Grid Map Construction and Terrain Prediction for Quadruped Robot Based on C-Terrain Path

Zhe Li,Hui Zhang,Yibin Li,Xuewen Rong

doi:10.1109/access.2020.2977396

Abstract

In general, grid map based path planning algorithms are employed in the robotics arena. The algorithm uses a grid map to represent environmental information, standardized. Compared with feature maps and topological maps, the algorithm realizes the construction of environmental maps in a more direct way, and has the characteristics of fast, simple and efficient.The integration and prediction of terrain is an unavoidable problem and the traditional raster map prediction method is based on the research of the terrain data itself, and lacks dynamic supplement for the path planning process. When the environmental data changes, the classification algorithm can only be re-executed, and the past data is completely discarded. Since the planned path is unlikely to change, the terrain tends to be stable. To solve this problem, this paper proposes a concept of C(circular)-terrain band following path nodes and terrain construction and prediction methods. The C-Terrain method first obtains an ordered set of passing points at the initial moment, based on the complete path planning. Then an ordered sequence of influence function values is obtained, which depends on the selection of the terrain band and the adjustment of related parameters. Finally, regression methods such as machine learning are used to complete the prediction of the path and location terrain, and the unknown path and terrain are predicted. The experimental results prove the accuracy and practical value of the C-T method.

Highlights

For quadruped robots, the ability to identify terrain is the key to improving motion efficiency in complex environments
Based on the concept of equidistant terrain, this paper proposes a method to judge the topographic features of grid nodes based on the characteristics of C-terrain
The terrain band is combined with the initial path to form a sequenced set of terrains following the path point

Summary

INTRODUCTION

The ability to identify terrain is the key to improving motion efficiency in complex environments. This paper focuses on the terrain classification method of robot vision system based on raster map. Z. Li et al.: Grid Map Construction and Terrain Prediction for Quadruped Robot Based on C-Terrain Path texture. Classifies the terrain for the existing path This ensures that when the environment map changes rapidly, the algorithm can provide more accurate terrain information for more important path points because the path offset is not large. A∗ algorithm is a common grid map path planning method, which was proposed by Nilsson in 1980. It can search the optimal path to target point by contrast evaluation function The core of it is to add heuristic search part based on Dijkstra algorithm. F (n) is the evaluation function, g(n) represents the path cost from the initial point n to any node. h(n) said heuristic evaluation price, from the node n to the target point

MAP CLASSIFICATION METHOD AND TERRAIN BAND

SIMULATION AND EXPERIMENT

CONCLUSION AND PROSPECT