Drones Can Dodge Obstacles at High Speeds, Thanks to Uncertainty

A little bit of uncertainty never hurt anyone, right? In tech, the answer to that question is almost always “no.” But with high-speed drones flying through cities with tall skyscrapers, busy warehouses, and thick forests, the answer could very well be “Yes!”

Recently, researchers from the Computer Science and Artificial Intelligence Lab (CSAIL) at MIT developed a new method to help fast-moving drones avoid obstacles in their flights. Their innovative approach takes the route of building a little bit of uncertainty into the drone’s mapping system.

The Drone Data Crunch

Most modern drones and self-driving cars employ a mapping technology called “simultaneous localization and mapping” (SLAM). SLAM scans the environment’s depth to build a 3D map that the drone or car then uses to guide itself to the destination.

But the issue with SLAM is that it’s very data- and processor-intensive. As a result, the drone’s movements can lag or react incorrectly to its surroundings. And the solution is a surprising, yet elegant one: simply don’t take in as much data about the surroundings.

A Glitch-Free Future

With CSAIL’s new approach, however, the drone development team focused on creating a method that “planned less and reacted more.” The team dubbed their creation the “NanoMap”; it works to measure 3D data from the surroundings, but the information isn’t aggregated into one large data-consuming map. The data is broken up into “chunks” of maps, like snapshots. As a result, the drone spends less time computing and more time figuring out how to react to incoming obstacles.

“Because we’re not taking hundreds of measurements and fusing them together, it’s really fast,” says one of the team members, Ph.D. candidate Pete Florence. “And when we want to plan our way around the world, we just search back through the views we already have.”

Accelerating Towards a Faster Future

Florence contends that NanoMap isn’t suited for drones that absolutely need high-quality maps, like those employed in search-and-rescue missions or those surveying vast swaths of land for farming. And being budget-friendly and accommodating of space restrictions don’t really make the NanoMap an optimal choice for self-driving cars since they’re both big in size and budget.

But NanoMap will really shine when put to use for smaller drones that work on smaller or more consistent tasks, like daily shifts in the warehouse. Early testing in simulations and real-world scenarios have been promising, but more testing is needed before the technology can be rolled out into consumer drones. In the meantime, Florence and his Boston-based development team are working hard to improve the reaction time, speed, and flight of the drones.