IEEE Robotics & Automation Magazine - March 2018 - 90

system is extensively evaluated both indoors using a motioncapture system and outdoors using a laser tracker while
performing hover and step responses and trajectory-following
tasks in the presence of external wind disturbances. We
achieve root-mean-square (rms) pose errors of 0.036 m with
respect to reference hover trajectories. We also conduct
relatively long distance (.180 m) experiments on a farm site,
demonstrating a 0.82% drift error of the total flight distance.
This article conveys the insights we acquired about the
platform and sensor module and offers open-source code with
tutorial documentation to the community.

are relatively expensive for research purposes, and part
replacements are difficult due to limited retail services.
For VTOL MAVs to become more widespread, they will
require lower costs and more easily obtainable parts. Recently,
the DJI Matrice 100 MAV (Shenzhen, China, https://www.dji
.com/matrice100) (Figure 1) has been introduced as a commercial platform with parts available at local retailers. Developers can
access sensor data, e.g., from the IMU and barometers and send
commands to the low-level attitude controller through the SDK
[12]. Although manufacturer documentation is provided, there
are limited scientific resources detailing aspects like attitude
dynamics, the underlying autopilot controller structure, and
Overview of VTOL Microaerial Vehicles
input command scaling. This information is critical for subseDuring the past decade, VTOL microaerial vehicles quent position control strategies, such as MPC.
(MAVs), which use counterrotating rotors to generate
A VI sensor is a favorable choice for aerial robotics due to
thrusting and rotational forces, have gained renown in both its light weight, low power consumption, and ability to recover
research and industry. Emerging applications, including unknown scales (monocular camera). This sensor suite can
structural inspection, aerial photography, cinematography, provide a time-synchronized wide field of view (FoV) image
and environmental surveillance, have spurred a wide range (≈133°) and motion measurements. There is an impressive
of ready-to-fly commercial platforms whose performance research-grade VI sensor [13] providing time-synchronized
has steadily improved in terms of flight time, payload, and and calibrated IMU-stereo camera images with supporting
safety-related smart features, allowing for more stable, easi- documentation and device drivers. Unfortunately, this sensor
er, and safer pilot maneuvers. However, a key challenge is is relatively expensive, and its production was discontinued.
directly adapting commercial platforms [1] for tasks requir- The Intel ZR300 (http://click.intel.com/realsense.html) has
ing accurate dynamic models, high-performance control- recently emerged as a compelling alternative. It is an affordable
lers, and precise, low-latency state estimators, such as and mass-produced module that enables applying the same
obstacle avoidance and path planning [2]-[5], landing configuration and calibration parameters to other sensors with
on moving platforms [6], object picking [7], and precision low performance penalties. Table 1 summarizes the total cost
agriculture [8].
for the VI sensor, MAV, and personal computer (PC).
Research-grade MAV platforms can alleviate these issues.
In this article, we address these gaps by using a ready-toFor instance, Ascending Technologies of Krailling, Germa- market affordable VTOL platform and a VI sensor to perform
ny, provides excellent products [9], [10] for advanced aerial system identification, sensor calibration, and system integration.
applications [11] with an accompanying software develop- The system can autonomously track motion commands with
ment kit (SDK), scientific resources (including self-con- high precision in indoor and outdoor environments. This ability
tained documentation), and online support. However, they is fundamental for any high-level application, enabling researchers to build custom systems without
tedious tuning procedures. Moreover,
Intel NUC i7 Processor
we provide self-contained documentation to cater for setups with different
inertial moments and sensor mount
configurations. The contributions of this
system article are
IR Reflector for
Motion-Capture Device
● a delivery of software packages [including a modified SDK, nonlinear
DJI M100 Platform
MPC (nMPC), calibration parameters, and system identification
tools] with accompanying documentation to the community
Corner Reflector for
● an evaluation of the control and
Intel RealSense ZR300
Leica Laser Tracker
state estimation performance in difVI Sensor
ferent environments
● a demonstration of use cases adapting our approach to build custom
research platforms with a step-bystep tutorial available at https://goo
Figure 1. A Matrice 100 VTOL MAV quadrotor platform with a downward-facing VI sensor. The
infrared (IR) reflectors and prism are mounted for obtaining the ground truth.
.gl/yj8WsZ.
90

IEEE ROBOTICS & AUTOMATION MAGAZINE

March 2018

https://www.dji.com/matrice100 https://www.dji.com/matrice100 http://click.intel.com/realsense.html https://goo.gl/yj8WsZ https://goo.gl/yj8WsZ

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - March 2018