00:00:00
so today we got probably one of the
00:00:02
craziest announcements this year that
00:00:04
could truly change the future of
00:00:06
Robotics this is The Genesis Project
00:00:10
this is a revolutionary new tool that
00:00:12
combines the creative power of
00:00:14
generative AI with the accuracy of real
00:00:17
world physics imagine being able to
00:00:19
create virtual environments and train in
00:00:21
a way that's not just realistic but a
00:00:23
way that is incredibly fast and
00:00:26
efficient that's exactly what Genesis
00:00:28
does now this project has been in
00:00:31
development for 2 years with over 20
00:00:34
research labs collaborating to make it a
00:00:36
reality the result is a powerful physics
00:00:39
engine that can create Dynamic
00:00:41
four-dimensional worlds and these worlds
00:00:43
are fully interactive and Incredibly
00:00:45
detailed making them perfect for
00:00:47
developing Advanced Robotics and
00:00:49
applications in physical AI which is a
00:00:51
field of teaching machines to understand
00:00:54
and interact with the physical world now
00:00:56
Genesis stands out because it's written
00:00:58
in Python a programming language that
00:01:00
many people already know and even so
00:01:02
it's shockingly fast 10 to 80 times
00:01:05
faster than the other high-end tools
00:01:07
that rely on powerful gpus like Isaac
00:01:10
Sim JY and mjx for example it can train
00:01:13
a robot to walk in just 26 seconds using
00:01:17
a single Nvidia RTX 4090 which is an
00:01:20
incredible achievement now the speed
00:01:22
isn't just for show it's about making
00:01:24
Cutting Edge technology available to
00:01:26
more researchers and developers now
00:01:28
what's crazy about this is that this is
00:01:30
actually open source this means anyone
00:01:33
can access this code and learn from it
00:01:35
and use it to create their own projects
00:01:37
I think this is probably going to be one
00:01:38
of the largest developments for robotics
00:01:40
because one of the main problems that we
00:01:42
do have in terms of you know trying to
00:01:45
advance robotics is the data collection
00:01:47
issue collecting data for robotics is
00:01:49
tedious because as you all know many
00:01:52
people currently use tele operation and
00:01:55
this makes sense because you're
00:01:56
interacting with the robot in a real
00:01:58
physical world and of course you can
00:02:00
collect that data because you know
00:02:02
exactly where the robot is how it
00:02:04
interacts with the world and how the
00:02:06
tasks are achieved successfully or
00:02:09
unsuccessfully but if we have something
00:02:12
like this which is a true physics
00:02:14
simulation something that is very
00:02:16
accurate in terms of the onetoone
00:02:18
mapping of the physical world then
00:02:20
things are going to speed up at a
00:02:22
ridiculous pace and we've seen before
00:02:24
with prior projects when we've used
00:02:27
previous simulations like nvidia's Isaac
00:02:29
Sim or videos Isaac Jim what that has
00:02:31
done for the pace of Robotics so if this
00:02:34
is something that is widely used and is
00:02:37
remarkably more efficient than those
00:02:39
other tools from Nvidia then this is
00:02:40
going to be something that completely
00:02:42
changes the game and we all know
00:02:45
companies and countries are really
00:02:47
really trying to work hard when it comes
00:02:49
to robotics so with something like this
00:02:52
being embedded into the ecosystem of AI
00:02:55
and Robotics we're probably going to see
00:02:58
things explode overall I'm I mean one of
00:03:00
the things that we have to understand is
00:03:02
that the real problem is not the
00:03:04
humanoid robots platform humanoid robots
00:03:06
are really successful at doing a variety
00:03:08
of different tasks the main problem is
00:03:10
of course the autonomy and ensuring that
00:03:13
we can get enough data to scale our
00:03:15
efforts across the world so this is
00:03:17
going to be something that I think is
00:03:18
going to be really fascinating
00:03:20
especially when we start seeing first
00:03:22
use cases of course right now this is
00:03:24
something that they have just released
00:03:26
so it will be a little bit of time
00:03:27
before we start to see the real world
00:03:29
implications of this I can imagine there
00:03:31
are some developers probably already
00:03:33
hacking away at their Labs figuring out
00:03:35
the best ways and the best settings to
00:03:38
get to grips with this kind of software
00:03:40
so for me this is largely one of the
00:03:42
craziest AI updates SL robotics updates
00:03:45
because this has a wide range of
00:03:47
applications that most of us simply
00:03:49
cannot fathom and before we get into all
00:03:51
of the futuristic sci-fi outcomes of
00:03:53
this let's take a look at how good this
00:03:55
simulator is and some of the key
00:03:57
features that differentiates it from any
00:03:59
software that's existed before now The
00:04:00
Genesis Project introduces a
00:04:02
groundbreaking feature a generative
00:04:04
agent that can handle every step of
00:04:06
teaching robots how to operate in
00:04:07
realistic environment entirely on its
00:04:09
own first it autonomously designs
00:04:11
virtual environments that mimics real
00:04:13
world spaces like kitchens living rooms
00:04:15
and other everyday settings then it
00:04:18
proposes the tasks for the robots to
00:04:19
perform such as opening a microwave
00:04:22
picking up objects or navigating through
00:04:24
Furniture but it doesn't stop there it
00:04:26
also creates the reward systems which
00:04:28
are like the incentives for robots
00:04:30
teaching them what succcess looks like
00:04:32
for instance if a robot successfully
00:04:34
