Researchers at Columbia University have developed a robotic system capable of physically modifying and expanding its structure by integrating components from its environment or other machines. The system, detailed in a study published in Science Advances, introduces a process referred to as “Robot Metabolism,” which allows robotic units to assemble, adapt, and repair themselves using modular parts.
The work was carried out by scientists at Columbia Engineering’s Creative Machines Lab in collaboration with the University of Washington. The team designed a modular robotic unit known as a Truss Link—a bar-shaped module inspired by the Geomag toy and equipped with magnetic connectors. These connectors enable the links to join at various angles, allowing the modules to self-assemble into increasingly complex forms.
Using this system, researchers demonstrated how individual Truss Links could self-assemble into flat shapes and subsequently evolve into three-dimensional robots. The assembled robots were shown to be capable of modifying their own structure by incorporating new modules. In one instance, a tetrahedral robot added a link to function as a support, which increased its downhill movement speed by over 66 percent.
According to lead author Philippe Martin Wyder, the development marks a step toward physically autonomous machines capable of sustaining themselves. Wyder and his colleagues suggest that such robotic systems could eventually be deployed in environments where human maintenance is impractical, such as disaster zones or space missions.
Co-author Hod Lipson, director of the Creative Machines Lab, described the process as a form of machine metabolism. He likened the modular robotic approach to biological systems that grow and adapt through the reuse of basic building blocks. Lipson emphasized that, while the concept of self-replicating robots may raise concerns, the technology addresses a growing need for autonomous maintenance in increasingly automated environments.
The researchers argue that the integration of physical adaptability into robotics adds a new dimension to machine autonomy, expanding current capabilities beyond cognitive functions such as artificial intelligence and machine learning.
Photo credit: Creative Machines Lab
