Scientists at MIT have developed a new drone that can fly autonomously up to 30mph, while dodging obstacles like trees.
Sensors like lidar can detect objects, as is the case car mapping tech, but they're expensive and heavy – not ideal for drones. This new system is possible thanks to an algorithm that allows the drone to detect objects at up to 20 times faster than existing software.
The drone itself is built from off the shelf parts, features a camera on each wing and a processor similar to that found in a smartphone. The result costs about $1,700 to make, has a 34-inch wingspan, and can fly faster than with most human controllers.
The software operates at 120 frames per second but, crucially, doesn’t compute as much data as traditional software – which tops out at 5mph. This new system just analyses objects at 10 metres away, since very little actually changes between frames.
The result is far faster processing while maintaining accuracy. Since that 10 metre horizon is constantly moving nothing should be hit, presuming it doesn’t pop up suddenly within that distance. The drone does also fill in missing depth information with the drone's odometry and previous distance data.
The scientists are aiming to further improve the algorithm for more varying depths. So it should soon be able to work in a dense forest. This is limited by hardware advances. Here's hoping we can see these smart drones appearing soon.