Sheet 2: Topic: Motion Models and Robot Odometry

Visual Navigation for Flying Robots
D. Cremers, J. Sturm, J. Engel, C. Kerl

Summer Term 2013
Computer Vision Group

Department of Informatics
Technical University of Munich
Sheet 2
Topic: Motion Models and Robot Odometry
Submission deadline: Tue, 21.05.2013, 10:15 a.m.
Hand-in via email to visnav2013@vision.in.tum.de
Introduction: Please read first!
In this exercise sheet, you will learn how to use a Kalman filter to estimate the pose
of the AR.Drone from its sensor measurements, using visual markers attached to
the ground: In the first exercise, you will set up everything required to fly the
drone with a joystick, and record a .bag file of the drone flying above visual markers.
In the second exercise, you will develop and implement the Kalman filter, and use
it to estimate the drones trajectory from the recorded .bag file from exercise one.
As we have more groups than AR.Drones we provide an example .bag file, such that
some of you can start with the second exercise (for which you do not require an
AR.Drone or joystick).
Exercise 1: Manual Flight
(a) Set up the joystick: If you are working on your own laptop, you first have to
install the ROS joystick packages using
sudo apt-get install ros-fuerte-joystick-drivers
Now plug in the joystick, and start roscore and the joystick driver using
rosrun joy joy node. Inspect the topic /joy, and verify that the joystick
works - you might have to press the P button once to initialize it.
(b) Set up the control node: We provide a node that translates the raw joystick
messages to the correct control commands sent to the drone, which you can
find here:
git://github.com/tum-vision/ardrone joystick.git
Clone it into your workspace, run rosmake ardrone joystick, and start the
node using rosrun ardrone joystick ardrone teleop. The axes and buttons are assigned as follows:
The R1 button toggles the emergency state of the robot. Pressing R1
while flying will stop the rotors immediately. If the LED beneath the
rotors are red (for example, after a crash), press R1 to reset the drone.
1
The L1 button starts the motors of the quadrocopter. It also works as a

deadman switch so that the robot will land if you release it during flight.
The quadrocopter will ascend to one meter above ground and tries to
hold this position.
The left stick can be used to control the roll and pitch angle of the
drone. Keep in mind that these velocities are given in the local frame of
the drone!
The right stick controls the yaw-rate and the altitude.
The select button can be used to switch between the two cameras. This
can also be done by executing rosservice call /ardrone/togglecam.
The triangle button can be used to switch off the on-board position stabilization: Per default, the drone hovers (i.e. stabilizes its horizontal
position by keeping vx and vy at zero) when you do not touch the left
control stick. It even actively decelerates when letting go of the left control stick.
Pressing and holding the triangle button will switch this off, i.e. give you
direct control over roll and pitch at all times note how this leads to
rapid drift in horizontal position.
(c) First Flight: Connect the drone as in the very first exercise sheet. Display the
camera image using
rosrun image view image view image:=ardrone/image raw
Fly a short round in the robot lab and practice your flying skills. Take a
group picture of your team and add it to your report.
(d) Graph Visualization: Use rxgraph to visualize the running nodes and used
topics. Take a screenshot, and attach it to your report.
(e) Record a .bag File: [put marker on ground]. Switch to the bottom camera
(the AR.Drone only streams one of the two camera images to the PC), and
use rosbag to record approximately a 30s-flight. The marker should regularly
be visible in the camera image, but should leave it temporarily. You will need
to record at least the camera image topic, the camera info topic, and the
ardrone navdata topic, i.e.:
rosbag record /ardrone/navdata /ardrone/image raw
/ardrone/camera info -O flight.bag
Exercise 2: Extended Kalman Filter
In this exercise, you will learn how to use a Kalman filter to estimate the pose of the
robot from a bag file or live data. You can either use the bag file from the website, or
2
your own bag file recorded in Exercise 1. For the screenshots in your report, please
use the bag file from our website. Exit all running nodes and disconnect the
drone and joystick, you will not need them for this exercise.
We provide you with a C++ framework of an extended Kalman filter for which you
will have to implement the correction step. For simplicity, we model the quadro>
copter only in the 2D plane, i.e., its state at time t is described by xt = xt yt t .
The prediction function is given by the odometry from the previous exercise sheet
using the measured yaw angle and horizontal velocities.
(a) Download the C++ framework for the Kalman Filter from
git://github.com/tum-vision/visnav2013 exercise2.git
Rename the folder to ex 2 and build it using rosmake.
(b) Download and compile the visual marker detection node from
http://robotics.ccny.cuny.edu/git/ccny-ros-pkg/ccny vision.git
Launch the marker detection node using the launch file provided in ex 2.
(c) Start rviz and add a grid, tf display, a marker-display and a markerArraydisplay as in the previous exercise sheet. Remember to change the base coordinate system to /world, and to select the correct topics.
(d) Start the Kalman filter by running rosrun ex 2 visnav 2. Replay the bag file
using rosbag play and watch the result in RVIZ; Take a screenshot after
a couple of seconds with the covariance ellipse. The Kalman filter also
publishes the estimated pose on /ardrone/filtered pose. Visualise the full
estimated two-dimensional trajectory from the bag file using
rxplot -p 48 /ardrone/filtered pose/linear/x:y
i.e., restart the playback after starting rxplot. Take a screenshot and
attach it to your report.
(e) Assume that the quadrocopter drifts more (say two times more) than this
in the global x-direction, for example, because there is strong wind in this
direction. Modify the process noise matrix Q accordingly in the source code,
re-run the experiment, and take another two screenshots.
(f) What would be a good way to determine the noise values empirically? Describe briefly an experimental setup that could be used to determine
these values.
(g) The framework detects markers in the environment and provides (in the case
>
of a detection) an observation z = x y
relative to the frame of the
quadrocopter. Specify the observation function z = h(x) and compute
.
its derivative (Jacobian) H = h(x)
x
(h) Implement the correction step in the Kalman filter using the observation function and the Jacobian. Take care when you compute the difference between
angles (as required for the computation of zt h(t ) in the correction step).
One simple solution is to normalize the angle afterwards, for example, using
normalized = atan2(sin(), cos()).
(i) use rxplot again to visualize the - now corrected - estimated two-dimensional
trajectory from the bag file, take a screenshot of the trajectory, and add
it to your report.
(j) From the screenshots, estimate roughly how far the pose-estimate drifted
without visual markers over the 48s flight (in meters).
Submission instructions
A complete submission consists both of a PDF file with the solutions/answers to the
bold questions on the exercise sheet and a TGZ (or ZIP) file containing the source
code that you used to solve the given problems. Make sure that your TGZ file
contains all files necessary to compile and run your code, but it should not contain
any build files or binaries (make clean, rm -rf bin). Please submit your solution
via email to visnav2013@vision.in.tum.de.

Sheet 2: Topic: Motion Models and Robot Odometry

Uploaded by

Copyright:

Available Formats

Sheet 2: Topic: Motion Models and Robot Odometry

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sheet 2: Topic: Motion Models and Robot Odometry

Uploaded by

Copyright:

Available Formats

Visual Navigation for Flying Robots

D. Cremers, J. Sturm, J. Engel, C. Kerl

Computer Vision Group

The L1 button starts the motors of the quadrocopter. It also works as a

You might also like