Conditional Generative Adversarial Networks for Optimal Path Planning

Path planning plays an important role in autonomous robot systems. Effective understanding of the surrounding environment and efficient generation optimal collision-free path are both critical parts for solving path-planning problems. Although conventional sampling-based algorithms, such as rapidly exploring random tree (RRT) its improved version (RRT*), have been widely used problems because their ability to find a feasible even complex environments, they fail efficiently. To solve this problem satisfy two aforementioned requirements, we propose novel learning-based algorithm which consists generative model based on conditional adversarial networks (CGANs) modified RRT* (denoted by CGAN-RRT*). Given map information, our CGAN can generate probability distribution paths, be utilized CGAN-RRT* with nonuniform sampling strategy. The is trained learning from ground-truth maps, each generated putting all results executing RRT 50 times one raw map. We demonstrate performance testing it groups maps comparing Informed-RRT* algorithm.