Skip to content
hlx1996 edited this page Aug 13, 2019 · 2 revisions

Q: Difference between different mapping frameworks for planning? In specifically, FIESTA vs Voxblox vs Ewok?

A: This comment is the detailed answer to this question

Q: Different mapping framework used in different repos of our group?

A: This comment is the detailed answer to this question

Q: Common tricks used in speedup the raycasting process?

A: While it is the common tricks in the mapping framework, it is omitted in the paper. The FIESTA mapping framework supports both deterministic occupancy grid map (for Velodyne LiDAR sensor) and probabilistic occupancy grid map (for depth map or point clouds). Only the latter needs the raycasting process. Raycasting task increases the probability of the grid of reprojected 3D point and decreases the probability of the grids between the origin and that 3D point. If we reproject all the voxels in the depth map one by one and deal with every grid along with all rays, the computing work will be enormous, making it the bottleneck of the whole framework. The tricks are thus yielded to deal with this problem.

First, multi-thread can be used, for this problem is high-parallelable. Second, when we need to increase/decrease a probabilistic, we will check if this has been increased/decreased before in this turn of updating. If not, then do it as usual; otherwise, stop the ray casting for this point. In this way, every grid will be each increased/decreased at most once. The effectivity of optimization will be much more significant when more points in the point cloud, especially for higher resolution of the camera and higher resolution of the grid map.

Clone this wiki locally