Google DeepMind's onboard robot brain

Good morning, robotics enthusiasts. Google DeepMind has unveiled its first vision-language-action model that runs entirely onboard a robot — no cloud, no cables, just self-contained intelligence.

It’s part of a new wave of untethered robotics — machines that can see, think, and act without phoning home. The promise? Smarter, faster, and more autonomous robots, ready to take on the real world.

In today’s robotics rundown:

• Google DeepMind’s on-device robot AI

• Apptronik’s ‘superhuman’ robotics spin-off

• MIT fast-tracks robot design with generative AI

• Eco-friendly robots made from rice paper

• Quick hits on other robotics news

LATEST DEVELOPMENTS

GOOGLE DEEPMIND

🧠 Google DeepMind’s on-device robot AI

The Rundown: Google DeepMind just unveiled Gemini Robotics On-Device, its first vision-language-action (VLA) model built to run entirely on a robot — with no internet needed. And it delivers nearly the same muscle as its cloud-powered counterpart.

The details:

• The new model allows robots to interpret what they see, understand instructions, and carry out complex physical tasks entirely on their own. 

• Google says that despite its compact size, the model’s performance is nearly on par with the cloud-based Gemini Robotics system.

• It can adapt to new tasks with as few as 50 to 100 demos, making an ideal solution for rapid deployment in changing environments.

• Google trained the model on its ALOHA robot, but says it is adaptable to different robot types, including humanoids and bi-arm robots.

Why it matters: As robot models like 1X’s Redwood and Figure AI’s Helix expand into varied physical environments, Google is charting a different course — prioritizing flexibility. Its Gemini-based robotics system, designed to run both in the cloud and locally, will enable teams to build latency-sensitive apps, even in fully offline settings.

APPTRONIK

🤖 Apptronik’s ‘superhuman’ robotics spin-off

The Rundown: Austin-based Apptronik, makers of the Apollo humanoid, launched a new subsidiary called Elevate Robotics — but instead of industrial humanoids, their focus is on “superhuman” machines that aren’t limited by the human form.

The details:

• The new subsidiary will build machines designed for heavy-duty industrial tasks, such as lifting massive loads and operating in extreme conditions.

• This strategic spin-off comes hot on the heels of Apptronik’s $403M funding round, led by B Capital and Capital Factory, with participation from Google.

• Driven by Plus One Robotics co-founder Paul Hvass, Elevate inherits Apptronik’s decade of breakthroughs, from exoskeletons to robotic arms.

• That said, unlike Apollo, Elevate’s robots will not be limited to the two-legged, human-like design, allowing for more specialized and robust architectures.

Why it matters: Elevate promises multipurpose automation robotics designed for extreme loads, relentless precision, and environments too dangerous or demanding for people — or even humanoids — to handle. The move signals a new wave of automation that might leave bipedal industrial humanoids in the dust.

MIT

🔥 MIT fast-tracks robot design with generative AI

The Rundown: MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) is leveraging generative AI to fundamentally change how robots are conceived and built — turning a historically slow, manual process into a fast and automated one.

The details:

• In two new projects, MIT used an AI platform to test thousands of underwater robot designs — including jumpers and gliders — inside a physics simulator. 

• The AI optimized every parameter, from fin shape to weight, and spat out the best designs — a task that would have taken months of trial-and-error.

• Once the AI crowned its champions, CSAIL’s team 3D-printed the winning blueprints and threw them into real-world testing.

• The jump-bot shattered depth records, propelling itself off ocean floors with biomimetic precision, while the glider sliced through water with minimal drag.

Why it matters: Critically, these weren’t digital fantasies: the physical prototypes performed within 5% of their simulated predictions, validating the AI’s ruthless efficiency. This workflow — AI-driven design, simulation torture-testing, and then instant fabrication — could make robotics’ slow-motion R&D loops a thing of the past.

UNIVERSITY OF BRISTOL

🍚 Eco-friendly robots made from rice paper

The Rundown: Scientists at the University of Bristol just found that Vietnamese spring roll wrappers work as well as silicone materials in prototyping flexible robots — plus, they’re nontoxic and biodegrade completely within a month.

The details:

• Unlike traditional plastics or silicones, this kitchen material is fully biodegradable, allowing soft robots to be composted after use.

• Because it’s edible, the material is completely nontoxic, making it ideal for applications in medical, educational, or food-handling environments.

• The team was able to shape it into grippers, actuators, and even sensors that mimic the softness and adaptability of biological tissues.

• Its natural transparency also allows for easy integration with optical sensors, while its malleability lets researchers customize thickness and texture.

Why it matters: Bristol’s rice paper robot turns a humble kitchen staple into a biodegradable, non-toxic wonder, making soft robotics accessible, affordable, and eco-friendly. Like similar robotics projects that use starch or algae, it’s a recipe for rapid prototyping, safer outreach, and single-use applications that won’t leave a trace.

QUICK HITS

📰 Everything else in robotics today

Waymo and Uber launched robotaxi rides in Atlanta, marking Waymo’s fifth major U.S. market and the second city under its deal with Uber after Austin.

Tesla CEO Elon Musk teased new progress on the company’s upcoming humanoid, Optimus V3, and confirmed that it is already using a Grok voice assistant.

Software firm Nominal raised $75M to expand its unified platform that accelerates hardware development and testing for aerospace, defense, and robotics.

Melco Mobility, a Mitsubishi spin-off, ordered nearly 100 industrial robots from U.S.-based Cartken to automate material handling operations throughout Japan.

Ontario introduced Canada’s first beach-cleaning robot, a fully electric rover known as BeBot, to remove litter from provincial park shorelines.

Cobot leader Universal Robots unveiled UR Studio, a next-gen online simulation tool designed to transform how robotic work cells are customized and deployed.

Bike-sharing firm Hellobike moved into the robotaxi sector by forming a new joint venture with tech powerhouse Ant Group and global battery giant CATL.

ROKAE Robotics, a robotics firm based in Beijing, has unveiled two new industrial humanoid models for factory jobs.

Galbot, a Chinese startup developing general-purpose AI robots, raised $152M, bringing total funding to $332M in just two years.

California-based K-Scale Labs announced that it will manufacture its $9K open-source humanoid robots in Dallas.

German startup NEURA Robotics unveiled its 4NE1 Gen3 at Automatica 2025 in Munich as it is reportedly seeking to raise up to $1.2B in new funding.

Northern Arizona University researchers unveiled OpenExo, the world’s first open-source robotic exoskeleton framework, to accelerate progress in the field.

COMMUNITY

🎥 Join our next live workshop

Join our next workshop this Friday, June 27th, at 4 PM EST with Dr. Alvaro Cintas, The Rundown’s AI professor. By the end of the workshop, you’ll be able to build advanced n8n agents that react, research, and automate like pros.

RSVP here. Not a member? Join The Rundown University on a 14-day free trial.

See you soon,

Rowan, Jennifer, and Joey—The Rundown’s editorial team