Introduction
Today I’m going to introduce you to several new humanoid robots with some costing as little as $5,500 US dollars. Let me first show you the five new tech breakthroughs Nvidia’s Groot AI demonstrated with multiple robots at GTC 2025.
How fast they’ve made this progress, starting with number 5, Nvidia’s Isaac Groot N1.
This is a foundation model for humanoid robots building on 2024’s Project Groot.

see back in GTC 2024, Project Groot focused on robots learning from human demos and language for basic skills.
But now, in 2025, Groot N1 introduces a dual system design featuring slow reasoning for planning but fast execution for its precision.
That raises to tech advance number 4, which is that Groot N1 is now open sourced.
See in 2024, Project Groot was Nvidia led with partners like Boston Dynamics showing their closed setups.
But now, in 2025, anyone can use it because at GTC, they open up access to all startups, researchers, and companies to adapt Groot N1 for their own factories, healthcare, and home environments.
The addition of MEGA to Nvidia’s
tool within Omniverse. See, while GTC 2024 owned individual robot training with Isaac Lab, MEGA scales it up, testing fleets of robots together, like in a virtual Blackwell factory.
Where humanoids coordinate and adapt in real time
Or, Shapler and Accenture used MEGA to test agility robotics digit fleets for material handling.
And Pegatron used it for their Metropolis video analytics to enhance factory safety.
And what’s even more important is that this addition of MEGA shows Nvidia is aiming for larger scale deployments of humanoid robots.
unlike 2024’s slower data methods, it now uses Omniverse and Cosmos to generate 780,000synthetic motion trajectories in just 11 hours, which boosts Groot N1’s performance by 40 percent.
And Hianna is already simulating Atlas robots on their assembly lines using Blueprint.
Plus Mercedes Benz is optimizing its Apollo robots in their vehicle assembly lines.
Nvidia’s Open Source Robotics and China’s Advanced Humanoid Robot Dubat

And Nvidia has released these training data sets for Groot N1 on platforms like Hugging Face, completely open source for everyone to use for worldwide development.
And all of this tech is combined thanks to tech advance number 1 and Nvidia’s Newton physics engine, which is a new open-source physics engine built in collaboration with Google’s D-Mind and Disney’s research.
Unlike 2024’s general simulation tools, Newton’s designed for robotics handling fine-brained dynamics and tactile feedback in super real-time.
This brings us to these humanoid robot updates coming out of China right now, starting with the Dubat humanoid robot for dexterous manipulation.
This thing features 28 degrees of freedom with 0.05 millimeters of precision, meaning that the robot can pick up something as small as a cherry without any problem at all.
Then there’s its straight knee walking setup, which saves up to 42 percent of its energy.
And it walks with a relatively natural gate across all kinds of surfaces, whether it’s a cluttered factory or a rough patch in a forest.
And as for this robot’s hardware, it’s running on edge computing with its robot operator model 1,
which gives it 7.7 times the usual brain power to handle real life chaos like a champ.
And as for the Dubat’s release, it’s still undetermined when it will begin selling and for what price.
But there’s another humanoid robot that’s going even further with flips and much more.
Meet Noatix N2 and Kepler K2: The Latest Advancements in Humanoid Robotics

The Noatix N2 boasts 40 TOPS AI, 15W power use, 8GB video memory, powered by NVIDIA Jetson.
And when it comes to hardware, it features two depth cameras for accurate 3D visual perception plus 18 degrees of freedom throughout
Its body using pure electric drive and 120 Newton meters of power for a maximum speed of 3.2 meters per second,
This robot walks naturally using deep learning, with autonomous obstacle avoidance, road adaptation, and dynamic anti-interference.
And as for the robot’s size, it stands 110 centimeters tall and weighs 20 kilograms,
giving it a light form factor to fit into small tight spaces that even humans may not be able to address.
But another brand-new humanoid robot demo was also released from Kepler.
Kepler K2
This is the K2 humanoid robot standing at 178 centimeters and weighing 85 kilograms.
It has 52 degrees of freedom total: 12 in ultra-dexterous hands and 40 in the limbs.
And it can handle uneven terrains, dodge obstacles and even lift heavy objects as well as chat with humans.
Precise hand-eye coordination and realistic human-like movement are enabled by planetary roller screws and rotary actuators.
Kepler K2’s Cognitive Abilities: Visual Recognition, Chat, and a Developer Ecosystem
But the Kepler K2’s nebulous system is actually what’s responsible for its four main cognitive jobs,
first of which is spotting things in the real world using its visual recognition.
Second is mapping its surroundings with visual slam or simultaneous localization and mapping.
Third is chatting through its multimodal interaction and finally fourth is nailing its hand-eye coordination.
And Kepler is going to the next level by pairing its cloud-based multimodal large language model with a smaller fast response industry specific model right within the robot.
Plus, they’ve rolled out their Kepler open-source system which is a global developer platform that syncs up with multiple robots.
it offers online tools for coding and debugging. which allows other developers to use this nebula cognitive system.
Kepler’s Nebula for All, and Engine AI’s PM01 Robot Shakes Things Up
There’s engine AI which just released a new video of its PM01 robot doing some insane dance moves..
Some people don’t even know whether or not this is real. Make sure to comment down below what you think.
This robot is138 centimeters tall, weighs 40 kilograms and has 23 degrees of freedom with five in each arm.
Its neck allows impressive moves, featuring a 320° waist twist and 330 Nm joint torque.
This robot walks up to 2 m/s and has 6 leg degrees of freedom: 3 in hips, 1 knee, 2 ankles.
As for onboard intelligence and hardware it features a dual chip setup with the Intel N97.
Features NVIDIA Jetson Orin, a 2-hour 10,000mAh battery, and an alleged $13,000 price tag.