Aimed at the problems that the traditional RRT algorithm pays high price in searching unnecessary areas and in planning path under conditions of complex environment, a fast expanding random tree algorithm is proposed, i.e. TNCGRRT* (twoway simultaneous noncollision goal based RRT*). In this algorithm, the bidirectional search strategy in BRRT* and the heuristic search fusion in BIT* are taken as a basic algorithm of this paper, and the batch size of neural network is introduced to determine the number of nodes sampled at one time, affecting the sampling speed. And then, the expansion of forward and reverse trees is carried out simultaneously to speed up the path search, and the expansion direction is defined by improving the extended vertex queue in the target bias strategy and updating the sampling area, and the growth range of random trees is reduced. Finally, the cubic Bspline curve is 〖JP2〗utilized for making the generated path smooth. The experimental results show that compared with the BRRT* algorithm and the BIT* algorithm, the TNCGRRT* algorithm can shorten the path generation time by 4.5%, the number of pruning increases by 80%, and the path cost (i.e. path length) is shortened by 9%, and is valid.