Ready For Tomorrow #78

In partnership with

Hello and welcome back. Grab a coffee, get comfortable, and let’s take a quick step into the world where humans and robots are moving closer together every day.

Fourier Robotics just unveiled their new humanoid – GR-3.
This is not another two-meter cyborg built for factories. GR-3 is designed for homes and schools. It stands 1.55 meters tall – about the height of a middle school student – with big, expressive eyes that are meant to spark sympathy rather than fear.

GR-3 will not lift a car engine or play the piano like its older brothers GR-1 and GR-2. Instead, it comes with a built-in language model, capable of teaching, talking and assisting – a mix of teacher, caretaker and chatty friend.

Imagine Alexa on legs, not stuck inside a speaker but walking around the classroom, helping kids with homework. Or reminding grandma about her pills, with a smile and a wink. That feels like a glimpse of the future finally stepping under our roofs.

And then we get to the price. 120,000 dollars. Yes, you heard that right – a number that knocks you off your chair. Which is a pity, because GR-3 looked like the first humanoid that could actually become part of everyday life. At this price it is more of a gadget for wealthy enthusiasts than a new middle-class family member.

To sum it up – GR-3 was supposed to be a robo-buddy, but for 120,000 dollars it feels more like a classmate asking you to pay downtown apartment rent just to hang out.

Training cutting edge AI? Unlock the data advantage today.

If you’re building or fine-tuning generative AI models, this guide is your shortcut to smarter AI model training. Learn how Shutterstock’s multimodal datasets—grounded in measurable user behavior—can help you reduce legal risk, boost creative diversity, and improve model reliability.

Inside, you’ll uncover why scraped data and aesthetic proxies often fall short—and how to use clustering methods and semantic evaluation to refine your dataset and your outputs. Designed for AI leaders, product teams, and ML engineers, this guide walks through how to identify refinement-worthy data, align with generative preferences, and validate progress with confidence.

Whether you're optimizing alignment, output quality, or time-to-value, this playbook gives you a data advantage. Download the guide and train your models with data built for performance.

Download the guide

Students at the Advanced Vertical Flight Laboratory built a drone that can… unfold mid-air.
Picture a sci-fi scene: villains throw a block the size of a smartphone into the sky, and in a split second it transforms into a fully functional drone. At that point the hero is doomed, because his main weapon happens to be a stick with some mud on it.

This little machine weighs just 112 grams, has foldable arms, launches straight from the hand, and its control system keeps it stable even while spinning at seven rotations per second. In other words – it flies like a champ, even when twirling like a turbocharged toy top.

The best part? The entire project was born in an academic lab. Hats off to Connor Elliott, Hunter Denton, Nicolas Belgum and their advisor, Dr. Moble Benedict.

In space there is no room for error. Every move has to be perfect. But what if robots could learn and decide on their own?

Dr. Chapen and the team at the Naval Research Laboratory just put that idea to the test – in collaboration with NASA and their Astrobee program. Astrobees are small cube-shaped robots that float freely around the International Space Station. Normally they are controlled like a video game – by an astronaut with a joystick or by an operator on Earth. In other words, until now they were more like puppets than independent robots.

NRL took a different path: reinforcement learning – learning through trial, error and reward. You do not tell the robot “move here and grab that.” Instead, you set a goal and give it a point for completing the task. The robot tries, fails, improves – until it masters the maneuver better than you could have programmed it.

After three months of training in simulation, it was time to test on the ISS. They planned for 20 minutes, got only 5. They prepared three maneuvers, one worked. But that one was enough – the robot docked autonomously, with no joystick and no human help.

It was the first time in history that reinforcement learning controlled a free-flying robot in space. A small step for Astrobee, a giant leap for robot autonomy.

Counting inventory in a warehouse – boring, full of errors, and often stealing your weekends. Sounds familiar?
In the US, LAPP had exactly that problem. Until Corvus One showed up – the autonomous drone from Corvus Robotics.

Here is the kicker: it took just 30 days to fully deploy in a warehouse of 134,000 square feet – about 12,500 square meters, roughly two football fields. The results? Thirteen times more inventory checks, 60 percent less labor cost, and no more weekend shifts.

What makes Corvus One special? It is powered by embodied AI – there is no pilot. The drone sees and decides how to move on its own. With cameras and computer vision it identifies labels, empty spots or misplaced reels. Every flight makes it smarter – the system learns, improves, and reports errors automatically.

Each night the drone flies through the facility, dodges obstacles, and scans everything that matters. In the morning workers get a ready-made report instead of spending hours with a clipboard and a handheld scanner.

For LAPP it means full control over every cable reel. For the logistics industry it is a clear signal – night patrols of robots can replace the most tedious jobs.

To sum it up – inventory once felt like running a marathon in Excel. Now it looks more like Batman on a nightly patrol with a barcode scanner.

Circus SE – and no, not a circus troupe, but a Munich-based company – has just completed its first CA-1 robot in their brand-new mass production factory. And here’s the surprise: the factory itself was built in just six months. Faster than most of us manage a kitchen renovation.

Speaking of kitchens – CA-1 is not another box-moving robot. It is a fully autonomous kitchen on wheels. Inside runs its own Circus OS, 36 silos for ingredients, induction pots with temperature control, robotic arms with electromagnetic grippers, and even an industrial dishwasher from Winterhalter. In short – it cooks, stirs, serves and cleans. It can prepare up to 500 meals on a single load, working 24 hours a day.

To be clear – this is not some student microwave with a “pizza” button. Each CA-1 goes through more than 150 quality tests and is built from 29,000 parts – about as many as a small car. Except this car does not drive, it makes you curry or pasta.

First deliveries are about to start, and waiting in line is CA-M – the military version. Circus SE also plans factories in Europe and the US to scale their global high-tech kitchen.

To sum it up – Circus SE builds kitchen robots faster than IKEA releases a new catalog. And in their version, instead of a cooking mess, all you are left with is a full stomach.

That’s all for today. Thanks for reading and sharing a moment of curiosity with me.

Until next time – stay ready for tomorrow.

Cheers, Jacek !