SLAM: Representations

1
SLAM Representations

• R. Chatila
• LAAS-CNRS
• Toulouse, France
• Raja.Chatila_at_laas.fr

2
Representations and Maps
Topology
Features
Objects
Appearance
Grids
3
Linear Features

• Popular for 2D lasers, sonars
• Piecewise linear approximation of sequence of
  points.
• Points have attached uncertainties
• Algorithms based on segmentation and grouping
  (split and merge).
• Hough transform, Least square - Kalman filter,
  RANSAC
• Model lines have attached uncertainties.

4
Linear Features
Local map
Measurements (2D laser scan)
Piecewise linear approximation uncertainties
5
Example

• 2D Sick Laser
• Linear Approximations
• EKF

6
Uncertainty Grids

• Regular grid.
• Free and occupied space
• Probability of presence of obstacle surface
• Update with new measurements
• Assumes independance between grid elements

Konolige
7
Uncertainty Grids
Christensen
8
Uncertainty Grids

• 3D point image (Laser, Stereo)
• Regular grid in 3D image
• Ground projection of 3D points

Lacroix
9
Uncertainty Grids Classification

• Attributes in each grid cell
• Density of points
• Point elevation difference and variance
• Average normal orientation and variance
• Bayesian supervised classification
• 4 classes obstacle, rough, flat, unknown

10
Uncertainty Grids

• Classification

• Reprojection

11
Fusion
12
Appearance Interest Points
13
Appearance interest points
110 images around a 10m diam. rubble

• Interest points (HARRIS detector) in each
  monocular image.
• Corresponding 3D points with EKF SLAM

14
Appearance Interest Points
Errors angles 1, position 1-3 cm
15
Mapping from Aerial imagery Appearance
Interest Points
Altitude 50 m, resolution 80cm
16
Mapping from Aerial imagery Appearance
Interest Points
17
3D Objects
18
Topological model and visual landmarks
Topological models
19
Kalman Filter
State
Observation
Prediction
Innovation
Update
20
SLAM et Kalman

• Etat
• Covariance
• Prédiction
• Observation
• Mise à jour

21
Quelques problèmes ouverts

• Cartographie 3D espace, objets.
• Environnements naturels.
• Environnements dynamiques.
• Grands espaces.
• Association de données (dépend des amers).
• Décision et navigation incertaines.