The year 2024 is shaping up to be the year of robots, particularly humanoid robots, and several major companies are leading the charge with groundbreaking advancements in artificial intelligence. OpenAI, 1x, NVIDIA, Boston Dynamics, and Anduril are just a few of the companies making waves in the robotics industry.
1X has introduced a fully autonomous robot built on a single neural network, capable of learning general purpose skills without explicit programming. NVIDIA’s Project Groot aims to create a general purpose Foundation model for humanoid robot learning, enabling robots to understand and perform a variety of tasks through multimodal instructions.
Boston Dynamics continues to impress with their reinforcement learning software, enhancing their robots’ capabilities to navigate complex terrains and perform tasks with precision. And companies like Mercedes, Xiaomi, and Yondu are also making strides in the robotics field with innovative humanoid and dog robots.
With each new development, the future of robotics looks brighter and more exciting. The possibilities for robots to collaborate with humans and revolutionize various industries are endless. As humanoid robots become more advanced and accessible, the potential for innovation and progress is limitless. Stay tuned for more groundbreaking advancements in the world of robotics in the coming years.
Watch the video by Matthew Berman
Video Transcript
2024 is shaping up to be the year of robots and specifically humanoid robots and now today the 1X company who also makes a humanoid robot dropped an update video also and so let’s take a look at that He He he hey hey hey hey Hey Okay what you saw there is a fully autonomous robot built on a single neural network end to endend that means it wasn’t specifically programmed to do any of the things that you saw in the video and if we look at the post right here it says onex’s mission is to create
An abundant supply of physical labor through Androids that work alongside humans we’re excited to share our latest progress on teaching evees general purpose skills the following is all autonomous all 1X speed all controlled with a single set of neural network weights okay now on to Nvidia so let’s
Talk about what they’re building with Groot so I made a video talking about Dr Jim fan building a super team for AGI within Nvidia and this is another project that is falling under his team’s umbrella now I already went over a bunch of the other projects that he has been
Working on so I won’t go into depth about those but a quick summary is he’s built autonomous agents within video games like Minecraft that can essentially explore and learn how to play as they go with no explicit instruction all the way through fully virtualized training environments in which robots can test their abilities
And iterate until they find something that works so they have training environments they have World models I mean Nvidia is just crushing it on all points but now let’s look at project Groot today is the beginning of our moonshot to solve embodied AGI in the physical world embodied AGI just means
It has a body so a robot I’m so excited to announce project Groot our new initiative to create a general purpose Foundation model for humanoid robot learning and this is something he’s been working on for a while and now it’s called project Groot and it basically
Means giving a single AGI to a robot with no explicit training on any individual task and it just knows how to do it or knows how to learn it so the group model will enable a robot to understand multimodal instructions such as language video and demonstration and
Perform a variety of useful tasks we are collaborating with many leading humanoid companies around the world so that Groot May transfer across embodiments and help the ecosystem Thrive so Groot is born in nvidia’s deep technology stack they use Isaac lab which we already talked about and I’ll give you another quick update
About that right now we are able to scale up Eureka thanks to Isaac Jim a GPU accelerated physics simulator that speeds up Reality by 1,000 times now I have mentioned something called synthetic data many times in the past synthetic data just means you’re using an AI to create data for another AI
That’s What Isaac Jim does it takes one example and actually is able to run a lot of different versions of the same task over and over again at 1,000 times speed so it is able to generate a ton of data to get good at you know flicking the little pencil around really really
Quickly it’s trained on osmo which is their new compute orchestration system to scale up models and deploying it on Jets and Thor which is their new Edge GPU chip designed to power Groot so let’s talk about osmo because that’s super interesting and they actually made an entire video about it which I touched
On already in yesterday’s Super cut so let me play that for you now it’s not enough for humans to imagine we have to invent and explore and push Beyond what’s been done am of detail we create smarter and faster we push it to fail so so it can
Learn we teach it then help it teach itself we broaden its understanding to take on new challenges with absolute precision and succeed we make it perceive and move and even reason so it can share our world with [Laughter] us this this is where inspiration leads us the next Frontier this is Nvidia Project group a general purpose Foundation model for humanoid robot learning the group model takes multimodal instructions and past interactions as input and produces the next action for the robot to execute we developed Isaac la a robot
Learning application to train group on Omniverse Isaac Sim and we scale out with osmo a new compute orchestration service that coordinates workflows across dgx systems for training and ovx systems for simulation with these tools we can train Groot and physically based simulation and transfer zero shot to the real
World the gro model will enable a robot to learn from a handful of demonstrations so it can help with everyday tasks and emulate human movement just by observing us this is made possible with invidious technologies that can understand humans from videos train models and simulation and ultimately deploy them directly to
Physical robots connecting group to a large language model even allows it to generate motions by following natural language instructions hi one can you give me a high five sure thing let’s high five can you give us some cool moves sure check this out all this incredible intelligence is
