INDEMIND Donutbot

BESCHREIBUNG

Donutbot is INDEMIND's home AI agent, launched at WRC 2026 and aimed squarely at households with seniors, children or pets. INDEMIND's argument for it is a distinction rather than a feature list: a companion toy chats, while an agent understands the space it is in, so Donutbot is built to move through the house, recognise who is in it and carry out tasks rather than answer questions from a fixed spot. It is small — 260 x 260 x 370 mm and 3.5 kg, or under 2 kg in the Lite version built for international markets with the same stated performance — and moves at 0.5 m/s on a wheeled base that clears 2 cm obstacles. A two-axis head and two hand joints give it the gestures a companion device needs. A 5,000 mAh battery runs it for about three and a half hours and recharges in two on a 50 W docking station, with Wi-Fi 6 and BLE 5.4 for connectivity. The navigation is purely visual. INDEMIND's spatial intelligence model builds sub-centimetre whole-house 3D semantic maps in real time and navigates on its own without remote control, closing the perception-cognition-reasoning loop on the device itself. Above that sits ROBOMIND, INDEMIND's embodied brain, which splits the work between low-computation on-device perception and full-parameter cloud reasoning — the architecture the company credits for cutting robot hardware cost sharply against conventional designs. A scheduling engine called ROBOMIND Claw coordinates tasks across the app, a PC and the robot, decomposing commands, re-prioritising dynamically and correcting its own execution drift. Two claims sit at the centre of the pitch. The first is open-ended task execution: task understanding without predefined boundaries, so the robot is not limited to a rule set. The second is a self-evolving mental model that combines multimodal perception, a per-household memory bank and a data-flywheel growth mechanism, so the machine's picture of a particular family sharpens with use. Privacy is handled on the device: faces, IDs and documents are anonymised locally by the model before anything is uploaded, and INDEMIND states multi-level encryption across transmission, storage and access.

FUNKTIONEN