WEBVTT

00:00:00.433 --> 00:00:02.235
Tilman, thank you so much for
joining us.

00:00:02.235 --> 00:00:04.187
What is physical intelligence?

00:00:04.788 --> 00:00:08.541
Well, physical intelligence
refers back to the ability to

00:00:08.541 --> 00:00:12.345
observe, understand, and
interact with the physical world.

00:00:12.595 --> 00:00:13.747
Let me give you an example.

00:00:14.080 --> 00:00:16.916
Let's assume you would ask to
get something for drink.

00:00:17.400 --> 00:00:21.438
So as a human being, I would
observe our surrounding.

00:00:21.521 --> 00:00:23.323
I would identify a bottle.

00:00:23.356 --> 00:00:27.961
I understand that a bottle is,
for example, a drinkable liquid

00:00:27.961 --> 00:00:28.828
for a human.

00:00:29.162 --> 00:00:32.232
And I also have the ability to
grab the bottle and to fill it

00:00:32.232 --> 00:00:32.882
into a glass.

00:00:32.882 --> 00:00:37.020
So it sounds quite easy, which
is easy for a human is a quite

00:00:37.020 --> 00:00:39.355
complex task for a
robotic system.

00:00:39.756 --> 00:00:42.242
And this is what physical
intelligence means.

00:00:43.193 --> 00:00:45.361
And so how can embodied AI help?

00:00:45.428 --> 00:00:46.513
And why now?

00:00:47.547 --> 00:00:52.318
Well, embodied AI is at first
when artificial intelligence

00:00:52.318 --> 00:00:57.006
meets the world of robotics and
we talk about three basic

00:00:57.006 --> 00:00:59.042
embodied AI capabilities.

00:00:59.476 --> 00:01:03.012
First and foremost, it starts
with a visual perception to

00:01:03.012 --> 00:01:04.481
understand the 3D world.

00:01:04.748 --> 00:01:09.152
We did great progress over the
past month, not only to identify

00:01:09.152 --> 00:01:13.640
any arbitrary object in 3D
space, we can also predict its

00:01:13.640 --> 00:01:18.628
exact location as well as
orientation in space, which is

00:01:18.628 --> 00:01:21.564
of importance because if you ask
a robotic system to grab

00:01:21.564 --> 00:01:25.268
something, the system needs to
understand where to grab what.

00:01:25.835 --> 00:01:28.471
Secondly, it's about
workflow planning.

00:01:29.239 --> 00:01:32.492
This brings us back to the
example with a bottle because

00:01:32.492 --> 00:01:34.594
you just ask for
something to drink.

00:01:35.161 --> 00:01:38.815
But to accomplish this task, you
need to plan accordingly

00:01:38.815 --> 00:01:41.518
incremental steps to accomplish
this goal.

00:01:42.152 --> 00:01:45.989
And also here we did great
progress in so-called concept learning.

00:01:46.639 --> 00:01:50.410
And last but not least, it's
about the manipulation of

00:01:50.410 --> 00:01:54.581
objects. And for doing this we
need to invest into so-called

00:01:54.581 --> 00:01:59.035
causal models of the real world
because the robotic system needs

00:01:59.035 --> 00:02:03.356
to understand the implication
and impact of certain actions on

00:02:03.356 --> 00:02:04.991
the physical embodiment.

00:02:05.558 --> 00:02:10.446
So overall this is what embodied
AI paves the the way towards

00:02:10.446 --> 00:02:11.698
physical intelligence.

00:02:12.549 --> 00:02:16.269
And how can companies capture
value with embodied AI?

00:02:17.137 --> 00:02:20.490
Well, embodied AI has a
potential to revolutionize the

00:02:20.490 --> 00:02:24.244
world of robotics because you
can't start automating handling

00:02:24.244 --> 00:02:28.148
operations too versatile and
complex for traditional automation.

00:02:28.148 --> 00:02:32.669
Now think of the millions of
handling operations which we do

00:02:32.669 --> 00:02:37.907
everyday on the shop floor, in
warehouses or even beyond the factory.

00:02:38.308 --> 00:02:42.362
And this is the great potential
of embodied AI, which can really

00:02:42.362 --> 00:02:46.065
change manufacturing, but also
the world we are living in.

00:02:46.783 --> 00:02:47.734
Thank you so much.

00:02:47.817 --> 00:02:48.768
Thank you for having me.