Scientists Unveil Cell-Sized Robots That Sense, Compute and Swim on Their Own

Under the glare of a blue-white LED in a University of Pennsylvania lab, the water in a shallow dish looks perfectly still. Only when the microscope’s magnification is cranked up do the machines come into view: flecks smaller than a grain of salt, shuddering and swinging through the fluid in tight arcs.

Each fleck is a robot—about the size of a single-celled organism—that carries its own solar panels, temperature sensors, computer and motor. And for the first time at this scale, those parts work together so the robot can sense its surroundings, make a decision and move on its own.

Engineers at the University of Pennsylvania and the University of Michigan report they have built what they describe as the world’s smallest fully programmable, autonomous robots, a milestone they say closes a decades-old gap between “smart dust” sensors and true micromachines.

The work, detailed Dec. 10 in the journal Science Robotics, builds on an earlier propulsion study published in July in the Proceedings of the National Academy of Sciences. Together, the papers describe cell-sized robots—roughly 200 by 300 by 50 micrometers—that can swim through liquid, follow temperature gradients and even “dance” to transmit data, all under programs stored in their own on-board memory.

“We’ve made autonomous robots 10,000 times smaller,” said Marc Miskin, an assistant professor of electrical and systems engineering at Penn who led the microrobotics effort. “That opens up an entirely new scale for programmable robots.”

How the robots are built

The robots look nothing like the humanoid machines in science fiction. Viewed from above, each is a flat rectangle etched on a silicon wafer, its surface blanketed with microscopic solar cells. Buried inside that structure is a custom-designed processor and digital memory circuit, built by collaborators at the University of Michigan’s College of Engineering, that runs on about 75 nanowatts of power.

“That’s over 100,000 times less power than a smart watch,” said David Blaauw, a professor of electrical engineering and computer science at Michigan whose lab designed the chip. “Working at that level forces you to rethink what a computer and its instruction set look like.”

Along two sides of each robot sit a pair of temperature sensors. On the other two sides, platinum-based electrodes act as motors. When light shines on the robot, its solar cells generate current that flows out through these electrodes into the surrounding liquid.

That current doesn’t spin any gears or hinges. Instead, it creates an electric field in the fluid, pushing nearby ions along the robot’s surface. The ions, in turn, drag the water with them, generating a gentle thrust that propels the robot forward. The mechanism, known as electrokinetic propulsion, has no moving parts and can run for months under steady illumination, the team reports.

From propulsion to autonomy

In the July PNAS paper, the group showed that simple, electronics-laden “chips” a few hundred micrometers across could scoot along at about one body length per second using this effect, and that their speed could be tuned by adjusting the electrical and chemical conditions. The new work adds something crucial: a true sense–think–act feedback loop.

In one demonstration described in Science Robotics, the researchers placed the robots in a tiny pool of water with one side gently heated and the other left cool. Because each robot carries a pair of sensors on opposite flanks, it can measure a slight temperature difference across its own body. That information goes into its microprocessor, which runs a short program—a dozen or so instructions—that determines how to drive the electrodes.

If the robot’s readings suggest it is drifting toward cooler water, the program switches to a wide, searching gait; if temperatures warm, the robot pivots to stay within the gradient. Over dozens of trials, the researchers reported, the tiny machines consistently changed their swimming patterns in response to changes in temperature, without real-time instructions from the outside.

“Being able to put a brain, a sensor and a motor into something almost too small to see is really just the first chapter,” Miskin said.

Communicating by “dance”

The robots can also communicate simple measurements through motion. In another set of experiments, a robot recorded the temperature at its location and then encoded that value in the amplitude and pattern of its swimming, which researchers recorded on video and decoded later. Blaauw likened the approach to a honeybee’s waggle.

“We have them report back in the wiggles of a little dance,” he said. “It’s a very low-bandwidth way to talk, but at this scale it works.”

Each robot has a unique identifier and can be reprogrammed after fabrication. The team uses modulated light—pulses projected through a microscope—to send new instructions that the on-board memory stores. In test runs, different robots in the same dish were given different behaviors, allowing simple swarm experiments in which some followed one path while others held position or reported sensor readings.

Manufacturing at scale—and current limits

The devices are fabricated in large batches on standard silicon wafers, using the same lithography techniques that produce computer chips. Hundreds to thousands of robots can be made at once on a piece of silicon about the size of a fingernail. The researchers estimate that, at industrial scale, individual robots could eventually cost about a cent apiece, although current laboratory prototypes are much more expensive.

The advance comes after decades of incremental progress in microscopic machines. In 2020, Miskin and colleagues at Cornell University unveiled four-legged robots just 40 to 70 micrometers long that could walk when jolted by a laser. Around the same time, other groups built micromotors driven by magnetic fields, ultrasound or chemical reactions, and Michigan’s lab shrank fully programmable computers into one-cubic-millimeter “micro motes.”

But most of those systems lacked one or more critical elements: either they could move only when steered continuously by external fields, or they could sense and compute but not propel themselves. By contrast, the new Penn–Michigan robots integrate sensing, computation, power harvesting and locomotion in a single, submillimeter body.

Outside experts say that combination is notable, but they caution against leaping too quickly to visions of medical nanobots patrolling the bloodstream.

For now, the robots operate only in carefully controlled, low-conductivity liquids. Salty environments such as blood or seawater damp the electric fields that drive electrokinetic propulsion. The machines also rely on light both for power and for data, which limits their usefulness deep inside opaque tissue.

Their computing power is modest—far less than a wristwatch, let alone a smartphone—and so far the only integrated sensor is for temperature. Adding chemical sensors, cameras or more complex communication would require further miniaturization and power savings.

Researchers and ethicists have also pointed to longer-term questions about what happens when vast numbers of microscopic robots are released into bodies or the environment. Many of the materials used—including silicon, glass-like coatings and platinum—are familiar from medical implants and electronics, but the risks of deploying swarms of such particles are not well understood.

Funding for the project includes grants from the National Science Foundation, the Air Force Office of Scientific Research, the Army Research Office and corporate support from Fujitsu, reflecting interest not just in medicine but in defense and manufacturing uses. Military agencies have for years explored “smart dust” and micro-robotic systems for sensing and surveillance, although the current devices are far from field deployment.

Despite the caveats, the Penn and Michigan engineers argue that demonstrating any autonomy at this scale is an important step. It suggests that future microrobots could be mass-produced and then assigned roles in software rather than being custom-built for a single task.

“You don’t have to decide what each robot will do at the moment you manufacture it,” Miskin said. “You can build a general-purpose platform and program it later.”

On a lab bench in Philadelphia, that future still looks humble. Under the microscope, the robots resemble specks of dust, jittering across a glass slide under an artificial sun. Yet in those tiny motions, engineers see the basic blueprint of a robot—sense, think, act—squeezed down to nearly the scale of life itself.