Powered by the new Jetson Thor robotics chips designed for group built for the future with Isaac lab osmo and Groot we’re providing the building blocks for the next generation of AI powered robotics so announced that Jensen’s keynote project Groot is a Cornerstone for the foundation agent road map of the
Newly formed gear lab and the gear lab is that lab that I told you about in a previous video that is Jim fans new super team to develop AGI so he then he talks a little bit more about what Gear’s doing but again very very cool and interestingly enough Brett Adcock
Said congrats and it’s interesting because Brett is the founder of figure robot again who just made an announcement last week with inedible updates themselves and I have a feeling they’re going to be working closely together with Nvidia and Jim says thanks Brett your team is doing Stellar work
Too the more talents working on humanoid the better for the world so if you like humanoid robots this is the time to be alive okay and Boston Dynamics not to be left out of all the commotion about robots they have now released their own video about their reinforcement learning
Software that they’re using for their own robots and if you’re not familiar with Boston Dynamics they are really the first company that I remember to have truly mindblowing robots doing things that we never thought were possible and they’ve been doing this for decades already and so now with this
Announcement they’re showing how some of their reinforcement learning is able to help these robots do tasks that previously they weren’t able to do and just in terms of capabilities Boston Dynamics is still far ahead of the pack so let’s take a look at this video so this was one of their original spot
Prototypes that you’re seeing here and here’s their Atlas robot which is running and I mean look how smooth the running is and the atlas robot is the same robot that can do parkour really well I mean the movement of these robots is incredible so what they’re talking
About here is the ability for robots like spot to walk over a different terrain and they have to make a ton of decisions in real time millions of decisions in real time for how to walk over difficult terrain and all of that is powered by their reinforcement learning so here’s another example
Example walking up and down stairs walking up and down platforms and they just have to make sure that maybe they have to slow down a little bit maybe they need to speed up maybe it’s different types of terrain slippery High traction it all depends and this is the
Software to make it work they’re basically using a lot of the same techniques as Nvidia is using with their new Groot software and they’re modeling all of this out in a world model they’re allowing these robots to explore and do things virtually and iteratively before program the actual physical robot and so
Here’s an example of the robot walking without reinforcement learning on a really slippery surface and this is it their new version with reinforcement learning and look how much better it is with all of those micro decisions that it’s having to make in real time I can’t even imagine the compute necessary to do
That so here’s another example we have spot version 3.3 on the left and the new version with better reinforcement learning on the right and it’s ability to walk up on this platform with a slippery surface navigate it properly so successfully and then come down and so
You can see this is a huge leap forward for spots abilities let’s take a look at a couple more examples of spot navigating environments Now And of course we’re starting to see robots in actual Factory environments and this is so exciting so Mercedes according to the verge is now triing humanoid robots for low skill repetitive tasks this is what it looks like and they are powered by apptronics Apollo robot so I’ve not actually heard of
Vtronic I need to do some research on them but this is what the robot looks like like and it is another humanoid robot that looks so cool let me show you Apollo’s demo video Now [Applause] And Company zami and I hope I’m pronouncing that right also has a robot dog named cyberdog and this is their new version and this is a $3,000 robot so affordable relatively speaking and this is a robot you can buy today it’s more like a toy than an actual
Robot that belongs in a working environment but it’s still incredible to see what can be bought as a consumer so for just $3,000 you can have your own pet dog robot and here’s Michael Chen who just left MIT last semester to join his friends and start yandu and I’m not
Super familiar with the company yet but essentially what they’re trying to do is give the ability to humans to speak natural language to any Ro robot and then that robot does what we say and it was actually something Microsoft did last year that I made a video on so this
Isn’t necessarily brand new but I love anybody who’s a builder and is getting into robots right now and I can’t wait to see what they build let’s watch this video now take off fly up 0.5 M turn around and then follow the guy with the white hoodie all right so we’re yandu we’ve
Created an app that allows to control DJ ey drone with your voice check it Out All right that’s it that’s a r thanks for watching so as you can see robots are going to get easier and easier to use to the point where we’re just going to describe what we want from them and they will do it hopefully and from andril industry series which is a defense
Contractor founded by the Oculus founder they just released a new robot the atlas 700m and this is a Warfare robot so a little bit different from what we’ve been talking about but it’s still really really cool what it’s capable of and what andril has done really well is integrated artificial intelligence into
All of their products so let’s watch this video now Pap A I’m so excited for the future I’m so excited to see what other humanoid robots are going to come out in 2024 and in fact there are a lot more than I’ve even started to cover there is somebody on X that I saw who is collecting a master list of all the humanoid robot
Companies and putting them in one place so I’m definitely going to keep a look and I will link that in the description below for you to check out what do you think of all these robots are you excited are you scared or somewhere in between if you liked this video please
Consider giving a like And subscribe and I’ll see you in the next one
Video “STUNNING Breakthrough “AGI Robot” From OpenAI, 1x, NVIDIA, Boston Dynamics, Anduril” was uploaded on 03/20/2024 to Youtube Channel Matthew Berman