Improved RRT* For Fast Path Planning in Underwater 3D Environment

Path planning for Autonomous Underwater Vehicle(AUV) needs to find a feasible path in three-dimension workspace and is a very difficult and challenging task. Traditional algorithms have failed to find the path effectively in high-dimensional space since it is proven that the complexity of the problem is NP-hard. Consequently, sample-based algorithms, such as Rapidly-exploring Random Tree star (RRT*), have been proposed to find the probabilistically complete and asymptotically optimal solution, which reduces the complexity of the algorithm. However, in underwater environment with undulating terrain and scattered floating obstacles, the global uniform random sampling strategy in RRT* costs too much memory and time, resulting in a slow convergence to optimal path. To deal with these problem, an improved RRT* is proposed in this paper. Goal-biased Gaussian distribution sampling with variable standard deviation is proposed to controls the randomness of search nodes. Furthermore, a focused optimal search strategy is introduced to improve the convergence rate. Finally, some simulations are conducted under various 3D underwater environments. The results show that the improved RRT* outperforms RRT* in search efficiency and convergence rate under water.