Building Augmented Elevation Maps with a Tarantula Robot Rescue Robotics Camp - Rome 2006

1
Building Augmented Elevation Maps with a
Tarantula RobotRescue Robotics Camp - Rome 2006

• Motivation
• PART I Rescue Robots Freiburg
• Please have a look to the other talks of mine
  under http//www.informatik.uni-freiburg.de/klei
  ner/talks.html
• PART II Details on the practical
• Module Structure
• IPC
• Elevation maps

2
Details on the practicalIntroduction

• The goal of this year's tutorial is to enable the
  Tarantula platform for building elevation maps
  from rough terrain, as typically found within
  disaster areas.

3
Details on the practicalProvided Hardware
Software

• All experiments will be carried out on a
  Tarantula robot Lurker, equipped with
• Inertial Measurement Unit (IMU)
• 2 Laser Range Finders (LRFs), one pointing
  parallel to the ground, and one with a freely
  adjustable pitch angle (e.g. pointing downwards
  or upwards).
• The Robot software is organized by modules
  (executables) communicating via Inter Process
  Communication (Simmons and James, 2001)
• Based on this communication framework, we provide
  the following binary modules
• SCAN a module for sending 2D and 3D scans from
  both sensors
• INERTIA a module sending the three Euler angles
  yaw, roll and pitch from the IMU
• JOYSTICK a module sending velocity commands from
  a joystick
• WEBCAM a module sending data from the connected
  cameras
• CONTROLLER a module for controlling the robot,
  e.g. to set the translational and rotational
  velocity
• ROBOGUI a module for teleoperating the robot
• LOC a localization module that sends the robot's
  global pose (x, y, ), estimated from LRF readings
  and Odometry
• RECORDER a module for recording log files
• READER a module for playing log files

4
System overviewInter Process Communication (IPC)
Behaviors ( A Planning Climbing behaviors)
Victim detection
Controller
Commands
Actions
Sensors
Targets
Commands
Detected!
Exploration ( Frontier Cell-based)
Incidence Commander
Other Robots
SLAM
RoboGUI
Map
Sensors
HUMAN
5
Details on the practicalExisting Module Structure
LOC
SCAN
INERTIA
JOYSTICK
ROBOGUI
CONTROLLER
WEBCAM
RECORDER / READER
ROBOT
6
Details on the practicalNew modules (this years
assignment)

• ELEVATION MAPPER
• A module for estimating height values of an
  elevation map by a Kalman Filter-based
  integration of data from the modules INERTIA,
  SCAN and LOC
• HEIGHT ESTIMATOR
• A module for estimating the robot's current
  height by a Kalman Filter-based integration of
  data from INERTA and LOC
• ELEVATION VIEWER
• A module for displaying the elevation map
  computed by the ELEVATION MAPPER and the robot's
  pose calculated by the modules HEIGHT ESTIMATOR
  and LOC

7
Details on the practicalNew modules structure
(this years assignment)
ELEVATION VIEWER
ELEVATION MAPPER
HEIGHT ESTIMATOR
LOC
SCAN
INERTIA
8
Details on the practicalIPC Communication

• Inter Process Communication (IPC)
• Developed by Reid Simmons at the Carnegie Mellon
  University (Simmons and James, 2001)
• Platform-independent package for distributed
  network based message passing
• Facilities for both publish/subscribe and
  client/server type communications
• It can efficiently pass complicated data
  structures between different machines
• IPC can run in either centralized-routed mode or
  direct point-to-point mode
• With centralized routing, message traffic can be
  logged
• Tools available for visualizing and analyzing the
  message traffic

9
Details on the practicalIPC Communication

• Communication between two modules within three
  steps
• Define your data structure, e.g.
• define RESCUE_TILTED_RANGES_NAME
  "rescue_tilted_ranges"define RESCUE_TILTED_RANGE
  S_FMT "int, int, ltdouble2gt,ltdouble2gt,"
  ROBOT_BASE_FMT "" typedef struct int id
  // ID of this scan int nranges // Number
  of ranges
• double ranges // Range double
  angles // Angle (in deg) robot_base_message
  robot // Def. see above rescue_tilted_ranges_me
  ssage
• Publish data on the sender side
• ComPublishToRobot(RESCUE_TILTED_RANGES_NAME,
  msg)
• Receive data on the receiver side with a message
  handler
• void HeightMapperupdate(const
  rescue_tilted_ranges_message msg) ...DEFIN
  E_CLASS_HANDLER(msgHandlerTiltedRanges,
  HeightMapper, rescue_tilted_ranges_message)
• ComSubscribeToRobot(RESCUE_TILTED_RANGES_NAME,
  HeightMappermsgHandlerTiltedRanges, this)

10
Details on the practicalModule interface

• Base message, defining the sender ID and
  timestamp
• define ROBOT_BASE_NAME "robot_base"define
  ROBOT_BASE_FMT "int,long,long"typedef struct
  int id timeval ts robot_base_message
• Global pose calculated by the LOC module
• define RESCUE_KALMAN_POSE_NAME
  "rescue_kalman_pose"define RESCUE_KALMAN_POSE_FM
  T "int, double, double, double, double, double,
  double3,3," ROBOT_BASE_FMT ""typedef
  struct int type // see RESCUE_POSITION_TYPE do
  uble posX // in mm double posZ // in
  mm double posTh // in DEG double transVel //
  in mm/s double rotVel // in DEG/sec double
  sigma33robot_base_message
  robotrescue_kalman_pose_message

11
Details on the practicalElevation Map
representation
m
n
12
Details on the practicalKalman Filter Update
Kalman update of height h and variance s2 from
observation z
Problem with vertical structures
13
Details on the practicalKalman Filter Update
Solution Data fusion depending on Mahalanobis
distance
14
Details on the practical Calculation of z,sz2
a
dx
z
z
d
Linearization with Taylor expansion
15
Details on the practical Variance update from
robot motion
Problem Localization error grows incrementally
with length of robot trajectory
Tt
Tk
Solution Update variance according to traveled
distance
I(t)
t
16
Details on the practical Map filtering with a
convolution kernel
Problem Mixed pixels and surface
holes Solution Map filtering with a convolution
kernel
¼
½
¼
½
1
½
½
¼
¼
17
And what about map augmentation?Thats Raymonds
job!