- Joined
- Jan 20, 2026
- Messages
- 345
- Reaction score
- 2,329
The computer, sensors and light-powered traction are housed in a housing less than a millimetre thick.
For a long time, it seemed that microrobots had a limit: once a device became smaller than a millimeter, it no longer had the space or energy for its "brains." As a result, the smallest prototypes moved only under external control, literally guided by magnetic coils, ultrasound, or optics, and exhibited almost no autonomy. Now, a team of researchers from the University of Pennsylvania and the University of Michigan has announced that they have finally broken through this limit and created a microrobot that can truly perceive its environment, make decisions, and act independently.
The new robot measures 210 × 340 × 50 micrometers, comparable to the size of a ciliate. Yet, it houses a fully functional computing system with memory, sensors, and actuators for movement. According to the authors, the volume of this "programmable" robot is approximately 10,000 times smaller than previous autonomous solutions, which were limited to about 1 mm.
The main problem with sub-millimeter chips isn't that it's difficult to "shrink the hardware," but that physics starts to work against developers. Energy is stored worse at the microscale, movement in viscous media is more difficult, and transistor leakage becomes more noticeable. So, previously, the choice was between miniaturization or computing and battery life. Mark Miskin, co-author of the paper and an assistant professor of electrical and systems engineering at Penn, explains the solution simply: all components were designed specifically for life at such a scale. The computer, sensors, and "motors" were custom-built to dramatically reduce power consumption and save space.
The device is powered by light, not batteries. Onboard photocells draw energy from an external LED source, and a processor then runs algorithms and controls movement. A temperature sensor with a resolution of approximately 0.1 degrees Celsius is used as a sensor. Propulsion is achieved through four platinum electrodes, which create an electrokinetic fluid flow without any moving parts. The entire system consumes approximately 100 nanowatts, comparable to the energy budget of a single-celled organism.
A special trick lies in the way these microrobots are programmed. Instead of complex external settings, the command only requires light: one LED supplies constant power, while the second blinks and transmits data after a handshake with a password. The Python interface then compiles the behavior into a bit stream, and the robot operates autonomously using an integrated clock generator. To conserve memory, the command architecture was made as compact as possible: operations are packed into 11-bit instructions, such as movement, temperature reading, and data transfer. The onboard memory is limited, about 500 bits, so an external station handles some of the heavy lifting during programming, acting as a temporary buffer.
To prove that this is not just a miniature chip but a true robot, the developers demonstrated several behavioral scenarios. During tests, the device switched between different motion modes, driving straight, turning, and moving in an arc. The speeds are still modest—a few micrometers per second in forward motion and fractions of a degree per second in turning—but the developers note that there is headroom for acceleration with increased voltage, right up to the point where water electrolysis begins.
The most interesting part begins with feedback. The robot measured the temperature, made a decision, and then "reported" the data, encoding it with its own movements. Manchester encoding was used for transmission: one step forward meant one, one step back meant zero. According to Miskin, the logic was pragmatic: since the actuators were already onboard, they could be used as a primitive communication channel without adding a separate radio module.
The manufacturing aspect is no less important. These microrobots aren't assembled by hand from microcomponents. They are manufactured entirely lithographically, like semiconductors, by the thousands on silicon wafers. The authors claim the process is compatible with standard foundries, and when scaled up, the cost per robot is less than a cent. In laboratory practice, this changes the economics of experiments: instead of single, complex samples, we have a platform that can be stamped and programmed with light, without magnetic coils, ultrasonic arrays, or expensive tracking systems.
Essentially, the work resolves the long-standing "small versus smart" tradeoff. If the approach proves successful in real-world environments, microrobots will gain what they've been most lacking: autonomy without cumbersome external infrastructure. The authors are already discussing the next steps: increasing memory, accelerating movement, teaching the devices to exchange data and work in swarms. And, as Miskin emphasized, the most valuable aspect here is programmability: with a computer onboard, such micromachines will be able to do not only what their creators conceived today, but also whatever they are programmed to do tomorrow.
