Summary: | 碩士 === 國立臺灣大學 === 資訊工程學研究所 === 94 === A hybrid Path Planner is proposed based on the Adaptive Medial Axis Probabilistic Roadmap Planner (aMAPRM). It addresses two key drawbacks of aMAPRM: 1) slow progress in vast regions having a major direction and 2) inability to sample through gates of narrow passages. It approaches the first issue by employing "Ellipsoid aMAPRM", which uses the Ellipsoid instead of the Sphere as its sampling boundary, covering more free space with possibly fewer samples. This is different from Covariance Sampling in that, rather than waiting for collisions to occur and then perturbing a sampling covariance matrix, it determines direction prior to collision using nearest obstacle information from the collision detector. The second issue is resolved by employing "Adaptive Voronoi-cuts", which samples across midsections of narrow passages instead of sampling inside them. The “ideal” sampling direction, according to this heuristic, is at the voronoi boundaries of known c-free. By iteratively bisecting midsections, the full connectivity of c-free is gradually understood. This research argues that disjoint roadmaps constructed by the modified aMAPRM can be bridged using Adaptive Voronoi-cuts to form a better, more complete roadmap.
|