The International Journal of Robotics Research Study, Ahead of Publish.
Computer optimum, collision-free trajectories for high-dimensional systems is a tough and vital issue. Sampling-based organizers have problem with the dimensionality, whereas trajectory optimizers might obtain embeded neighborhood minima as a result of intrinsic nonconvexities in the optimization landscape. Using mixed-integer shows to envelop these nonconvexities and locate internationally optimum trajectories has actually just recently revealed fantastic guarantee, many thanks partially to limited convex leisures and reliable estimate techniques that considerably lower runtimes. These strategies were formerly restricted to Euclidean arrangement rooms, preventing their usage with mobile bases or constant revolute joints. In this paper, we manage such circumstances by modeling arrangement rooms as Riemannian manifolds, and we explain a decrease treatment for the zero-curvature instance to a mixed-integer convex optimization issue. We better offer an approach for getting approximate remedies by means of piecewise-linear estimates that applies to manifolds of approximate curvature. We show our outcomes on numerous robotic systems, consisting of creating reliable collision-free trajectories for a PR2 bimanual mobile manipulator.
发布者：Thomas Cohn，转转请注明出处：https://robotalks.cn/non-euclidean-motion-planning-with-graphs-of-geodesically-convex-sets/
