Roborock Saros Z70 Gains AI Object Labeling, Drag-and-Drop Programming

Big update, bigger promise

Roborock today is rolling out a software upgrade that adds AI object labeling and a drag-and-drop programming interface to the Saros Z70, bringing consumer-grade robotics closer to personalized household automation. The update lets owners teach the robot to name and remember objects from a favourite slipper to a child's toy and then create simple cleaning routines by dragging labeled objects into visual workflows in the app. This shift transforms the Z70 from a high‑end cleaner into a reactive home agent that adapts to life’s clutter without frequent human intervention.

What it can now do

The new object-labeling capability uses the Z70’s camera and StarSight perception stack to attach semantic tags to items and store them in room maps, enabling rules like “avoid or remove slippers at night.” The drag-and-drop programming canvas exposes building blocks (identify, avoid, nudge, return-to-base) so non‑technical users can chain behaviors into one‑tap routines. Roborock positions this as an accessibility and convenience play: households can automate repetitive decisions keep toys off carpets during mopping, clear pathways for the arm without writing code.

Under-the-hood advances

The capability rests on the Saros Z70’s StarSight Autonomous System 2.0, which fuses LiDAR, structured light and camera feeds into a unified SLAM and recognition pipeline that runs on-device, reducing latency and preserving privacy. Time-of‑Flight sensors and a ToF-enabled VertiBeam give denser 3D point clouds for robust object segmentation, while on-board AI prunes false positives so the drag‑and‑drop actions execute reliably even in cluttered homes. Engineers noted this approach minimizes cloud dependence while allowing over‑the‑air model updates to improve recognition over time.

Everyday scenarios made smarter

In practice, the update opens practical workflows: parents can set the Z70 to scoot small toys into a basket before mopping, pet owners can flag feeding areas as off-limits, and older adults can create daily escort routes that clear walkways. The robotic arm gains behavioral context for example, it can gently reposition a fallen sock and then vacuum the freed patch turning previously futile cleaning attempts into completed tasks. These scenarios emphasize time saved and fewer interruptions for users, a real step toward meaningful hands-off maintenance.

Hardware blueprint snapshot

The Saros Z70 is a compact, wheeled vacuum with brush roll and suction system measuring 35.3 x 35.3 x 7.98 cm and weighing 4.9 kg, designed to slip under low furniture while carrying a five‑axis OmniGrip arm and sensors that include LiDAR, camera (StarSight system), Time‑of‑Flight, cliff and bump sensors, plus structured light arrays. It runs Roborock’s proprietary OS with StarSight navigation firmware, and Roborock rates typical battery lifespan at 2–3 years; safety features include cliff detection, anti‑collision bumpers, AI obstacle avoidance and an app child lock. Mobility is conventional wheeled locomotion with aggressive suction and a brush-roll for mixed surfaces.

How it stacks up

Against rivals, the Z70’s strengths are its mechanical arm and advanced StarSight recognition features none of the four competitors combine in the same way giving it unique abilities to clear small obstructions and act on labeled objects; however, rivals like the Ecovacs X9 Pro Omni and Dreame Ultra series still challenge on docking feature sets and mop‑station sophistication, while the iRobot Roomba j9+ maintains a reputation for dependable mapping and broad ecosystem integrations. The Z70 may trail certain competitors in base‑station cleaning or replacement‑ecosystem maturity, but it leads where behavior-level autonomy and object interaction matter most.

Competitors

Ecovacs X9 Pro Omni

Ecovacs

Dreame L50 Ultra

Dreame

Dreame X50 Ultra