Video Friday: Robot Decommissioning Takes a Fun Turn

The Humanoid Fatigue is Real

I’ve spent the last decade watching founders try to build things that look like us, talk like us, and move like us. Usually, it’s a distraction. We see another week of robotics updates, and the signal-to-noise ratio is getting worse. Figure is doing stunts, and while the engineering is impressive, it feels more like a PR play for the next funding round than a breakthrough in utility. As a builder, you have to ask: are we solving problems, or are we just making better toys?

We are seeing a flood of humanoid videos where the robot climbs a suitcase or catches a box. It looks great on Twitter, but if you look closely at the recent Workhorse demonstrations using the Unitree G1, the real story isn't the hardware. It's the move toward learning manipulation from robot-free human demonstrations. This is where the AI side of the house is actually helping. We’re moving away from hard-coding every joint movement and toward models that understand intent. But let’s be honest: until these things can handle a communication failure without looking like they’ve had a nervous breakdown, they aren't ready for the job site.

The Plastic Ball Pivot

One of the more bizarre developments this week is Devanthro selling plastic spheres for a thousand dollars a pop. They’re marketing them as a piece of robotics history, essentially decommissioning parts of their Reachy units. It’s a clever way to recoup R&D costs, I suppose, but it highlights a bigger issue in the hardware space. Burn rates are lethal. When you start selling the literal scrap of your project as a holiday gift, it tells me the path to monetization for general-purpose humanoids is still a long, winding road.

For those of us in the trenches, this is a reminder that hardware is hard and expensive. If you aren't building for a specific vertical—like the KUKA robot arms currently trying (and mostly failing) to automate car washes—you're just burning cash. The KUKA car wash project is a perfect case study. It sounds great on paper: a robot that models your car and scrubs it perfectly. In practice, the reviews are mediocre. Why? Because the real world is messy, unpredictable, and doesn't fit into a clean simulation. Builders should take note: the last 5% of the user experience is where 90% of the work lives.

Navigation vs. Novelty

If you want to see where the real progress is happening, look at the boring stuff. MIT’s SANDO navigation system isn't flashy. It doesn't dance, and it doesn't rap. What it does is provide mathematical guarantees for obstacle avoidance in unfamiliar environments. In the crypto world, we talk about trustless systems. In robotics, mathematical safety guarantees are the equivalent. If you can’t prove the robot won’t hit a human, you don't have a product; you have a liability.

We’re also seeing interesting developments in soft robotics. NC State researchers are looking at feather stars to create underwater robots that use only two pneumatic inputs. This is the kind of efficiency I love. Instead of a hundred sensors and twenty motors, they’re using physics and smart material design to achieve complex movement. Founders often over-engineer the solution when they should be simplifying the problem. You don't always need a humanoid to do a task; sometimes you just need a smart piece of plastic and some air pressure.

The Transparency Problem

I have to call out the editing in some of these recent demo videos. We’re seeing clips from companies like Sanctuary AI where the camera cuts right as the robot achieves a difficult grip. As someone who has audited enough pitch decks to see through the fluff, this is a red flag. If you’re a builder, don't hide the failures. The industry knows this stuff is hard. When you hide the 'jump' in the video, you lose the trust of the very people who might actually help you solve the problem.

We’re seeing a lot of 'Generalist' grippers that look simple but handle complex manipulation. That’s the sweet spot. The hype is in the humanoid body, but the utility is in the hand. The moment we stop trying to make robots look like C-3PO and start focusing on the end-effector—the part that actually touches the world—is the moment we start making real money.

What This Means for the Next Cycle

The robotics space is currently where AI was three years ago: lots of impressive demos, very few sustainable business models. The winners of the next five years won't be the ones with the best dancing videos. They’ll be the ones like the team at Ecole Polytechnique Fédérale de Lausanne, working on thread-like linear motors. It’s small, it’s niche, and it’s a fundamental building block.

If you're building in this space, my advice is to ignore the humanoid arms race. Let the big-money firms fight over who can make a robot walk most like a human. Focus on the navigation bottlenecks, the sensor fusion, and the specific tasks that currently require a human to do something dangerous or mind-numbingly boring. The decommissioning of robots into $1,000 plastic balls should be a wake-up call. Innovation is great, but utility pays the bills.

The real breakthrough isn't a robot that can dance; it's a robot that can fail gracefully and navigate safely without a scripted path.

We’re heading toward a period of consolidation. The companies that are relying on 'cool' will fade, and the ones providing 'boring' infrastructure will become the backbone of the next industrial layer. Whether it's underwater soft robots or better navigation algorithms, the focus is shifting from form to function. And honestly, it’s about time.

Read the original at IEEE Robotics →