confidence of the grids in the initial semantic grid map based on the topographic gradient amplitude and the obstacle distance field; and judging the grids with the passing confidence higher than the preset confidence as passable areas of the initial semantic grid graph, and generating a two-dimensional passable area map in the advancing direction.
  6. 6. The image segmentation-based humanoid robot dynamic obstacle avoidance method as set forth in claim 5, wherein the calculation formula of the traffic confidence is as follows: ; In the formula, Representation grid Is used for determining the traffic confidence of the vehicle, Representation grid Distance values in the obstacle distance field, Representation grid Is used for the gradient magnitude values of (1), Indicating that the preset obstacle affects the scale factor, Representing a pre-set terrain sensitivity factor, An exponential function based on a natural constant e is represented.
  7. 7. The method for dynamically avoiding the obstacle of the humanoid robot based on image segmentation as claimed in claim 1, wherein the step of performing incremental path search on the two-dimensional passable region map with the current pose of the robot as a starting point to obtain the local obstacle avoidance path of the robot comprises the following steps: Extracting a local grid window taking the current pose of the robot as the center in the two-dimensional passable region map, and determining the target grid position in the local grid window according to the local target point of the robot; Taking the grid where the current pose is located as an initial node, taking the target grid position as an end point, and performing neighborhood traversal iteration on the local grid window to obtain the accumulated path cost of the robot; And performing reverse backtracking connection on the target grid position based on the accumulated path cost to obtain a local obstacle avoidance path of the robot.
  8. 8. The method for dynamically avoiding obstacles in humanoid robot based on image segmentation according to claim 1, wherein the adjusting the safe distance threshold between the local obstacle avoidance path and the obstacle in combination with the motion state vector comprises: Judging the movement trend of the obstacle relative to the path point based on the included angle between the instantaneous speed direction in the movement state vector and the direction of the path point pointing to the mass center of the obstacle in the local obstacle avoidance path; According to the movement trend, carrying out interactive urgency judgment on the path points to obtain obstacle threat levels of the path points; And based on the change trend of the obstacle threat level, performing obstacle avoidance margin adjustment on the safety distance threshold between the local obstacle avoidance path and the obstacle to obtain an adjusted safety distance threshold between the local obstacle avoidance path and the obstacle.
  9. 9. The method for dynamically avoiding the obstacle of the humanoid robot based on image segmentation according to claim 1, wherein the step of planning the gait phase of the robot based on the local obstacle avoidance path and the safety distance threshold value to generate a joint space trajectory instruction of the robot comprises the following steps: converting path points in the local obstacle avoidance path into a three-dimensional coordinate system with the hip joint center of the robot as an origin to obtain foot drop point positions of the robot in the running direction; adjusting the foot drop point position based on the motion state vector of the current obstacle and the safe distance threshold; Based on the adjusted foot drop point position, carrying out gesture time sequence planning on the robot to obtain a gait phase sequence of the robot; And coding motion parameters of the gait phase sequence according to the time sequence to obtain a joint space track instruction of the robot.
  10. 10. The dynamic obstacle avoidance system of the humanoid robot based on image segmentation is characterized by being used for realizing the dynamic obstacle avoidance method of the humanoid robot based on image segmentation, and the system comprises the following components: The semantic segmentation module is used for carrying out pixel-level semantic segmentation on the environment original data in the robot travelling direction to obtain an environment mask map of the environment original data; the motion tracking module is used for tracking the outline centroid of the obstacle in the advancing direction based on the environment mask map, and mapping the tracked centroid position into a three-dimensional coordinate system so as to obtain a motion state vector of the obstacle; the map construction module is used for carrying out grid terrain analysis on the environment mask map and constructing a two-dimensional passable area map in the advancing direction; The path planning module is used for carrying out incremental path search on the two-dimensional passable region map by taking the current pose of the robot as a starting point to obtain a local obstacle avoidance path of the robot, and adjusting a safety distance threshold between the local obstacle avoidance path and an obstacle by combining a motion state vector; the gait control module is used for planning gait phases of the robot based on the local obstacle avoidance path and the safety distance threshold value and generating a joint space track instruction of the robot.

