Putting a basket back, collecting a bin and moving on to the next room: Xiaomi has shown a robot tidying an apartment complex. It rolls on wheels and moves objects with two arms. A single assignment involves a whole sequence of smaller actions.
Xiaomi says the trial at its Youth Apartments uses Xiaomi-Robotics-1, an AI model for robot control. In a 30-minute round it reported more than 50 task sequences across three spaces. This is a trial of a system still in development.
Learning from people with grippers
The test draws on a large set of examples. Xiaomi’s technical overview describes more than 100,000 hours of recorded actions across over 1,700 settings. Much of it comes from people demonstrating movements with wearable grippers. The hours do not represent time one robot spent practising in an apartment.
The model first learns how actions change a scene. An object might start on a table and end up in a basket. The long recordings are broken into short examples, each with a description of what happens.
Next the system practises with real robots and instructions such as ‘put the basket back’. It brings together what the robot sees, what it is asked to do and how it moves. The robot has to recognise the object and work out a movement its own arms can make.
A chore takes more than one move
In the demonstration, the robot moves items around a sofa and table, tidies another space and returns supplies to a cleaning cart. It must pick up where its previous action left off.
The footage is sped up, with labels including 4×, 10× and 20× playback. The short edit is therefore no measure of how fast the robot works in real time.
The challenge is the sequence of small decisions. After picking something up, the robot has to set it down properly before the next task starts somewhere else. The trial shows how Xiaomi is extending its control system from individual actions to longer jobs.

























