1. Autonomous Systems Laboratory
Chang Liu, Supervisors: Dr Stephen D. Prior, Prof James Scanlan
Faculty of Engineering and the Environment
http://www.southampton.ac.uk/engineering/postgraduate/research_students/cl21g11.page?
email: cl21g11@soton.ac.uk
Boldrewood Campus, Burgess Road, University of Southampton, Southampton, S016 7QF, UK
Visual Control of a Small Quadrotor in
GPS-denied Environments
Abstract — This project focuses on the design of quadrotor control
architecture, based on commercial off-the-shelf (COTS) components. The
aim of the project is to provide a test-bed for vision-based autonomous
navigation systems in GPS denied environments. In the current stage, by
utilizing the PX4FLOW sensor, the quadrotor is capable of horizontal
velocity control and altitude hold, as well as real-time 3D feature mapping
in onboard computer and visualisation in ground station. And thanks to
the visual-inertial fusion algorithm, the visual scale of the feature map is
sufficiently recovered. The vehicle is also equipped with a forward facing
analogue camera for first person view (FPV) capability.
rescue, surveillance, exploration, agriculture, monitoring and military
applications in both indoor and outdoor environments.
Over the last decade, Global Positioning System (GPS) has been the key to
enabling autonomy of UAVs. However, recently, due to the proven
weakness of the GPS signal and rapid development of onboard sensing
and computation capability, there has been growing interest in developing
and researching alternative navigation methods for UAVs in GPS denied
environments. The successful implementations will not only improve
system robustness under GPS failure, but also enable a new range of
applications out of GPS coverage (typically in military, disaster, indoor and
urban environments) , as shown in Fig. 1.
A camera has significant advantages over other sensors, such as low mass,
low power consumption, low price, adjustable field of view (FOV), high
accuracy, additional colour information and long range.
Fig. 1. GPS-denied Application Scenarios.
The fully customised autopilot control and sensor fusion algorithm was
developed in Arduino compatible open source electronics, including servo
controller, Inertial Measurement Unit (IMU) and Xbee radio, and an
interface printed circuit board (PCB) designed in-house. Besides, a
forward facing analogue camera together with a 5.8 GHz video transmitter
are equipped to provide first person view (FPV) capability, as shown in Fig.
2.
Platform Hardware Introduction
The first subplot in Fig. 6. shows the
velocity command-response in y-axis.
The second subplot shows the
intermediate roll angle command-
response signal between velocity
controller and attitude controller. The
third subplot shows the altitude
command-response and the fourth
subplot shows the yaw hold.
Conclusion
This poster has shown the design and
implementation of an autonomous
quadrotor platform suitable of GPS-
denied navigation and control based on
vision approaches. The flight test
results have shown an acceptable
velocity control performance, even with
significant wind. The fully customized
design makes it easy to integrate new
sensors and manipulating controller.
Future work includes developing safety
procedures for tracking failure;
implementation of position controller
to involve SLAM in the loop.
References
[1] S. Madgwick, “Estimation of IMU
and MARG orientation using a gradient
descent algorithm,” Rehabilitation
Robotics (ICORR), no. 1945-7898, pp. 1
– 7, 2011.
[2] C. Forster, M. Pizzoli, and D.
Scaramuzza, “SVO: Fast Semi-Direct
Monocular Visual Odometry,” Proc. IEEE
Intl. Conf. on Robotics and Automation,
2014.
Fig. 3. Platform Hardware Interaction Diagram.
Platform Software Introduction
The inner loop operates at 333 Hz to control the vehicle attitude. The
attitude control algorithm performs directly on SE(3), thus is able to
effectively recover from any initial orientation. The attitude measurement
is obtained by fusing the data from inertial measurement unit (IMU), which
includes a gyroscope and an accelerometer, with gradient decent algorithm
[1]. The outer loop operates at 100 Hz to control vehicle altitude and
horizontal velocity, which are obtained from the downward facing
PX4FLOW camera.
Introduction
The quadrotor is one of
the most popular subset
of UAVs. Because of its
ability of agile maneuver,
vertical take-off and
landing (VTOL) and stable
hover, it is commonly
agreed to be an ideal
candidate for search and
Main
Controller
Servo
Controller
XBee
RC receiver Velocity and
Altitude
Sensor
Interface
Arduino
IMU
Additional Vision Computer
ESCs
UART1
UART1
UART2
400 KHz I2C
100 KHz
I2C
UART3
333 Hz PWMs
USB
Manual RC
controller
PC Ground
Station
Global Shutter
CameraUSB
The design also includes an additional downward facing optical flow sensor
(PX4FLOW camera) for horizontal velocity estimation and vehicle altitude
estimation, and a separate Linux embedded computer (Odriod-U3) with a
downward facing global shutter monocular camera for Simultaneous
Localization And Mapping (SLAM) vision algorithm development. The
interaction between different electronic hardware components are shown
in Fig. 3.
Fig. 2. Front and bottom-up view of the quadrotor.
Fig. 4. Onboard SLAM algorithm (SVO). [2]
The Robotic Operating System (ROS) is installed in Odroid-U3 single board
computer as the central software package coordinator. As shown in Fig. 4.,
the state-of-the-art open source monocular simultaneous localisation and
mapping (SLAM) algorithm (Semi-direct Visual Odometry [2]) operates in
Odroid-U3 with the downward facing global shutter camera onboard the
vehicle. The algorithm is efficient enough to estimate the vehicle 6 degree
of freedom state at 40 Hz, even in fast manoeuvre. Moreover, a 13 state
extended kalman filter (EKF) is implemented to fuse inertial measurement
with the SLAM state estimation, which not only gives much more robust
estimation at 100 Hz, but also recovers the visual scale factor for the
monocular SLAM.
Fig. 5. Operating View.
Results
Outdoor flight test results are shown in Fig. 5. and Fig. 6. to demonstrate
the control performance and validate the theory. Note that the weather
forecast states 8 mph wind speed at the time of testing.
Fig. 6. Test Result.
Fully
Customized
Autopilot
FPV Camera
Odroid-U3
Onboard
Computer
PX4FLOW
Optical
Flow Sensor
Global
Shutter
Camera
Historical
Path
6 DOF
Quadrotor
State
Estimation
Natural Features
being reconstructed
in 3D map
Natural Features being tracked in 3D
map (re-projected back in video)