Mesh RRT* enables path planning on triangular mesh terrain

Otobong Jerome and Tiago Nascimento propose Mesh RRT*, for path planning on non-flat terrain represented by triangular meshes. They point out that the original RRT* cannot run directly on meshes: the surface has zero volume, so uniform sampling cannot hit the surface. The authors claim that Mesh RRT* solves on terrain with slopes up to 75 degrees in every run. However, numerical simulations show that after 30,000 iterations, its reported cost is on average up to 2.7% lower than the exact geodesic, yet no actual path can achieve this cost; the returned paths are on average 0.1%–0.5% longer than the geodesic. The authors also claim that if each chord is replaced by a path on the surface connected across mesh edges by projected points and charged by its own length, then each edge cost equals the exact length of a feasible path; with other conditions unchanged, the exact length of the returned paths under each roughness is on average within 0.07%–0.16% of the geodesic, and on rougher terrain is close to 2 to 3.4 times closer than when using chords. The method is in MeshNav3D 0.2.0 and has not yet been released.

AI 根据报道生成 · 6 days ago更新