‘Make me a chair’: robotic assembly system responds to text requests
23 December 2025 
Image: Courtesy of the researchers
US researchers have developed an AI-driven robotic assembly system that builds physical objects requested as text descriptions. It builds the objects automatically from a set of prefabricated parts, and can iterate on the design based on user feedback.
The project aims to make design faster and more accessible for non-experts. Current CAD (computer-aided design) platforms require expertise to master, and often incorporate such a high level of detail that they are not suitable for brainstorming or rapid prototyping.
The new system, developed by experts from MIT and elsewhere, including Google Deepmind and Autodesk Research, uses a generative AI model to build a 3D representation of an object’s geometry based on user prompts. A second generative AI model then figures out where different components should go, according to the object’s function and geometry.
The researchers have used the system to build furniture, such as chairs and shelves, from two types of premade components (structural components and panel components). The parts can be disassembled and reassembled, reducing waste. The researchers found that more than 90% of participants in a survey preferred the objects made by the AI-driven system to those made using other approaches.
The technique could be particularly useful for rapid prototyping of complex objects such as aerospace components and architectural objects. In the longer term, it could be used to fabricate furniture or other objects in homes, avoiding the need to transport bulky products.
“Sooner or later, we want to be able to communicate and talk to a robot and AI system the same way as we talk to each other to make things together,” says lead researcher Alex Kyaw, a graduate student in the MIT’s departments of Electrical Engineering and Computer Science (EECS), and Architecture. “Our system is a first step towards enabling that future.”
While generative AI models are good at generating 3D representations, known as meshes, from text prompts, most do not produce uniform representations of an object’s geometry that have the component-level details needed for robotic assembly. Separating these meshes into components is challenging because it depends on the geometry and functions of the object and its parts.
The researchers tackled these challenges using a vision-language model (VLM) – a powerful generative AI model that has been pre-trained to understand images and text. They tasked the VLM with figuring out how the two types of prefabricated parts could fit together to form an object.
“There are many ways we can put panels on a physical object, but the robot needs to see the geometry and reason over that geometry to make a decision about it,” Kyaw explains. “By serving as both the eyes and brain of the robot, the VLM enables the robot to do this.”
The user prompts the system with text – such as “make me a chair” – and provides it with an AI-generated image of a chair to start. The VLM then determines where to place the panel components on top of the structural components, based on example objects that it has seen before. For instance, it can determine that a chair’s seat and backrest need panels to act as surfaces for someone sitting and leaning on it.
The system outputs this information as text – such as “seat” or “backrest.” Each surface of the chair is then numbered, and the information is fed back to the VLM. The model chooses labels that correspond to the geometric parts of the chair that should receive panels on the 3D mesh to complete the design.
The user can refine the design by giving new prompts, such as “only use panels on the backrest, not the seat.”
“The design space is very big, so we narrow it down through user feedback,” Kyaw explains. “We believe this is the best way to do it because people have different preferences, and building an idealised model for everyone would be impossible.”
Once the 3D mesh is finalised, the robot assembly system builds the object from prefabricated parts.
The researchers compared the results of their technique with an algorithm that places panels on all upward-facing horizontal surfaces, and another that places panels randomly. More than 90% of those surveyed preferred the designs made by AI system.
The researchers also asked the VLM to explain why it chose to put panels in particular locations. “We learned that the vision language model is able to understand some degree of the functional aspects of a chair, such as leaning and sitting, to understand why it is placing panels on the seat and backrest,” Kyaw reports. “It isn’t just randomly spitting out these assignments.”
The researchers now want to enhance their system to handle more complex and nuanced user prompts – such as tables made out of glass and metal. They also want to incorporate additional prefabricated components, such as gears, hinges and other moving parts, giving the objects more functionality.
“Our hope is to drastically lower the barrier of access to design tools,” says senior researcher Professor Randall Davis, a member of MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). “We have shown that we can use generative AI and robotics to turn ideas into physical objects in a fast, accessible, and sustainable manner.”
The researchers presented a paper on the project at the recent Conference on Neural Information Processing Systems in Sydney, Australia.
