The first consumer-ready manipulator?
Roborock announced a version of its robot vaccuum with an arm at CES 2025
It’s finally happened: a company has announced a consumer home robot with a mobile manipulator, for sale, releasing this year. That company was not one of the big humanoid robot companies, it wasn’t Tesla, it was Roborock: a Chinese consumer goods company well-known for making some of the best quality robot vacuums on the market.
The Saros Z70 has a five DOF robot arm and can even “serve as a companion.” According to their sales reps, it can recognize 108 objects out of the box, with potential for teaching more (socks, sandals, light dog toys were all mentioned as options). The payload is, unsurprisingly, low — only 300 grams — but it’s enough for clutter and, importantly, is extremely safe.
Similar Robots
There are three similar robots that I know of, which are either on the market now or are landing soon (i.e., this year):
YC2024-funded Innate is working on an AI-powered robot called Maurice, which is selling for about $2000 and is supposed to ship in 2025
Similarly, Hackerbot is working on a low-cost robot for… hackers
There is always the Hello Robot Stretch, which is designed for researchers and developers and has a vibrant open-source ecosystem
Of course, that’s not a complete list. Matic Robots showed their dev kit at CoRL 2024 - with a 3d printed example of an arm mounted on top. If this works for Roborock, expect everyone else to follow suite in short order.
Why this task?
Tidying up small objects from the floor is a nice, well-defined robot task.
It meshes well with existing capabilities — moving objects out of the way of your vacuum-robot makes the vacuum robot itself much more valuable, because you need to do less tidying before the robot can act.
You can focus on a relatively small set of objects. Open-vocabulary perception (i.e., handling any object the robot might find) is very difficult, and grasping is likewise very hard.
The robot itself is inherently safe. A robot that’s about a foot tall is not going to be able to harm anyone. It only needs the strength to pick up a sock, so that arm is not going to be crushing any fingers.
The latter two points dramatically decrease the difficulty of the AI problem that needs to be solve, while also dramatically decreasing the necessary compute. You don’t need ChatGPT with its hundreds of millions of parameters to perform this task.
This may seem disappointing to fans of, for example, Tesla’s Optimus — who are looking forward to a robot that can cook and clean. But by making these AI requirements minimal and complete, that means we can definitely see something useful later this year.
And if you’re not looking for useful, but want to experiment with AI in your home — well, take a look at that list in the previous section.
When does it come out?
Roborock says it comes out this summer, with preorders shipping starting in April. The robot should cost about $2000. It’s exciting to see the first useful, fully autonomous consumer mobile manipulators starting to come out, and I expect we’ll see many more in the next couple years.
"A robot that’s about a foot tall is not going to be able to harm anyone" is a great point that hasn't occurred to me about household robots.
That is a pretty clever use case, though doing some vacuuming this weekend the things I had to move were often much heavier. Also, $2000 is way more than I would consider for such a device.