Engineers at Northwestern University have developed modular robots designed with artificial intelligence that can recombine into different configurations, recover from damage and continue operating in outdoor environments. The machines, referred to as “legged metamachines,” consist of autonomous modules that function as individual robots but can connect to form larger systems with varied movement capabilities.
Each module contains its own motor, battery and computer, enabling it to operate independently. A single unit can roll, turn and jump. When multiple modules connect, they form structures capable of more complex movement patterns such as bounding, undulating or hopping. The robots can also right themselves if flipped over and perform movements including jumping over obstacles and spinning in midair.
The research describing the system appears in the journal Proceedings of the National Academy of Sciences. The project was led by Sam Kriegman, assistant professor of computer science, mechanical engineering, and chemical and biological engineering at Northwestern University’s McCormick School of Engineering. Chen Yu, David Matthews and Jingxian Wang, doctoral students in the university’s Center for Robotics and Biosystems, are co–first authors of the study.
To determine effective configurations for the robots, the researchers used an evolutionary algorithm that generates and evaluates different body designs. Starting from a set of modular components, the algorithm simulated combinations of modules within a virtual environment. Designs demonstrating efficient movement were retained and iteratively modified through processes analogous to mutation and selection.
The modules themselves are approximately half a meter long and resemble two rigid limbs connected by a central sphere. Inside the sphere are the electronic and mechanical components that enable operation, including a circuit board, battery and motor. The modules rotate along a single axis but can be combined into multiple arrangements that function as legs, spines or tails depending on the overall configuration.
After the simulation stage, the researchers assembled several of the designs identified by the algorithm, including robots with three, four and five modules. These systems were tested outdoors on terrain including gravel, grass, sand, mud and uneven brick surfaces. The robots were able to move across these environments without additional training or reconfiguration.
The modular design allows the system to remain functional even when parts are damaged. If a component breaks off, the remaining structure can continue operating with reduced capability while the detached module retains the ability to move independently and potentially reconnect with the larger assembly.
According to Kriegman, the research builds on earlier work in which artificial intelligence algorithms generated robot designs automatically. Those earlier prototypes demonstrated basic walking behavior but lacked the ability to sense their bodies or coordinate multiple modules. The new system integrates sensing and coordination across modules while maintaining the ability to generate unconventional designs through automated evolution.
Photo credit: Sam Kriegman/Northwestern University
