IEEE Robotics & Automation Magazine - December 2022 - 143

on the robotic system. According to Abbeel, " For Covariant,
we actually spent the first half of the year meeting with 200
other companies and asking them, 'If they could get a smart
robot in their facility, what would that smart robot be doing?' "
Can researchers in ML and robotics prepare, already during
their academic careers, for a potential future transition
from novelty to customer needs? First of all, gathering experience
in teams working on integrated systems is certainly
beneficial. As Tenorth explains, " One important thing I
learned also during my Ph.D. time is how to build a system,
rather than only one part of a system. Other groups focused
on individual algorithms, but we did things such as making
pancakes with a mobile manipulator. In these cases, it was
important that components are put together in a way that it
results into an integrated system. I think that's the most
important learning experience. "
Second, open source publishing of code and the choice
of license should be done with foresight. More companies
continue to use and contribute to open source, i.e., according
to Tenorth, " Magazino also releases open source software
to the outside. So we try not to just consume but also
to give back to the community. " The choice of license has a
big impact on potential innovation paths. Tenorth goes on
to say, " The Apache, BSD (Berkeley Source Distribution),
and MIT (Massachusetts Institute of Technology) licenses
are good. Copyleft licenses such as General Public License
are problematic because if you link against them in your
system, the copyleft clauses apply to all other code that
depends upon it. This implies having to share your source
code with your customers, which is something that investors
are very sensitive about. "
Finally, when spinning off the start-up, " it's detrimental if
you come up with a solution and try to find a problem that
fits. It's better to see which problems need to be solved and
which solutions are appropriate, " says Tenorth. In this context,
Konidaris recommends to " make sure to include some
hard-nosed industry people in your spin-off. "
Reliability Is Key
What sets robots apart from other digital agents is that they
physically interact with the real world. This radically changes
the requirements on reliability and safety, and these requirements
are not always compatible with trial-and-error ML
and extrapolation to novel situations. " Learning is very tricky
for an industrial robot that literally weighs a ton and is working
on a production line where the costs of downtime are in
the tens or even hundreds of thousands of dollars a minute, "
says Konidaris. In this context, Abbeel corroborates the previous
remarks by Tenorth on the divergence of academic
requirements (novelty) and industrial requirements (reliability):
" At Berkeley, we try to do things that have never been
done before. If we can do it for the first time, that's great. It
doesn't matter if it only succeeds 50-60% of the time. It's
already a big deal because it was never done before.'
Which strategies are used to ensure reliability in robotic
systems that use AI methods, especially data-driven ML?
Professor / Assistant
Professor (Tenure Track)
of Robotics
→ The Department of Mechanical and Process
Engineering (www.mavt.ethz.ch) at ETH Zurich
invites applications for the above-mentioned position.
→ Successful candidates must demonstrate an excellent
international record of research accomplishments
as robotics engineers and scientists. The
new professor is expected to establish an ambitious,
world-class research program in the fast-evolving
field of robotics. We encourage applications from
scientists and engineers from the entire spectrum of
robotics, from modelling and design to perception,
control, and robot learning. They should demonstrate
a core-area of scientific expertise and solid theoretical
foundation with a desire to apply this knowledge
to solve important challenges in fields like service
and industrial robots, aerial or underwater robotics,
construction, precision agriculture, or mobility and
logistics. Specific areas of research may include
(but are not limited to) intelligent perception and
scene understanding, manipulation and enhanced
human-machine interaction, robot and sensor
design, soft robots, modeling and control, or robot
learning. Successful candidates should hold a PhD
degree or equivalent in engineering and have an
outstanding record of accomplishments in robotics.
Furthermore, a strong motivation and indisputable
commitment to undergraduate (in German or
English) and graduate (in English) student teaching
is expected.
→ Assistant professorships have been established
to promote the careers of younger scientists. ETH
Zurich implements a tenure track system equivalent
to that of other top international universities. The
level of the appointment will depend on the
successful candidate's qualifications.
→ Please apply online: www.facultyaffairs.ethz.ch
→ Applications should include a curriculum vitae, a
list of publications and projects, a statement of
future research and teaching interests, a description
of the leadership philosophy, three key publications,
and a description of the three most important
achievements. The letter of application should be
addressed to the President of ETH Zurich,
Prof. Dr. Joël Mesot. The closing date for applications
is 31 December 2022. ETH Zurich is an equal
opportunity and family-friendly employer, values
diversity, and is responsive to the needs of dual
career couples.
http://www.mavt.ethz.ch http://www.facultyaffairs.ethz.ch
IEEE Robotics & Automation Magazine - December 2022

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2022
