Generative Adversarial Network Based Heuristics for Sampling-Based Path Planning

Sampling-based path planning is a popular methodology for robot planning. With uniform sampling strategy to explore the state space, feasible can be found without complex geometric modeling of configuration space. However, quality initial solution not guaranteed, and convergence speed optimal slow. In this paper, we present novel image-based algorithm overcome these limitations. Specifically, generative adversarial network (GAN) designed take environment map (denoted as RGB image) input other preprocessing works. The output also an image where promising region (where probably exists) segmented. This utilized heuristic achieve non-uniform planner. We conduct number simulation experiments validate effectiveness proposed method, results demonstrate that our method performs much better in terms solution. Furthermore, apart from environments similar training set, works well on which are very different set.