2 weeks ago
10
AI has already transformed our digital world. But now, it’s extensively stepping into the physical one. The next great wave of AI isn’t just about software; it is about physical AI—intelligent machines and systems that think, reason, and interact with the real world.
If NVIDIA’s GTC 2025 is any indication, this shift is moving at a pace faster than anyone anticipated. Notably, NVIDIA’s leap into physical AI places it at the forefront of a rapidly expanding ecosystem.
During his keynote speech, CEO Jensen Huang declared that robotics powered by physical AI could become “the largest industry of all”. From humanoid robots to autonomous vehicles, NVIDIA promises to reshape industries and individual societies.
Yet, NVIDIA isn’t the only player in this high-stakes game. Across the US and beyond, tech giants and ambitious startups are racing to build intelligent machines that will define our future. Moreover, while Silicon Valley leads much of the narrative, China is making steady moves of its own.
NVIDIA’s Bold Bet on Intelligent Machines
“Physical AI and robotics are moving so fast. Everybody, pay attention to this space,” Huang said.
NVIDIA unveiled Isaac GR00T N1, its first generalist foundation model for humanoid robots. GR00T N1 enables robots to reason, plan, and act in complex environments, combining ‘slow thinking’ with reasoning and perception and ‘fast thinking’ with precise and real-time actions.
Notably, GR00T N1 was also announced as open source, inviting a wave of innovation across the robotics ecosystem. NVIDIA also introduced Newton, an open-source physics engine for robotics simulation developed in collaboration with Google DeepMind and Disney Research.
Huang then brought on stage an innovation that took over the internet for the next few days. Blue, an advanced AI-powered robot, followed real-time tactile feedback and soft-body simulations. This was done at super-real-time speeds, enabling robots to train in highly realistic digital environments.
Alongside Omniverse, Cosmos, and Isaac Lab, NVIDIA is building what Huang refers to as the “operating system for physical AI”.
In autonomous vehicles, NVIDIA unveiled a major partnership with General Motors, aiming to power the automaker’s next-generation self-driving car fleet, along with the announcement of NVIDIA Halos, a comprehensive safety system for AVs.
But NVIDIA isn’t running this race alone.
Who Else is Building the Future?
As NVIDIA dominates headlines, other US-based companies and startups are busy shaping the future of physical AI and robotics, introducing updates much faster than expected.
Just three months into the year, robotics has made such significant progress that this could genuinely be the year of robotics. From industrial robots to humanoids for homes, there has been an immense expansion in their use cases.
Moreover, American engineering and robotics company Boston Dynamics continues to evolve and update its humanoid robot, Atlas.
Recently, in partnership with RAI Institute, it demonstrated new levels of agility and dexterity using reinforcement learning (RL) policies developed by a motion capture suit.
Atlas is demonstrating reinforcement learning policies developed using a motion capture suit. This demonstration was developed in partnership with Boston Dynamics and @rai_inst. pic.twitter.com/bEk6wEDbIn
— Boston Dynamics (@BostonDynamics) March 19, 2025In an X post, the company announced it is “designing Atlas to do anything and everything”. Through part sequencing, a common logistics task in automotive manufacturing, the Atlas robot can carry out monotonous tasks like pick, carry, place and repeat.
Boston Dynamics also demonstrated another example of RL in its Spot robot in a separate partnership with RAI Institute.
Using reinforcement learning, we trained policies for @BostonDynamics Spot that allow the robot to achieve record running speeds of 11.5 mph (5.2 m/s) — over three times faster than Spot's default max speed. pic.twitter.com/CuPWtj2Pqk
— RAI Institute (@rai_inst) March 4, 2025Meanwhile, Austin-based Apptronik has provided updated glances at its general-purpose humanoid robot, Apollo. With collaboration from NASA, backing from Google DeepMind and latest funding from Mercedes Benz, the company focuses on industrial and space applications.
Apptronik’s latest partnership with Jabil to use Apollo robots on the production lines that create them marks progress towards robots constructing robots.
The company announced on X that it is “ultimately paving the way for Apollo to build Apollo.”
In January this year, the company had announced a collaboration with Texas Instruments as part of its commitment to building the safest humanoid robots in the world.
The company partnered with Google DeepMind in March to build the next generation of humanoid robots with DeepMind’s new AI model, Gemini Robotics, which enables Apollo to understand natural language and perform real-world tasks in environments it has not seen before.
Moreover, Silicon Valley startup, Figure AI, is another company that has made significant progress this past year and has contributed to America’s robotic ecosystem.
