Mysterious Robot Team Releases New Demo Videos of 'MVP' Model, QbitAI Reports

The report said the model behind the robots is internally codenamed 'MVP,' an abbreviation for 'Make Veritable People.' When asked why the team did not publicly claim ownership of the earlier video, the team said the model, like a person, should speak for itself. It added that the robots would soon be put on streets in open environments for public viewing.

The report describes one demo in which a robot holding a cup of water is pushed by a person with a stick. The robot shifts its body to avoid the push and returns to its original position without spilling water. When the person pushes several soda cans on a table, the robot reaches out with its other hand to keep them from falling. The report says this indicates an understanding of physical dynamics, in contrast with mainstream vision-language-action and world-action models that often fail when tasks involve occlusion, contact or deformation.

Another clip shows a robot completing a zero-shot household tidying task. It returns a toy to a shelf, pulls a storage basket closer, removes a black hat and a purple fitness ring from the basket and hangs them on a rack, then throws a large flower-shaped cushion onto a sofa. The team claimed an 80% zero-shot success rate for such tasks, according to the report. The robot's actions suggested it understood spatial relationships, object attributes and human habits, such as dragging a heavy basket along a smooth floor and avoiding needless effort when tossing the cushion.

A third demo features two different types of robots working together to straighten a bed sheet. The team said both robots were trained exclusively on human videos. One robot pulls two corners while the other bends and steps forward to make the sheet taut, then they shake it together. The report says this shows cross-embodiment coordination and what the team calls self-evolution.

In the fourth clip, a robot demonstrates energy-efficient decision-making by turning itself into a moving clothes rack: it drapes a black scarf around its neck, slides a purple ring onto its arm and lets it fall to the elbow, then picks up other objects before carrying them away in one trip. The report says the robot appears to use an understanding of physical rules and human behavior to choose an optimal energy path, rather than relying on statistical guesses that would typically lead to multiple trips.

The report noted that the MVP model appears to unify physical reasoning and human behavioral logic, enabling stable long-horizon tasks and continuous improvement. It also said the robots showed individual characteristics, with each one having its own way of doing tasks, and several observers called the demos an important milestone toward functional embodied intelligence.