Welcome, sci-fi realists…

…to the first-ever send of our new Robotics Insider Weekly newsletter. We’re excited to get this ramped up so we can bring you all of the most interest, relevant, and occasionally weird news from the robotics space. Reply and let us know what you think! 

Now, for the good stuff…

Recently, Boston Dynamics and Google DeepMind have teamed up to give embodied AI, that’s AI with a physical presence in the world, a new home inside four-legged robot dogs. The goal: teach them enough human-like reasoning skills to handle simple tasks, like perhaps walking actual dogs (our idea) or picking up laundry (their idea). 

Eventually, Boston Dynamics and Google DeepMind hope this work could make robots like Spot more useful in the real world, like the autonomous mobile robots (AMRs) that scurry across warehouse floors inspecting safety hazards. But that’s a story for another day. 

For now, Spot will have to settle for the humble dream of becoming the world’s most overqualified laundry assistant.

Anyway, here's what you need to know about robotics right now:

As advanced as industrial automation tech seems today, there are still constraints: speed, lThe company introduced an open humanoid reference design built around its Isaac GR00T robotics stack, Jetson Thor onboard compute, Unitree’s H2 Plus robot body, and Sharpa’s five-fingered robotic hands.

That may sound like a parts list because, well, it is. But that’s the point.

Humanoid robotics is still a fragmented mess of custom hardware, custom software, training bottlenecks, simulation challenges, and demos that look great until someone asks the robot to do useful work twice in a row.

NVIDIA’s pitch is that a more complete reference design could give researchers and robotics companies a stronger starting point — not a finished robot worker, but a blueprint for building and testing one faster.

What this likely means for the future: The Isaac GR00T model will be widely available from Unitree in late 2026. If the reference design gets broad adoption, it could become a baseline for what serious humanoid robotics platforms are expected to include. 

Our read: if this proves reliable, factories will be the first serious testbed — not homes.

Here's where we highlight a new feature about robots in action

NVIDIA’s Isaac GR00T mimics human movements, as dictated by humans, during simulations to learn the tasks they’re expected to perform in industrial settings. What are they capable of so far? Anything ranging from: 

A brief video on the company’s YouTube channel shows humans teaching the model these tasks in simulation environments, as well as what it would look like once they’re deployed in actual factories or warehouses. 

If you know us from The Neuron, then you know what this section is about: a round-up of the top bot stories from this week.

Each week, we'll share a funny scenario where bots aren't working as intended. Put another way: bots behaving badly (we were going to call it Naughty Botties but idk if you'd forward this to your grandma if we did)

Now, what isn’t working so well? Headlines keep churning out about the impressive capabilities of humanoids, the physical robots meant to resemble and mimic humans. Still, these are scattershot efforts by robotics companies to test the waters of what humanoids can do, without concrete plans to apply them to industry use cases anytime soon. Scaling will take time, even as CEOs believe they hold promise. 

Putting this aside, the real elephant in the room when it comes to humanoids is safety. There’s still a long way to go to ensure they don’t endanger humans and that they have built-in mechanisms to comply with safety regulations. Still, it’s worth keeping an eye on headlines to see if and when companies announce they finally have deployment-ready humanoids that are fully compliant. 

Every week, we share our top insights in this section to give you a sense of what's important to us.

P.S. If you haven’t already… follow us on Youtube!