Increased visibility sampling for probabilistic roadmaps

Sampling based planning algorithms solve the problem of Robot Motion Planning by sampling a number of vertices to make a roadmap or a tree, which is then searched for a solution. The sampling strategy denotes the mechanism to generate samples used to construct the tree or the roadmap. In this paper new sampling strategies are proposed for the Probabilistic Roadmap technique that generate samples aiming at maximizing the sample visibility. The increased visibility makes it easier to construct edges with the neighboring samples and thus contribute to get a solution early. Based on this principle three new samplers are pro-posed. The first sampler generates samples inside corridors and promotes them exactly to the corridor centres. The second sampler uses a distance threshold bi-nary search to approximately place the samples in the corridor centre. The last sampler attempts to bias the sampling towards narrow corridors, while still placing the samples approximately at the corridor centres. The increased visibility pays off for the increased computation effort incurred therein. The approach is tested for narrow corridor scenarios and is experimentally found to surpass all state-of-the-art sampling techniques of Probabilistic Roadmap.