Introduction to
Navigation using ROS
Giorgio Grisetti
The material of this slides is taken from the Robotics 2 lectures given by
G.Grisetti, W.Burgard, C.Stachniss, K.Arras, D. Tipaldi and M.Bennewitz
Outline
Navigation Concepts
Map, Robot Pose, and Path
Taxonomy of Navigation
Localization
Planning
Building a Map
Building a Map
Sense - Plan - Act
To accomplish a task, a
robot should understand
the environment from it
sensor measurements.
Understanding: capturing
the information at the right
level of abstraction.
An autonomously navigating
robot should know
what the environment
looks like and
where it is in the
environment
[courtesy of Rainer Kuemmerle and Dirk Heahnel]
What is this Talk about?
SLAM
localization
mapping
integrated
approaches
(SPLAM)
active
localization
exploration
path planning
[courtesy of Cyrill and Wolfram]
Map
A map is a representation of
the environment where the
robot is operating.
It should contain enough
information to accomplish a
task of interest.
Representations:
Metric
Grid Based
Feature Based
Hybrid
Topological
Hybrid
Robot Pose and Path
A metric map defines a
reference frame.
To operate in a map, a
robot should know its
position in that
reference frame.
A sequence of
waypoints or of actions
to reach a goal location
in the map is a path.
Localization
Determine the current robot position, the
measurements up to the current instant and
a map.
Path Planning
Determine (if it exists) a path to reach a
given goal location given a localized robot
and a map of traversable regions.
Mapping
Given a robot that has a perfect egoestimate of the position, and a
sequence of measurements, determine
the map of the environment.
A perfect estimate of the robot pose is
usually not available.
Instead we solve a more complex
problem: Simultaneous Localization
and Mapping (SLAM).
SLAM
SLAM= Simultaneous Localization and Mapping
Estimate:
these quantities
the map of the environment
are correlated
the trajectory of a moving device
using a sequence of sensor measurements.
Putting Parts Together
To navigate a robot we need
A map
A localization module
A path planning module
These components are sufficient if
The map fully reflects the environment
The environment is static
There are no errors in the estimate
However
The environment changes (e.g. opening/closing doors)
It is dynamic (things might appear/disappear from the perception
range of the robot)
The estimate is noisy
Thus we need to complement our ideal design with other
components that address these issues, namely
Obstacle-Detection/Avoidance
Local Map Refinement, based on the most recent sensor reading.
ROS Navigation Stack
Map provided by a Map Server
Each module is a node
Planner has a layered architecture (local and global planner)
Obstacle sensing refined on-line by appropriate modules (local
and global costmap)
Building a Map
There are gozillions of SLAM algorithms around.
ROS uses GMapping, which implements a particle
filter to track the robot trajectories.
To build a map you need to
Record a bag with /odom, /scan/ and /tf while driving the
robot around in the environment it is going to operate in
Play the bag and the gmapping-node (see the ros wiki and
the live demo), and then save it.
The map is an occupancy map and it is represented
as
An image showing the blueprint of the environment
A configuration file (yaml) that gives meta information
about the map (origin, size of a pixel in real world)
Localizing a Robot
ROS implements the Adaptive Monte Carlo Localization
algorithm
AMCL uses a particle filter to track the position of the robot (see
paper of Fox et al.)
Each pose is represented by a particle.
Particles are
Moved according to (relative) movement measured by the odometry
Suppressed/replicated based on how well the laser scan fits the map,
given the position of the particle.
The localization is integrated in ROS by emitting a transform
from a map-frame to the odom frame that corrects the
odometry.
To query the robot position according to the localization you
should ask the transform of base_footprint in the map frame.
Localization and MARRtino
AMCL relies on a laser
Unless you want to spend 5K euro, you will
not get a laser, so your robot will not
localize with this provedure
However
You can get a kinect/xtion sensor, that
provides data useful to simulate a laser
scanner (how?)
These data can then be plugged in AMCL et
voila you get your system running.
Homework
Bring up a working navigation stack by downloading the git
repository on the web page
Steer the robot through rviz by issuing target locations with
the mouse
Once you have all these components running,
Write a node that whenever it receives a laser scan
computes the position of the laser scan
look in the header of the laser message)
Uses a tf listened to determine the position of the laser w.r.t. the
/map frame
Writes a line in a file with the following format
LASER_POSE <timestamp.sec>.<timestamp.usec><x y z qx qy
qz qw>