In a latest update, using RL, the company’s humanoid robot, Figure 02, learned to balance, shift weight, and walk naturally through trial-and-error simulations. The approach adapted the robot for industrial and domestic applications by mimicking natural human locomotion.
Introducing learned natural walking
Figure can now walk naturally like a human
Years of data was trained in simulation in just a few hours pic.twitter.com/HvOq9s1krS
In February this year, the company also announced Helix, a vision-language-action (VLM) model that allowed humanoid robots to perform complex tasks using natural language. Watch the robot pick up virtually any object just by natural prompting like, ‘Pick up the desert item’.
We find that just by prompting the robot, we can pick up virtually any object
When asked to "Pick up the desert item," Helix identifies the toy cactus, chooses the nearest hand, and executes precise motor commands to grasp it securely pic.twitter.com/2l70kKG1GV
In March, the company also unveiled BotQ, its high-volume robot manufacturing facility, capable of producing 12,000 humanoids annually. It plans to grow the fleet to 1 lakh robots within four years. BotQ uses Swedish-Swiss automation company ABB’s robotic arms and autonomous equipment for manufacturing.
Today we're introducing: BotQ
BotQ is our high-volume robot manufacturing facility
Initially designed to produce 12,000 robots per year, it will scale up to support a fleet of 100,000 pic.twitter.com/4HmXEa9C7R
Figure AI founder, Brett Adcock, also announced that the company moved up its home timeline by two years, starting Alpha testing this year and launching humanoids for home.
While Tesla’s Optimus was making big strides last year, the company has not had any updates this year.
Norwegian 1X Technologies, another rising humanoid-for-home company, launched major updates this year on top of its existing humanoid NEO Beta and industrial robot EVE Industrial.
It launched NEO Gamma, its latest personal assistant and home companion, which has abilities like tidying, deep cleaning, and home management.
“There is a not-so-distant future where we all have our robot helper at home, like Rosey the Robot or Baymax. But for humanoid robots to truly integrate into everyday life, they must be developed alongside humans, not in isolation,” Bernt Børnich, founder and CEO of 1X, said.
Introducing NEO Gamma.
Another step closer to home. pic.twitter.com/Fiu2ohbIiP
In other interesting updates, NEO Gamma unboxed the latest model of the Nothing phone, showed housework competency by vacuuming at the GTC summit 2025, went around hugging people, and even gifted Huang a jacket!
NEO got Jensen a jacket :) pic.twitter.com/uGz65iLA1X
— 1X (@1x_tech) March 21, 2025Agility Robotics has gained attention with Digit, its bipedal humanoid robot designed for warehouse automation. In late 2024, Agility introduced RoboFab, claimed to be the world’s first factory dedicated to humanoid robots, with plans to produce over 10,000 robots annually.
At the summit, Digit showed off a new RL-based whole-body controller, fetching snacks without seeing them before or being programmed how to manipulate them. The company also expanded its partnership with NVIDIA, adopting Isaac Sim and Isaac Lab robot simulation and learning frameworks to train and test behaviors on Digit.
Last week, this small Digit fleet worked continuously for 28 hours spread across the three-and-a-half days of #promat2025. They performed three different tasks at a 98.96% success rate. pic.twitter.com/rOGNef6aAg
— Agility Robotics (@agilityrobotics) March 25, 2025China is Not Resting
China has quietly—and not-so-quietly—advanced its physical AI capabilities, with companies making significant strides in robotics, manufacturing, and AI integration. Two significant examples, with a continuous line of updates in 2025, are UBTECH Robotics and Unitree Robotics.
Shenzhen-based UBTECH has its hands in various domains, from humanoid robots to AI education, commercial bots, healthcare bots, and consumer bots.
The company recently completed what is claimed to be the world’s first multi-humanoid robot collaborative training program at Zeekr’s 5G Intelligent Factory.
The initiative involved UBTECH’s Walker S1 humanoid robots collaborating across various production zones, including assembly workshops and quality inspection areas.
Best known for its affordable quadruped robots, Unitree, a Zhejiang-based company, has expanded extensively into humanoid platforms.
The company’s G1 robot boasts mobility and dexterity, which is driven by imitation and RL. It has been seen demonstrating highly dynamic tasks like running and jumping. But this is not it. G1 has also achieved records in the first robotic kick-up, the first successful side flip, a 720° spin kick in Kung Fu, and even dancing!
What Dance Would You Like to Perform with Unitree G1?
With the upgraded algorithm, G1 can learn any dance. Leave a comment to tell us what dance you'd like to see！😘#Unitree #AGI #EmbodiedAI #SpringFestivalGalaRobot #AI #Humanoid #Bipedal #WorldModel #Dance pic.twitter.com/pskis1nycg
