A disc-shaped machine rolls quietly across the living room floor.
A five-axis arm rises from its top like a curious periscope. It spots a crumpled sock, swings toward it, and then misses. It tries again. Misses again. That is the real-world story of the most talked-about home robot of the past year, and the question it raises is whether a $1,000 arm premium is worth a coin-flip.
Why the arm fails more than half the time it tries
The Roborock Saros Z70’s mechanical arm, branded OmniGrip, is a five-axis folding system that unfolds from the top of the vacuum body. It was designed to clear obstructed areas and pick up small items before the main brush ever touches the floor.
The problem is in how the arm decides what to grab. It can only pick up items that the onboard AI recognizes from its training library, a short list that includes socks, paper balls, tissues, small towels, and slide-style slippers. If it encounters something outside that catalogue, it flags the object for avoidance rather than pickup. That is not a software glitch but a fundamental design boundary: the arm’s grip strategy is tuned to specific shapes from a finite list.
A further limit compounds that one. The moment the machine switches to mopping, the arm retracts and stays there. A floor strewn with small objects gets one job done at a time, not the seamless sweep-and-sort the demos implied.
What the machine actually promises at its $2,599 price point
Coming out of CES 2025, no single product generated more buzz than this robot vacuum with the robotic arm. The concept was genuinely compelling: anybody who has used a modern robot vacuum knows you need to tidy up before the robot can clean. The Saros Z70 promised to change that, with an arm that would move small debris into a box and sort loose slippers into a dedicated zone.
Beyond the arm, the cleaning hardware itself is legitimately impressive. The vacuum earned a Most Innovative award from Vacuum Wars for the OmniGrip, the first arm to autonomously stow small objects into a user-defined compartment. It also delivers industry-leading mopping performance, near-flawless 3D obstacle avoidance, and a multifunction dock that washes, dries, and dispenses detergent for its mop pads.
The arm moves slowly and can only lift objects under 300 grams. A standard 12-ounce can of cola weighs roughly 370 grams full. Socks, small towels, tissue paper and slide-style sandals are about as far as its ambition reaches.
The numbers that reviewers put on paper
Multiple independent hardware evaluations arrived at similar verdicts. Independent lab testing found the arm’s roughly 50 percent success rate on supported items like socks and paper balls makes it more proof-of-concept than daily workhorse. Half the time, you still need to pick up the sock yourself before the robot can vacuum under it.
Reviewers also noted that the machine takes a step backward in value relative to competing models at lower price points. The Saros 10R delivers the same 22,000 Pa suction with a comparable dock for roughly $1,000 less, meaning the entire premium over that model is the arm.
Consumer tech has a pattern of selling a headline feature at a headline price while the fine print lives in the spec sheet. Just as cloud storage egress fees surface only after data is already uploaded, the arm’s limitations surface only after the receipt is signed. The gap between demo conditions and a real, cluttered American living room is where the $1,000 premium gets spent.
Where the gap between demo and daily life is widest
Demo environments are controlled. Objects sit in expected positions, evenly lit, on clear floors. The OmniGrip arm is a sharp illustration of that gap because its recognition system was trained on a curated set of objects that does not represent the average messy household.
Many questions from press at launch drew answers like “no,” “not yet,” and “we’re considering it.” Those answers are honest. But they are also answers a buyer pays $2,599 before hearing.
The sequential job rule, arm, then vacuum, then mop, never together, means the machine adds multiple passes to what many buyers assumed would be one. Hardware categories that promise a breakthrough feature tend to carry this pattern into pricing before the feature is ready for daily use. The open-ear audio shift showed a similar dynamic: the old form factor promised one thing, real-world comfort delivered another, and the market corrected. Robot arm integration is at a comparable early stage.
What a second-generation arm would need to change
The two fixes that would close most of the gap are a larger object-recognition library and the ability to run arm sorting in parallel with vacuum suction. Roborock says more objects will be supported in future software upgrades, though it has not committed to heavier weight limits or custom object training.
The mechanical constraint of sequential-only operation is a harder problem. The arm’s retracted position is part of the machine’s idle and mopping posture by design, which means software alone cannot fully solve it without a hardware revision.
None of that makes the Saros Z70 a bad machine. Strip away the arm and it is still arguably the best-cleaning robot vacuum available, with obstacle avoidance that navigates around more than a hundred object types. The arm itself is a promising but unfinished idea wearing a $1,000 price tag, and the second generation is the version that earns the premium.