Description

Dynamic obstacle avoidance method and system for humanoid robot based on image segmentation Technical Field The invention relates to the technical field of intelligent obstacle avoidance, in particular to a dynamic obstacle avoidance method and system for a humanoid robot based on image segmentation. Background The existing humanoid robot dynamic obstacle avoidance technology lacks of fine processing on the perception of the travelling environment, does not realize pixel-level semantic segmentation, has insufficient tracking precision on the outline and the mass center of the obstacle, is difficult to accurately capture the real-time motion state of the obstacle, does not quantitatively evaluate the analysis of the terrain, cannot accurately define a passable area, and reduces the scientificity of obstacle avoidance decisions. The path planning and the safety distance setting in the prior art lack of dynamic adjustment capability, the safety threshold of the obstacle avoidance path cannot be adjusted in real time according to the motion state of an obstacle, the adaptability of the gait phase planning and the obstacle avoidance path is insufficient, and a matched joint space track instruction is difficult to generate quickly, so that the overall efficiency and the accuracy of the dynamic obstacle avoidance of the humanoid robot are limited, and therefore, how to improve the accuracy and the efficiency of the dynamic obstacle avoidance of the humanoid robot becomes a problem to be solved urgently. Disclosure of Invention The invention provides a dynamic obstacle avoidance method and a system for a humanoid robot based on image segmentation, which aim to solve the problems in the background technology. In order to achieve the above purpose, the invention provides a humanoid robot dynamic obstacle avoidance method based on image segmentation, comprising the following steps: Performing pixel-level semantic segmentation on the environment original data in the robot advancing direction to obtain an environment mask map of the environment original data; Based on the environment mask map, tracking the outline centroid of the obstacle in the travelling direction, and mapping the tracked centroid position into a three-dimensional coordinate system to obtain a motion state vector of the obstacle; Carrying out grid terrain analysis on the environment mask map, and constructing a two-dimensional passable area map in the advancing direction; performing incremental path search on the two-dimensional passable region map by taking the current pose of the robot as a starting point to obtain a local obstacle avoidance path of the robot, and adjusting a safety distance threshold between the local obstacle avoidance path and an obstacle by combining a motion state vector; and planning gait phases of the robot based on the local obstacle avoidance path and the safety distance threshold value, and generating a joint space track instruction of the robot. In a preferred embodiment, the performing pixel-level semantic segmentation on the environmental raw data in the robot travelling direction to obtain an environmental mask map of the environmental raw data includes: Acquiring an environment original image in the moving direction of the robot in real time through a binocular vision sensor of the head of the robot; contrast enhancement is carried out on the environment original image, scene semantic analysis is carried out on the enhanced environment image, and an initial segmentation map of the enhanced environment image is obtained; And (3) carrying out conditional random field optimization on the initial segmentation map to obtain an environment mask map of the environment original data. In a preferred embodiment, the tracing of the centroid of the outline of the obstacle in the traveling direction based on the environmental mask map includes: based on connectivity of the environment mask map, performing neighborhood attribution division on the environment mask map to obtain obstacle pixel clusters of the environment mask map; Analyzing the center point of the obstacle pixel cluster to obtain the two-dimensional centroid coordinates of the obstacle pixel cluster; Performing multi-frame association matching on the two-dimensional centroid coordinates to obtain continuous centroid coordinates of the obstacle; and carrying out recursive average filtering on the continuous centroid coordinates to obtain the centroid position of the obstacle. In a preferred embodiment, the mapping the tracked centroid position into a three-dimensional coordinate system to obtain a motion state vector of the obstacle includes: performing bilinear interpolation reading on the centroid position to obtain a sub-pixel depth sequence of the centroid position; converting the mass center position and the sub-pixel depth sequence into a three-dimensional coordinate system with the hip joint center of the robot as an origin to obtain a three-dimensional space p