Episode: 1798
Title: HPR1798: Machine learning and service robots.
Source: https://hub.hackerpublicradio.org/ccdn.php?filename=/eps/hpr1798/hpr1798.mp3
Transcribed: 2025-10-18 09:24:56

---

This is HPR episode 1,798 entitled, Machine Learning and Service Robots.
And in part of the series, Interviews, it is hosted by MiWid and is about 9 minutes long.
The summer is, Interview with Prof. Dr. Wolfgangertel at the 2014 Maker World in Germany.
This episode of HPR is brought to you by an Honesthost.com.
Get 15% discount on all shared hosting with the offer code HPR15, that's HPR15.
Better web hosting that's Honest and Fair at An Honesthost.com.
Hello Hiccup Public Radio, this is Michael, also known as Mervy.
I want to finally bring you a short interview from the Maker World 2014, which was end
of June last year alongside the HEM Radio, Europe's biggest amateur radio fair in Friedrichshafen,
Southern Germany.
It was the first time with the Maker Fair adjacent to the amateur radio conference and
it turned out quite well.
One complete exhibition hall was dedicated to the Maker World and the Raffenzburg Weingarten
University of Applied Sciences had a big booth there, but they demonstrated machine learning
and assistive robotics.
I myself was captivated when I was just strolling along the main stage where Prof. Wolfgang
ertel was just giving one of his introductory talks.
There were lots of kids in the audience and he did a great job in explaining the fundamentals
of robotics and machine learning, starting out by explaining the differences between humans
and machines, what other things each one can do very well and what other tasks that pose
more of a challenge to either one.
He went on about classical programming approaches and how they fail in certain situations.
That made it obvious that we have to enable the machines to adapt to various environments
to be able to dynamically learn things and extend their capabilities beyond what the program
has initially put in them.
As an intuitive example of machine learning, he showed a crawling robot with a pair of wheels
in the simple arm controlled by two servos, which is able to lift the device up or if
controlled correctly, drag it along or push it away.
He showed some videos, which I have links to in the show notes, where we could watch
the robots learning from scratch how to move forward on different surfaces.
It was interesting to see how the same algorithm can come up with completely different movement
patterns depending on the surface or even just for different runs of the same program
on the same surface.
Now with the basic concept introduced how machines can learn things, they were not initially
programmed to do.
He went on talking about application of those techniques in service robots, which are meant
to assist people in their daily living.
He talked about their prototypes of service robots, Kate and Marvin, which they had both
live at their booth with demonstrations and I have links to their home page in the show
notes where you can see videos of both of them.
At that point the new robot Marvin wasn't able to do very much because it was so brand
new they had just gotten the parts days before the event and they decided to bring it along
and to show it off and it really is impressive to go have a look at the videos.
At the end of his presentation he showed a short video of some dog-like robot carrying
military supplies over rough surfaces, commenting that whatever noble goals the researchers
might have about human assistive robots and so on.
No one should fool themselves about where the money is and where this technology will
be also used.
He did not overemphasize it but I think it was a fair point to make for the overall picture.
So far for setting the stage so let's jump into the interview.
We are at the Maker World in Friedrichshafen and I am talking to Wolfgang Erdel and can you
just summarize what we are seeing here, what you are doing here.
Yes I mean our competence is machine learning.
This is a subfield of artificial intelligence and what you can see here is applications
of machine learning to service robotics.
So our goal is to make service robots learn their tasks and this is a big advantage compared
to classical programming because these tasks that a service robot has to solve are extremely
difficult and it's almost or even really impossible to program such tasks and our solution
is we apply machine learning techniques which is kind of the same as we humans do.
A kid learns its behavior and it's not being programmed and this is what we do.
And yeah we can do it quite successfully.
For example our service robot Kate here is able to learn by demonstration.
I mean this is quite obvious.
A human trainer demonstrates the task a couple of times, maybe two or three times.
The robot watches this human trainer and after that the robot can I apply some generalization
and machine learning algorithms reproduce the task.
Okay I saw you talked before where you explained how there are problems which we cannot approach
by normal programming, why we need machine learning to be able to adapt to varying environment
and such subject and can you talk a little bit about your institution, about your department
or collaborations you have.
This is the Institute for Artificial Intelligence at Hochschule Ralfensburg Weingarten.
And there we have a couple of different research projects.
Our latest project is on assistance robots for physically disabled people.
And for this project we just built a new service robot which we present here at the
Makerworld for the first time.
And this is really a very innovative robot by example because this robot is able to grab
objects from the floor in any height up to about two meters.
So this robot can reach any position in a typical living environment.
And so what we have at the moment is the hardware of the robot and some software.
And we are now starting to implement all the service robots software on this new robot.
We have of course a lot of software modules already from its predecessor.
The predecessor robot was Kate and you can see on the internet videos of Kate and we
of course we are now going to adapt all this to the new robot which is called Marvin.
Yeah I saw the live presentation of Kate before fetching a cup of coke and so on.
And it was obvious that there's a high potential but also it's still tricky where all the disciplines
from voice recognition to the video processing have to interact to be successful.
Yes you are absolutely right this is an extremely difficult, extremely complex software engineering
task.
Even though we are able to learn particular tasks there still is of course the software
engineering effort to bring all these disciplines like image processing, planning, artificial intelligence,
machine learning.
We have to bring all these modules together.
This extremely complex software engineering process costs of course very much human resources
very much main power and this is actually the problem at the moment in the service robotics
research community.
There are very many universities all around the globe but all these university institutes
they are not very large they are typically like between two and ten researchers and this
is by far too small for doing all the engineering for a complex robot and therefore really making
a stable product out of this that works in everyday environments in changing scenarios
such a product cannot be developed by the universities and it's actually not the purpose
of a university to do this.
But we do at the universities we do research and we show that it works and we are finished.
And now comes a big company or should come that puts a lot of money into such a project
and would then be able to deliver a good product at an affordable price.
And I was waiting for such a big player to do this for many years and quite recently
about half a year ago now there is rumors that Google is going to develop the first
commercially available affordable service robot and I mean the indications are quite clear
because Google bought all the premium robotics companies and so I guess that in about five
years there will be such products on the market.
So where can people find on the web to see, to learn about your new robot, your prototype?
Yeah I mean you just saw from the website of our institute which is ikki.hazvinegarden.de
Okay I will have a link in the show notes.
So thank you very much Professor Wolfgang Ertl and thanks for talking to us.
So far for now my apologies to the HBR community and to Professor Wolfgang Ertl
for taking so long to put this out and my sincere thanks to him for being such an approachable
guy and for making it easy for me to take my 900 ticket in front of him and record this interview.
And if you are listening to this episode the moment it hits the HBR feed we are just days away
from 2015's event as the hem radio and the maker world 2015 will take place this weekend.
Please don't take me as a good example and do your shows in a more timely fashion.
Here you are hopefully sooner. Bye for now.
You've been listening to Hacker Public Radio at Hacker Public Radio dot org.
We are a community podcast network that releases shows every weekday Monday through Friday.
Today's show, like all our shows, was contributed by an HBR listener like yourself.
If you ever thought of recording a podcast then click on our contributing to find out
how easy it really is. Hacker Public Radio was founded by the Digital Dove Pound and the
Infonomicom Computer Club and is part of the binary revolution at binrev.com.
If you have comments on today's show please email the host directly leave a comment on the website
or record a follow-up episode yourself. Unless otherwise status.
Today's show is released on the Creative Commons,
Attribution, ShareLife, 3.0 license.