NVIDIAs New GR00T Dreams + N1.5 AI Humanoid Robot Tech Is Insane ($9000 K-BOT HUMANOID)

So now, developers can simply post-train using Cosmos’ Predict World Foundation models for their robots using just a few human demonstrations recorded via teleoperation for a single task in one setting. And these videos are then filtered by Cosmos’ reason to produce action tokens which train the robots. Then a single image with new instructions can be fed into Cosmos, and it spits out the model’s dreams, which are brand new synthetic videos of the robot’s future actions. Plus, because Cosmos is powered by Generative Artificial Intelligence, it handles all the new commands without needing extra demo data, making it super flexible and fast.

But the next step is where the magic happens as Cosmos churns out heaps of new dream videos and picks the very best ones all by itself, and then it transforms them into 3D action trajectories for the real world. And the result is that these trajectories function like a cheat sheet for the robots to learn all the actions required to perform any given task. And this is a breakthrough for physical AI and robotics because a small handful of humans can now generate data that would have previously required thousands of people to produce, meaning the playing field has been leveled for both small scale and large scale developers.

And here’s an example because Nvidia released an even bigger bombshell, as they also announced Groot N1.5 as their latest update to their Isaac Groot humanoid robot model. And it was reportedly developed in just 36 hours using Groot Dream’s synthetic data. And this is compared to three months for manual data collection without using the blueprint. And as a result of this Groot N1.5 now adapts much better to new spaces, it recognizes objects via user instructions, and it nails tasks like sorting and storing. And several early adopters like iRobot are already using it to power complex industrial pick and place jobs, and NeuroRobotics is using it to explore new home automation techniques.

And scheduled for later in 2025 Groot N1.5 will also be deployable on Nvidia’s Jetson 4. Plus, Fourier Robotics is also using Nvidia to significantly reduce its physical AI training times as the company just released a new video showcasing the future of its robots operating in the home, hospital, office and more. Starting with its GR2 humanoid, which stands at 175 centimeters and weighs 63 kilograms, and it features an impressive 53 degrees of freedom in total, which allows for smoother, more natural human like motions. And for real world tasks, the GR2 has a payload of up to three kilograms in each arm, and this is to handle tasks like object sorting more effectively, and this broadens its practical applications.

But the GR2 standout feature is its highly dexterous hand which is equipped with 12 degrees of freedom and six array type tactile sensors, which provides the robot with an exceptional level of sensitivity to execute highly delicate assemblies or even medical procedures in healthcare settings. And this combined with teleoperation could be a game changer for industries requiring fine motor skills. And to power it all, the GR2 features the second generation of Fourier Smart Actuators, which deliver peak torques of over 380 Newton meters for more precise, more powerful movements. Plus, the robot’s dual encoder system further enhances control accuracy even in dynamic environments.

And complementing the GR2’s hardware is the Fourier Toolkit, which is a suite of optimized software tools that work with leading frameworks including NVIDIA, Mujoko, and Robot Operating System. And pushing the envelope of physical AI even further, Tesla also just released a video of its Optimus Gen 2 robot performing a ballet dance as well as an insane techno dance that rivals the style of most humans. And Elon Musk said earlier this year that Tesla’s Optimus Gen 2 production goals work 5,000 units this year, with plans to scale to 50,000 units produced in 2026.

And what stands out about the Optimus Gen 2 is that it features 22 degrees of freedom in its hands, and it has a target price range of between 20 to $30,000, with Tesla’s production costs being around $10,000 per robot. But tell us in the comments whether a 3D projector built into the robot’s face would be the cherry on top for the ultimate user experience. But for a more affordable robot, there’s a new five fingered open source humanoid that only costs $9,000 named the Kbot, and it’s paired with the KOS for its operating system, and KSim for its high performance machine learning framework, and together they form a fully integrated open source robotics stack designed to streamline development.

Plus the K operating system is built in the Rust programming language for greater speed, while offering a Python SDK to make it more accessible for developers of all skill levels. And this is huge because it eliminates the need for complex system configurations while KSim delivers GPU accelerated reinforcement learning, and it processes over 100,000 samples per second. And on top of this, it’s sim to real transfer lets developers train humanoid robot policies and simulation, and then deploy them directly to physical robots without any code changes, which dramatically accelerates both prototyping and development.

And by open sourcing their entire stack, including their hardware, software and machine learning infrastructure, K scale labs aims to democratize robotics development with the K bot. Meanwhile, AGI bot just released a new video of its Ling Shih x two robot, a bipedal humanoid that’s powered by the genie operator one model or go one for short. And this is a generalist AI model that enables the x two to seamlessly switch between tasks like doing house chores or riding a bicycle and go one outshines RDT benchmarks by 0.12 points, featuring 28 degrees of freedom throughout its body, five fingered hands and bionic ankles to move with human fluidity.

Plus the link CX two is trained via deep learning and 1 million plus trajectories from the AGI bot world data set with go one’s vision language action framework linking the robots 3d vision to precise actions in order to master dynamic skills and environments. And when it comes to human to robot interaction, the x two’s multimodal AI helps it react in milliseconds when talking to a human. And it even features 3d holographic projection for the naked eye. And as per intelligence, the robots go one zero shot generalization handles untrained tasks with 30% higher success rates when being compared to OXC data sets.

Plus the robot includes a cerebellum controller for movement, a domain controller for decision making and an intelligent power management system as well as modular joints and advanced cooling. Plus AGI bot also just introduced a new robot called the NASA codename x two n and it features feet that turn into wheels for even faster travel. But there’s still no pricing or release information. So tell us how much you think it’ll cost. And finally, open AI just launched codex, its newest cloud based AI agent for automating software development tasks like fixing bugs and building out new features.

And because it’s built on the codex one model, it handles code repos via text prompts and even runs isolated containers. Plus codex documents changes supports agents dot MD files for project rules. And it’s already being used by Cisco and superhuman. Plus it’s available right now to chat GPT pro users for free with pricing plan for other users in the near future. But if you want to know more about the latest in AI news, make sure to like and subscribe and click this video here. [tr:trw].