Mechanical intelligence is an emerging concept in robotics, one that focuses on embedding sensing and response capabilities directly into a machine’s physical structure rather than relying solely on electronics and software. A research team led by Professor Katia Bertoldi at the Harvard John A. Paulson School of Engineering and Applied Sciences has demonstrated a new example of this idea with a robot that operates entirely through mechanical design. Their study, describes a small walking robot that uses only one motor and no onboard computer, relying instead on rubber bands and the geometry of its structure to perform tasks such as walking through mazes and avoiding obstacles.
Kamp, L. M., Zanaty, M., Zareei, A., Gorissen, B., Wood, R. J., & Bertoldi, K. (2025). Reprogrammable sequencing for physically intelligent underactuated robots. Proceedings of the National Academy of Sciences, 122(38). https://doi.org/10.1073/pnas.2508310122
Traditional robots depend on a complex system of sensors, circuits, and algorithms to make decisions and navigate their environments. As those environments become more complicated, so do the control systems, making robots heavier, more expensive, and more prone to failure. The Harvard team’s work suggests a different approach: allow the robot’s body itself to perform part of the computation.
Professor Katia Bertoldi at the Harvard John A. Paulson School of Engineering and Applied Sciences stated,
“This is kind of an extreme version of ‘form follows function,’ where functionalities like memory, adaptability and intelligence can be enabled by geometry and material parameters.”
The project, led by graduate student Leon Kamp in Bertoldi’s lab, explored whether it was possible to design a robot whose behavior was determined by the placement and tension of elastic components instead of by software. By arranging a network of levers and rubber bands, the researchers created a system where the physical configuration of the materials dictates how the robot moves and reacts to its surroundings.
The robot’s structure is made up of a chain of flat plastic blocks joined by small levers and rubber bands. The tension in each rubber band determines the energy cost of a particular movement. When the robot moves, it naturally follows the sequence of motions that require the least amount of energy, a behavior that can be “programmed” simply by adjusting where the rubber bands are placed.
By attaching legs to this mechanism, the researchers built a small walking robot capable of moving forward, backward, and even turning, all controlled by one motor. The robot can navigate mazes and avoid obstacles without any sensors or code. It uses a pair of simple antennae to physically detect when it has hit an obstacle. When one antenna touches an object, the mechanical system automatically adjusts, changing the robot’s direction of movement.
In another experiment, the researchers showed that the same principle could be used to sort objects by mass. By tuning the elastic energy stored in the rubber bands, the robot could pick up and release objects of specific weights at different locations. This type of “mechanical programming” shows that functions such as decision-making and classification can emerge from structural design alone.
This work demonstrates that intelligence in robotics does not need to depend entirely on electronics. With the right combination of materials, geometry, and mechanics, robots can achieve simple adaptive behaviors with very little control hardware. Such robots could be smaller, lighter, and more energy-efficient, and might be useful in environments where electronics would be impractical or costly to maintain.
However, the current prototype can only perform a limited range of tasks, such as walking and object sorting. Its movements are relatively slow, and the system’s behavior can change if friction or wear alters the mechanical balance. Future research will explore how these principles could be applied to faster, more capable robots made from flexible materials.
Bertoldi’s team believes that combining this kind of mechanical intelligence with minimal electronics could produce hybrid systems that are more efficient and resilient than traditional robots. A future generation of physically intelligent machines might be able to adapt to the world around them through structure alone, blending the boundaries between design, material, and computation.
This research highlights a shift in how engineers think about robotics. Instead of viewing a robot’s body as a passive structure that simply carries sensors and motors, it suggests treating the body as an active participant in intelligence and control. The design itself can process information and make decisions based on physical laws.
For engineers and designers, this opens new directions. Mechanically intelligent systems could play a role in soft robotics, micro-scale devices, and low-cost autonomous systems that do not require extensive electronic control. The Harvard team’s robot may be small and simple, but it offers a glimpse into a future where the line between structure and intelligence is increasingly blurred.
Adrian graduated with a Masters Degree (1st Class Honours) in Chemical Engineering from Chester University along with Harris. His master’s research aimed to develop a standardadised clean water oxygenation transfer procedure to test bubble diffusers that are currently used in the wastewater industry commercial market. He has also undergone placments in both US and China primarely focused within the R&D department and is an associate member of the Institute of Chemical Engineers (IChemE).
