ekfvmap

ScanMatching. More...

ScanMatching.

This driver implements segment-based EKF localization:

"Mobile Robot Localization and Map Building: A Multisensor Fusion Approach" J.A. Castellanos and J.D. Tardós Kluwer Academic Publishers, Boston, 1999 ISBN 0-7923-7789-3

Compile-time dependencies

• none

Provides

• position2d
  • The pose estimation
• localize
  • The full data about the estimation
• opaque
  • Publishes a 9-tuple of doubles which is the covariance matrix.
  • Under the key "covariance"

Requires

• position2d : odometry source of pose and velocity information
• laser : Pose-stamped laser scans (subtype PLAYER_LASER_DATA_SCANPOSE)
• map : Vector map with segments modelling the environment
• graphics2d : optional. If provided, it will be used to display debug data

Configuration requests

• Position requests are forwarded to the odometry position2 interface.
  • They are expected to be in the ekfvloc ref. frame, and will be automatically converted to odometry reference.
• Velocity requests are forwarded to the odometry position2 interface

Configuration file options

• max_laser_range (float)
  • Default: 7.9 m
  • Maximum laser range.
• laser_noise (tuple (doubles): [m rad])
  • Default: [0.004 0.045]
  • Noise in laser range/bearing readings; default is for a SICK LMS200
• odom_noise (tuple (doubles): [m m rad])
  • Default: [0.4 0.2 0.2]
  • Noise in motion model
  • This is, in terms of covariance, 2*sigma
• robot_pose (tuple (doubles): [m m rad])
  • Default: [0.0 0.0 0.0]
  • Initial pose estimation
• robot_pose_initial_error (tuple (doubles): [m m rad])
  • Default: [1.0 1.0 0.2]
  • Initial error the initial robot pose. It EX = 1.0, robot is at X +/- (E/2)
• mapfile (string)
  • Default: none
  • Parameter that allows the loading of a mapfile instead of using a map interface. Mapfile formap is #segments [x0 y0 x1 y1]...
  • When not using this but a requires, use alwayson 1 in the vmapfile driver
  • if you get errors initializing this driver
• truth_model (string)
  • Default : none
  • Named model in stage for ground truth. Used for debugging.
  • When using this, also use a "simulation" requires
• FINE TUNING OPTIONS (units (default))
  • Note that player unit conversions are applied, so these units are only informative for the defaults
• max_region_empty_angle (rad (0.035 ~ 2deg))
  • Angular distance to split regions
• max_region_empty_distance (m (0.1))
  • Euclidean distance to split regions
• min_region_length (m (0.2))
  • Minimum distance between region endpoints
• min_points_in_region (int (4))
  • Minimum laser returns in a region
• split_confidence (percent (95.0))
  • chi-2 confidence test for segment splitting
• check_residual (boolean_int (0))
  • perform residual testing for segment splitting
• max_ang_ebe (rad (0.0))
  • ???
• min_split_segment_distance (m (0.0))
  • minimum dist between new segment endpoints for segment splitting
• min_odom_distance_delta (m (0.0))
• min_odom_angle_delta (rad (0.0))
  • minimum change in odometry before performing update
• backoff_period (s (0.05))
  • minimum time between updates; if lasers arrive faster they'll be dropped
• DEBUG OPTIONS
  • send_debug (int (0))
  • port for custom socket connection to an external debugging GUI
  • If != 0 it will be used.

Example

# Using degrees
driver
(
  name "ekfvmap"
  provides ["position2d:1" "localize:0" "covariance:::0"]
  requires ["position2d:0" "laser:0" "map:0"]

  max_laser_range 31.9
  laser_noise [0.004 0.5]
  odom_noise  [0.4 0.2 10]
  robot_pose  [-49 -39 0]
  robot_pose_initial_error [1.0 1.0 20]
  # mapfile "upc.vec"
  # Use either "requires" or "mapfile"

  # truth_model "robot"
)

Author

J.A. Castellanos, J.D. Tardós (underlying algorithm), Mayte Lázaro (code), A. Mosteo (player integration)