In high-tech labs, robots are now being trained to poke, prod, and measure materials just like human experts. These robotic systems can test how materials respond to light, a property called photoconductance, which is key for developing things like solar cells and sensors.
But here’s the twist: instead of relying on tons of labeled data or pixel-perfect precision, the MIT team created a self-supervised system. That means the robot learns on its own, using smart algorithms to copy how scientists work; no hand-holding is required.
The result? A fully autonomous robot that can quickly and reliably test materials, boosting both speed and accuracy in self-driving labs. It’s like giving your lab assistant a PhD and a pair of laser-focused eyes.
By blending expert knowledge with machine learning, the team taught a robotic probe how to smartly choose the best spots on a material to measure its photoconductance, and how well it responds to light. A clever planning system also helps the robot move quickly and efficiently between those points.
In a 24-hour test run, this self-driving lab assistant took over 125 precise measurements per hour, outperforming other AI methods in both speed and accuracy.
Why does this matter? Because faster, smarter testing means scientists can discover better semiconductor materials more quickly, paving the way for more powerful, efficient solar panels and a brighter, greener future.
Earlier research focused on quickly creating and imaging new perovskite materials to study their properties. But when it comes to measuring photoconductance and how materials respond to light, nothing beats a hands-on approach: placing a probe, shining light, and recording the electrical reaction.
To make this process both fast and accurate, researchers had to come up with a solution that would produce the best measurements while minimizing the time it takes to run the whole procedure. Doing so required the integration of machine learning, robotics, and material science into one autonomous system.
Here’s how it works:
The robot system starts by taking a photo of a printed perovskite sample. Using computer vision, it slices the image into segments. These segments are analyzed by a custom neural network trained with insights from chemists and materials scientists. This lets the robot think like a human expert, identifying the best spots to probe based on shape and composition.
A path planner then maps out the fastest route for the robotic probe to hit all the key points. Surprisingly, adding a bit of randomness to the algorithm helped it find even shorter paths.
The neural network is self-supervised; it doesn’t need labeled data to learn. That means it can adapt to all kinds of optimal contact points directly on a sample image.
As Professor Buonassisi puts it, “It’s almost like measuring snowflakes.” No two samples are the same, but this robot can handle them all, quickly, precisely, and with a scientist’s intuition.
After assembling their robotic system from scratch, MIT researchers put it to the test, and the results were nothing short of impressive.
The system’s custom neural network outperformed seven other AI models, finding better probe contact points with less computing time. Its path-planning algorithm also beat the competition, consistently charting shorter, more efficient routes.
In a 24-hour fully autonomous run, the robot made over 3,000 unique photoconductance measurements; that’s more than 125 per hour, without any human help. This high-speed precision allowed researchers to spot both high-performing hotspots and areas of material degradation in the samples.
“Gathering such rich data so quickly, without human guidance, opens the door to discovering new high-performance semiconductors,” said researcher Siemenn. This is especially promising for sustainable tech, like next-gen solar panels.
The team now aims to expand this system into a fully autonomous lab, one that could revolutionize how we discover and develop materials for a cleaner, more energy-efficient future.
Journal Reference:
- Alexander Siemen, Basita Das, Kangyu Ji, et al. A self-supervised robotic system for autonomous contact-based spatial mapping of semiconductor properties. Science Advances. DOI: 10.1126/sciadv.adw7071