Skip to content

Latest commit

 

History

History
37 lines (28 loc) · 1.76 KB

slam_planetary_global.md

File metadata and controls

37 lines (28 loc) · 1.76 KB

SLAM for autonomous planetary rovers with global localization

Author: Dimitrios Geromichalos

Year: 2019

Notes:

  • relative positionning via scan matching + global matching of local map to an orbiter global map
  • Factor graph SLAM for planetary exploration : Adaptive localization and mapping with application to planetary rovers
  • " However, LiDARs are hardly used in planetary rovers due to the heavy weight and high power consumption and to the authors knowledge , there exists no LiDAR sensor qualified for planetary missions. "
  • No dense 3D reconstruction needed, just 2,5D elevation grid map

GA SLAM

  • for local mapping, scan to map with a particle filter combined with ICP
  • for global pose, use template matching wrt orbital images

Relative pose

  • PC are downsampled, transformed in the global map frame, cropped and the uncertainty of the eight of each point is computed using a sensor model
  • Then update the local map (whose cells also have a 1D Gaussian distribution in z) using a Kalman Filtering operation
  • Why map update is done before pose estimation?
  • The scan matching particle filter estimates $[x, y, \theta]$
  • Use a particle filter where the weights of each particle is computed using the fitness score of a single ICP iteration
  • The estimated pose is then the average of the particles with the highest weights

Global matching

  • template matching of local elevation map and global orbital map
  • Uses the sobel operator to generate gradient images
  • template matching cannot deal with rotated images: 20 rotated images are generated
  • pose registration limited to the orbital view resolution
  • the surroundings must be rich in elevation features

Experiments

  • data collected in tenerife
  • a flying drone produced the DEM for the global map matching