Matic rolled out Cues on August 13, adding opt-in voice and gesture controls to its robot vacuum. Owners can point at a mess and say "clean this," issue room-level commands in more than 70 languages, or ask the robot to follow them. Gesture and wake-word detection run on the robot, while post-wake voice commands are sent anonymously to Google's Gemini API for parsing.
Matic rolled out Matic Cues on August 13, adding voice and gesture controls to its robot vacuum through a software update. The new interface lets an owner point at a spill or debris and say "clean this," direct the robot to a named room, or ask it to follow them. Matic says the voice interface supports more than 70 languages; The Verge described Cues as a free update.
Pointing becomes a cleaning command
Matic's launch page says the robot detects a pointing hand, estimates its direction in 3D, and uses that direction to locate the area to clean. The same interface accepts commands such as vacuuming or mopping a room, cleaning carpet edges, running a schedule, and coming to the user. Matic says the microphone array can hear a command from up to 10 meters away.
The controls are opt-in. Matic's privacy documentation says microphones are off by default and can be disabled again in settings. The wake phrase, voice direction, and gesture recognition use three models running on the robot. Camera data used for gestures stays on the device and is discarded, according to the company.
The cloud boundary matters
Voice processing is not entirely local. After the robot detects "Hey Matic" on-device and plays a chime, it sends the subsequent command over an encrypted connection to the Gemini API. Matic says the request is sent without user information and is not stored. The company plans a smaller on-device command parser, but says that version will probably cover fewer commands and may initially support only English.
For robotics and ML teams, Cues is a concrete hybrid-architecture example: latency- and privacy-sensitive perception remains at the edge, while broad multilingual language understanding uses a cloud model. That split makes the consent boundary and the audible wake confirmation important product controls. It also leaves open practical questions about recognition reliability across rooms, accents, background noise, pointing angles, and unusual floor layouts that will require evidence beyond launch demonstrations.
Key Points #
- 1Matic Cues adds opt-in voice and pointing controls for spot cleaning, room commands, schedules, and follow-me behavior.
- 2Wake-word, voice-direction, and gesture models run on the robot, while post-wake commands are parsed through the Gemini API.
- 3The hybrid edge-cloud design preserves local camera processing but makes consent, wake confirmation, and command-data handling material product controls.
Scoring Rationale #
Cues turns natural speech and pointing into operational commands on a shipping home robot, making it a useful edge-AI and human-robot interaction case. Impact is moderated because launch materials and one hands-on report do not yet establish reliability across varied homes and users.
Sources #
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.