Avionics News February 2023 - 32

AVIDYNE AND DAEDALEAN PARTNER
Continued from page 31
The company brings its expertise in hardware design,
manufacturing and certification to this product development
effort. Schwinn sees artificial intelligence/machine
learning as a " core, fundamental technology of the next
generation of avionics and aircraft. Specifically, I am
talking about using convolutional neural networks that
have been trained on some kind of function or image, not
a traditional image processing system. " Convolutional
neural networks are a type used to analyze visual imagery.
When Schwinn first looked at Daedalean's website, he
found it claimed the company planned to build artificial
intelligence that could do anything a human pilot could do.
" As a human pilot, I was thinking, well, you know, maybe,
but I don't know, " Schwinn said. " But I
was definitely intrigued enough to send
them a note to their generic mailbox. I
said I'm a traditional avionics guy and
interested in what you are doing. I don't
know; it seems a little far-fetched. Let
me know if you want to talk. "
Luuk van Dijk, co-founder and chief
executive officer of Daedalean, replied,
and Schwinn decided to visit the
company in Zurich. It was then that the
idea of a partnership gelled. " I realized
what seems obvious today, " Schwinn
said. " On the way to doing everything
that human pilots can do, there were a
bunch of really interesting applications
that can be done. And this is similar to
what's been done in cars. "
Avidyne is in the business of certifying
software using established traditional
means, but convolutional neural network was going to
be a different kettle of software than certifying more-commonly
used avionics software. CNN is analogous to how
neurons connect in the human brain. CNN can be trained
to do image analysis using a deep learning algorithm.
Schwinn noted that software, like other things on the
aircraft, meet different levels of safety depending on how
critical the function is. Avidyne uses different levels of
process development to reach different levels of design
assurance and to deliver different levels of reliability.
Failure rates can be forecast with statistics. " For neural
networks, having a statistical safety validation mechanism
is not actually that big a leap from where software has
been developed if you're willing to look at things a little
bit differently, " Schwinn said. " Avidyne has put a ton of
effort into this. "
32 avionics news * february 2023
AVIDYNE AND
DAEDALEAN
FORMED A PARTNERSHIP
IN 2020 THAT TAKES
ADVANTAGE OF
AVIDYNE'S LONG
EXPERIENCE IN
CERTIFYING AVIONICS
FOR BUSINESS AND
GENERAL AVIATION WITH
THE FEDERAL AVIATION
ADMINISTRATION
AND THE EUROPEAN
UNION AVIATION
SAFETY AGENCY.
The partnership formed two years ago has Daedalean doing
neural network technology and training for applications,
while Avidyne brings the hardware platform and the aircraft
installation expertise to get the product into the field and supported.
The team is working with the FAA with concurrent
validation with EASA to get PilotEye certified.
PilotEye will be as good as the best human eye that exists
in spotting aircraft, according to the partners. The system
will have three cameras initially and a fourth for special
applications in the future. The forward-facing camera has
a 60-degree field of view, and the two side-facing cameras
both have 80-degree fields of view. Due to some overlap,
the total panorama is 210 degrees; there is not a rear-facing
camera, but that could be a potential use of a fourth camera.
Helicopter operators also might want to have a fourth camera
facing down to identify helipads or fields where an
emergency landing can take place.
Initially, PilotEye will use visual
cameras to detect aircraft and some types
of obstacles. Once that capability is
certified, Schwinn said it will " provide
another level of situational awareness of
non-cooperative traffic. "
The beauty of having cameras involved
is they survey the full 210-degree field of
view constantly and are never distracted,
while a human pilot can only focus on
one area when not looking down for an
instrument crosscheck or to tune the radio.
Schwinn said larger repair shops should
be able to qualify to design and install
the camera system, which requires proper
placement to avoid being blocked by a
propeller, for example. The camera system
must be calibrated after it is installed.
Schwinn added that most repair shops
should be able to work on maintaining the PilotEye system.
It's like having a robotic co-pilot if the only job is to
keep an eye out for other traffic, a task that can be a lifesaver.
This co-pilot doesn't have the limitations of the Mark 10
human eyeball; it can see farther and detect and identify the
type of target faster than a human. Because cameras can see
non-cooperative traffic not squawking to air traffic control
or not broadcasting position on ADS-B Out, Pilot Eye may
see the one that is under no surveillance.
Regulators are going to require that an autonomously piloted
aircraft be able to fly like it has a human pilot aboard
to be able to see and avoid other aircraft. To cooperate with
the existing air traffic control system, it will be necessary
to accept a visual clearance and maintain visual separation
with a designated aircraft. Once the PilotEye system is certified,
it will be possible to add other sensors such as long
Avionics News February 2023

Table of Contents for the Digital Edition of Avionics News February 2023
