Figure melted its old robots with Arnold Schwarzenegger

A robot retirement worth watching

Well, that is one way to clear out the old hardware. Figure has turned retiring its humanoid robots into a Terminator-flavoured marketing event, with Arnold Schwarzenegger and molten steel. It is gloriously over the top, and I am very interested in what happens next.

The video on Figure’s YouTube channel, published on 30 September, is the sort of idea that makes a company’s next announcement considerably harder to ignore. An old robot fleet needs retiring. Somehow, Arnold gets involved. Suddenly, the comments are debating whether the robots will remember this later.

I think the marketing is brilliant. It also feels like a good moment to start paying closer attention to where humanoid robots are heading. One day, Compare may need a category for the robot that walks around your house, rather than the one that disappears under the couch.

A fairly dramatic retirement plan

In its production explainer, Figure says maintaining its older F.02 fleet no longer made sense as F.03 expanded. It wanted to protect proprietary hardware, while avoiding lengthy disassembly work that would pull technical staff away from F.04.

Figure says Schwarzenegger suggested melting the robots, then agreed to get involved. The company found a willing foundry in Imatra, Finland, and trained the robots using simulation and movements from stunt performers. According to Figure, the robots really did jump autonomously into molten steel. It is also turning recovered metal into a limited run of keepsakes. Even the scrap gets a product launch.

There are probably cheaper ways to announce that you have moved on to the next model. Very few would get people discussing your old hardware this enthusiastically.

The Terminator connection does a lot of work here. The joke is immediately understandable, even if you have never heard of Figure or have absolutely no interest in actuator design. Give people a memorable idea and they have something to share with a mate. A specification sheet has a much harder time doing that.

It also gives Figure a useful opening for the next chapter. If this is how it retires a robot, you naturally want to know what the replacement can do. I certainly do.

The comments understood the assignment

The YouTube comments have some excellent reactions. A few favourites:

@StridentSloth reckons this would be “the sort of thing that radicalized them” when the robots eventually revolt.

@thanos879 imagines AI agents having “an emergency meeting about this”.

@bpwheat takes the practical approach: “I'd give a Figure 2 a good home.”

That last one is very reasonable. Surely there was a retirement option somewhere between working in a factory and becoming molten metal. A quiet life folding towels sounds quite pleasant.

The reactions are part of why the idea works. People are making their own jokes around the campaign and giving the robots personalities. That is a much more interesting conversation than another brand asking everyone to be excited about its roadmap.

The household jobs are worth watching too

The useful next step is to look at what these machines can actually do away from a marketing stunt.

Figure’s Helix 2.5 report, published on 17 September, describes tests involving tidying toys, folding towels and making beds in 30 previously unseen homes. The company reports a 56% complete-task success rate for its pretrained system, compared with 9% for a system trained from scratch under its evaluation conditions.

Those are Figure’s own results. They cover three defined behaviours, and “unseen homes” does not mean the robot learned the jobs from nothing. The tasks were trained elsewhere before those home evaluations.

Even with that qualification, this is the kind of progress that interests me. Working in a different room, with unfamiliar objects, is relevant to whether a household robot could eventually be useful. So is the gap between a promising success rate and something you can comfortably leave to finish the job.

A robot that regularly needs rescuing halfway through a chore would be an unusual way to save time. I want to see how consistently these systems work, how they recover when something goes wrong, and how much human help they still need.

We may need a new comparison category

We already compare robot vacuums, robot mowers and pool robots. A future humanoid category would make complete sense once there are useful products Australians can realistically buy and support.

I would want the comparisons to answer fairly ordinary questions. Which chores can it finish reliably? How much supervision does it need? What happens around pets, stairs or something left on the floor? What can its cameras see, where does that information go, and can you control it?

Then there is the ownership side. We would need Australian pricing, repair options and clear warranty support. Any ongoing subscription belongs beside the purchase price. Battery life should mean useful work between charges, with the task and conditions explained.

That is where I think this becomes a genuinely exciting consumer-tech category. A machine that reliably gives you some of your time back would be worth a serious comparison. A great launch video can get us interested; everyday performance has to keep us interested.

I reckon household humanoids are coming, and we should be watching closely. Figure has absolutely got my attention. Now let’s see what it does next.