Tip:
Highlight text to annotate it
X
Follow us with InsideScience
and take a peek over the shoulders of researchers.
In this episode: "What do we need for a humanoid robot?"
Mechatronics. Learning. Interaction
Hearing about robots we associate images
that trigger strong emotions,
but most of us have never seen a robot in reality.
Industrial robots revolutionized production processes in the last 50 years.
Today we want robots to get out of the factories
and closer to people as modern helpers in private homes.
How must a machine be designed
that has to move in an environment
built for humans,
that is supposed to help humans
and that can be controlled by humans as easily as possible.
At the Karlsruhe Institute of Technology researchers of many different fields are working
on the construction of a humanoid robot.
The goal of the Collaborative Research Center „humanoid robots" was to develop
a multimodal, learning, co-operating robot.
It was supposed to have a humanoid, a human-like form,
move in a natural, human-like way and to be interactive above all else.
Now we are capable
of building robots that, in contrast to the industry robots
that statically complete tasks,
can be described
as humanoid.
This is an ancient human ambition
that can be achieved today, for about the past 20 years.
On the other hand no one has any household robots
that clean up the table,
do the dishes
or load the dishes into the dishwasher.
We still have to do all that ourselves.
If we want a flexible machine
that can pretty much execute any movement after instruction by a human
we are still a long way away from that.
„Humanoid" means acting like a human.
If we want that
we need to transfer
certain human attributes
to the robot.
First of all it has to be able to walk, to grip things, it has to feel,
it has to move.
And of course it has to see, that means it has to record
what is happening around it.
We have to see
how our environment is actually arranged,
what can we do and how we can give instructions to a machine
so that the machine can understand
what we expect of it.
A robot that helps with household
tasks has to act like a human.
It has to perceive the world like a human
and understand the instructions of a human.
How do the research fields work together?
Robotics in general is an inter-disciplinary field.
That means that we have to combine the knowledge and expertise of the seperate fields
and create synergies.
If we want to build a robot,
first the mechanical engineers have to build a machine.
First they build the body.
That means that we need the fields of mechanics,
mechatronics and mechanical engineering.
Furthermore we have the interface to the sensors,
the preprocessing of the signals and their interpretation.
A robot that is supposed to see needs cameras.
One that is supposed to hear needs microphones.
Also they need arms and legs.
Additionally we have the conversion into movement and actions.
In order for the robot to move
we need motors,
generally electric motors.
That is why we need the field of electrical engineering.
On the other hand we need seeing machines,
that means the automated conversion of picture, sound, video, language
and acoustics or rather the processing of signals.
We also need the ideas of the humanities and social sciences.
Of these specifically the sports science field is integrated with us.
In robotics the mechatronics, the electrical engineers, the computer scientists,
the mechanical engineers and even the sports scientists come together.
What role do the three main columns of mechatronics, learning and interaction play?
The robot and the car are prime examples for the basic idea of mechatronics.
The basic idea of mechatronics is to work together with multidisciplinary fields.
That means that mechanical engineering, electrical engineering,
automation engineering and computer science are combined.
Not all situations in everyday life are predictable.
That means that the robot needs knowledge about what it can do, about the consequences of its actions
so that it can estimate their outcome.
Learning from the human brain is important.
how do we recognize it? How can we automatically learn from that knowledge?
A robot learns in a variety of ways through experiments, through trial and error.
But it can also be taught by a human,
or it watches humans executing actions.
Learning through knowledge of the world and its environment,
that means knowledge about the components
that are found in a kitchen for instance
and the functionality of the relevant appliances is necessary for a robot.
At the same time it has to observe the human,
what he does, what he needs,
how he reacts,
so that the robot can adapt to the situation in the environment.
Imagine for instance
that I'm sitting in the living room and say: "I'm thirsty now!"
Then the robot should be able
not just to say:„That is very nice!",
But it should say:
and possible even tactile signals.
„Should I get you a beer, or do you want a glass of milk?"
This differentiation has to combine
the comprehension of what the person said
and the implicit request for something to drink.
Here we have, I can say without much ceremony, the best hand in the world.
We also have developed the legs so far
that we can say
that we have legs that work very well.
These four fields build the body, measurement technology,
sensor technology, control technology, data processing,
computer science and the realization in actuation, so that it can move.
Those are the four central fields of mechatronics.
We call this multi-modality
when we combine different signals.
We combine video, acoustic
All that happens in the interaction.
We try to understand our environment through interaction
and this allows the robot to react to its surroundings accordingly.
The robot is supposed to adapt to its surroundings
and react to it in a meaningful way.
What was the biggest challenge?
The biggest challenge of the Collaborative Research Center was
that we had underestimated the complexity of the mechatronics
and the integration of all components.
The biggest challenge with such a robot from the perspective of interaction
is to come to terms with all the uncertainties and variables of the surrounding environment
and the interaction with humans.
We are virtually always guessing:
was it this or was it that?
Maybe he just said „table" or maybe „cable".
We solve this problem
with the help of additional knowledge.
In this sentence maybe „cable" is more plausible,
in another sentence „table" makes more sense.
That means that we understand sentences
because we combine different sources of information.
I believe the biggest challenge is that, and you need to know that,
in a collaborative research center every year twenty to twenty four
young people that are each working on single projects come together,
that they co-operate, that they exchange their ideas,
that they use the creativity that young people especially have
when they come directly from school or from university.
That they use this here
so that we really build
what these young people imagine.
So building a robot is not easy at all.
Especially because the surroundings keep changing.
And in the end the young students should be able to use their creativity.
And the future?
If you look at robots today they are still very slow.
I think that future robots will be more precise and will perform better.
Today the sentences spoken to the robot have to be very precise,
partially still in written form.
The interaction is limited to pointing gestures
and gripping processes
while more complex action sequences,
like repairing a device or manufacturing an object,
making a cupboard,
a chair or a table
are very difficult and need much effort to be transferred onto a robot.
It would be interesting
if a robot could learn from another robot.
That means that it could transfer its acquired knowledge to another machine,
if they could exchange experience, algorithms
or visual and perceptive skills.
This can be raised even further
when many robots exchange information
or if robots generally surf the web and acquire the information
they need worldwide.
The KIT series InsideScience takes a peek over the shoulder of researchers
of the Collaborative Research Center.
Take a look at the other episodes
and see how a robot is designed, how it learns,
how it interacts with humans
and which social aspects have to be discussed.