opens a microwave the system gives it a
00:04:36
reward helping it learn faster and
00:04:38
better and finally it automates the
00:04:41
entire process of writing robotic
00:04:43
policies which are the detailed
00:04:45
instructions or strategies that guide
00:04:48
how robots move think and interact with
00:04:50
their environment that means that robots
00:04:53
don't just get trained they become
00:04:55
smarter and more capable of handling
00:04:57
complex tasks all while developers and
00:05:00
researchers save time and effort and by
00:05:02
combining these capabilities Genesis
00:05:04
sets a new standard for how robots learn
00:05:07
and adapt to the physical world now
00:05:09
what's crazy about this physics engine
00:05:11
that can simulate an incredibly diverse
00:05:13
range of physical phenomena this engine
00:05:15
actually combines state-of-the-arts
00:05:17
physics solvers like mpm material Point
00:05:19
method SP smooth particle hydrodynamics
00:05:23
F finite element method rigid body
00:05:25
Dynamics and pbd among many others and
00:05:28
this makes it cap of realistically
00:05:30
simulating virtually anything you can
00:05:33
imagine from rigid objects to objects
00:05:36
like metal to soft deformes like cloth
00:05:38
and rubber to complex materials like
00:05:40
liquid and smoke and even elastic or
00:05:43
plastic bodies and it doesn't stop there
00:05:44
Genesis even supports Advanced
00:05:46
simulations for articulated systems like
00:05:49
robot muscles and thin shell materials
00:05:51
which are crucial for Robotics and
00:05:53
physical Ai and what this means is that
00:05:56
Genesis can simulate real world physics
00:05:58
at an unprecedented level of detail and
00:06:00
accuracy imagine being able to replicate
00:06:02
how a piece of fabric Moves In The Wind
00:06:04
how liquid Splash and flow or how robot
00:06:07
muscle stretch and contract all in a
00:06:09
virtual environment these simulations
00:06:11
are not just for show they play a
00:06:13
critical role in training AI models and
00:06:15
robots to interact with the physical
00:06:16
world in realistic Ways by bringing all
00:06:18
of these physical capabilities into one
00:06:21
unified engine Genesis makes it easier
00:06:23
for developers to research and create
00:06:25
highly realistic simulations Beyond its
00:06:27
incredible ability to simulate rigid
00:06:29
object objects liquids and deformable
00:06:31
materials Genesis also takes robotics to
00:06:33
the next level by supporting Soft
00:06:35
Robotics this is a GameChanger because
00:06:37
soft robots like those made to mimic
00:06:39
muscles or organic movements are
00:06:41
notoriously difficult to simulate
00:06:43
Genesis bridges that Gap it's the first
00:06:44
ever platform to provide full support
00:06:46
for soft muscle simulations and soft
00:06:48
robot interactions with rigid structures
00:06:50
what's even better is that Genesis
00:06:52
includes a configuration system similar
00:06:54
to urdf which is commonly used in
00:06:57
robotics this makes it easier than ever
00:06:59
to design test and refine soft robotic
00:07:02
systems in Virtual environments for
00:07:04
example you can simulate how a soft
00:07:05
robot might crawl or interact with its
00:07:07
surroundings which has huge potential
00:07:09
for Industries like healthcare search
00:07:11
and rescue and more and if Soft Robotics
00:07:14
wasn't impressive enough Genesis
00:07:15
continues to push the boundaries with
00:07:17
its unparalleled computational
00:07:18
capabilities one standout feature is its
00:07:21
GPU parallelized ik solver ik or inverse
00:07:25
kinematics is a critical process in
00:07:27
robotics that calculates The Joint
00:07:29
movements needed for a robotic arm to
00:07:31
reach a specific Target normally this is
00:07:33
computationally expensive and timec
00:07:35
consuming especially when dealing with
00:07:36
multiple robotic systems but Genesis
00:07:39
takes this to a whole new level it can
00:07:40
solve ik for an incredible 10,000
00:07:43
robotic arms like the highly Advanced
00:07:45
Frank arms simultaneously in under 2
00:07:47
milliseconds using a single RTX 4090
00:07:50
this level of speed and efficiency is
00:07:52
unprecedented and opens up new
00:07:54
possibilities for large-scale robotic
00:07:56
simulations Industrial Automation and
00:07:59
complex research projects with Genesis
00:08:01
the limits of what we thought was
00:08:02
possible in robotics are being redefined
00:08:04
another groundbreaking capability of
00:08:06
Genesis is its support for Native
00:08:09
non-convex Collision handling this might
00:08:11
sound technical but it's a huge deal in
00:08:12
the world of simulations in simpler
00:08:14
terms most physics engines struggle to
00:08:16
handle collisions between objects with
00:08:18
complex irregular shapes like gears
00:08:21
chains or any object that isn't smooth
00:08:23
or simple these shapes are called
00:08:24
non-convex and accurately simulating
00:08:26
their collisions requires Advanced
00:08:28
algorithms is tackles this challenge
00:08:30
head-on with built-in support for these
00:08:32
types of interactions this means it can
00:08:34
accurately simulate how irregularly
00:08:36
shaped objects bump slide or even
00:08:38
interlock with one another all while
00:08:40
maintaining physical realism for example
00:08:42
if you're working on a robotic system
00:08:44
that interacts with intricate Machinery
00:08:45
or components Genesis ensures that every
00:08:48
collision and movement behaves as it
00:08:50
would in the real world this capability
00:08:52
makes Genesis a powerful tool for
00:08:53
Industries like manufacturing Robotics
00:08:55
and game development where precise
00:08:57
Collision handling is crucial