For a long time, it seemed that microrobots had a limit: once a device became smaller than a millimeter, it no longer had the space or energy for its "brains." As a result, the smallest prototypes moved only under external control, literally guided by magnetic coils, ultrasound, or optics, and exhibited almost no autonomy. Now, a team of researchers from the University of Pennsylvania and the University of Michigan has announced that they have finally broken through this limit and created a microrobot that can truly perceive its environment, make decisions, and act independently.
The new robot measures 210 × 340 × 50 micrometers, comparable to the size of a ciliate. Yet, it houses a fully functional computing system with memory, sensors, and actuators for movement. According to the authors, the volume of this "programmable" robot is approximately 10,000 times smaller than previous autonomous solutions, which were limited to about 1 mm.
The main problem with sub-millimeter chips isn't that it's difficult to "shrink the hardware," but that physics starts to work against developers. Energy is stored worse at the microscale, movement in viscous media is more difficult, and transistor leakage becomes more noticeable. So, previously, the choice was between miniaturization or computing and battery life. Mark Miskin, co-author of the paper and an assistant professor of electrical and systems engineering at Penn, explains the solution simply: all components were designed specifically for life at such a scale. The computer, sensors, and "motors" were custom-built to dramatically reduce power consumption and save space.
The device is powered by light, not batteries. Onboard photocells draw energy from an external LED source, and a processor then runs algorithms and controls movement. A temperature sensor with a resolution of approximately 0.1 degrees Celsius is used as a sensor. Propulsion is achieved through four platinum electrodes, which create an electrokinetic fluid flow without any moving parts. The entire system consumes approximately 100 nanowatts, comparable to the energy budget of a single-celled organism.
A special trick lies in the way these microrobots are programmed. Instead of complex external settings, the command only requires light: one LED supplies constant power, while the second blinks and transmits data after a handshake with a password. The Python interface then compiles the behavior into a bit stream, and the robot operates autonomously using an integrated clock generator. To conserve memory, the command architecture was made as compact as possible: operations are packed into 11-bit instructions, such as movement, temperature reading, and data transfer. The onboard memory is limited, about 500 bits, so an external station handles some of the heavy lifting during programming, acting as a temporary buffer.
To prove that this is not just a miniature chip but a true robot, the developers demonstrated several behavioral scenarios. During tests, the device switched between different motion modes, driving straight, turning, and moving in an arc. The speeds are still modest—a few micrometers per second in forward motion and fractions of a degree per second in turning—but the developers note that there is headroom for acceleration with increased voltage, right up to the point where water electrolysis begins.
The most interesting part begins with feedback. The robot measured the temperature, made a decision, and then "reported" the data, encoding it with its own movements. Manchester encoding was used for transmission: one step forward meant one, one step back meant zero. According to Miskin, the logic was pragmatic: since the actuators were already onboard, they could be used as a primitive communication channel without adding a separate radio module.
The manufacturing aspect is no less important. These microrobots aren't assembled by hand from microcomponents. They are manufactured entirely lithographically, like semiconductors, by the thousands on silicon wafers. The authors claim the process is compatible with standard foundries, and when scaled up, the cost per robot is less than a cent. In laboratory practice, this changes the economics of experiments: instead of single, complex samples, we have a platform that can be stamped and programmed with light, without magnetic coils, ultrasonic arrays, or expensive tracking systems.
Essentially, the work resolves the long-standing "small versus smart" tradeoff. If the approach proves successful in real-world environments, microrobots will gain what they've been most lacking: autonomy without cumbersome external infrastructure. The authors are already discussing the next steps: increasing memory, accelerating movement, teaching the devices to exchange data and work in swarms. And, as Miskin emphasized, the most valuable aspect here is programmability: with a computer onboard, such micromachines will be able to do not only what their creators conceived today, but also whatever they are programmed to do tomorrow.
