I. Introduction
With the advancement and development of robot technology, robots are continuously applied in important fields such as military, aerospace, medical, family, agricultural and industrial production[1]. At the same time, the working environment and tasks performed by robots have also become diversified and complex, so how to make robots independently plan a collision-free, stable and better path when performing tasks has become a hot spot in today's robot path planning research. Path planning is to find a stable and collision-free optimal route from the starting point to the target point under the working space[2], there are three main types of path planning, the first is based on grid search and Dijkstra path planning algorithm[2] [4], these two algorithms can find an optimal path with the increase of the resolution of the raster, but the path node is more, and the computer processor requirements are relatively high, and the search time is relatively long. The second type is genetic algorithms[5] and ant colony algorithms[6] based on intelligent algorithms, which can find a better global path, but their local search ability is weak and the convergence speed is slow[7]. Although the ant colony algorithm has strong robustness and strong search ability, the parameters are not easy to set, and the time to solve the optimal path is long[8]. The third type is RRT (fast search random tree) path planning algorithm based on random sampling[9], compared with the above other algorithms, RRT algorithm search faster and suitable for high-dimensional path planning, but there are shortcomings such as poor convergence ability and the search path is not the optimal path [10].