This Autonomous Drone Can Track Humans Through Dense Forests at High Speed

Autonomous drones could revolutionize a wide range of industries. Now, scientists have designed a drone that can weave through dense forests, dodge thin power lines in dim lighting, and even track a jogging human.

Rapid improvements in sensor technology and artificial intelligence are making it increasingly feasible for drones to fly themselves. But autonomous drones remain far from foolproof, which has restricted their use to low-risk situations such as delivering food in well-organized cities.

If the technology is ever to have an impact in domains like search and rescue, sports, or even warfare, small drones need to become both more maneuverable and more reliable. That prompted researchers from the University of Hong Kong to develop a new micro air vehicle, or MAV, that can navigate challenging environments at speed.

The new drone, named SUPER, combines lidar technology with a unique two-trajectory navigation system to balance safety and speed. In real-world tests, it outperformed commercial drones in both tracking and collision avoidance, while flying at more than 20 meters per second (45 miles per hour).

“SUPER represents a milestone in transitioning high-speed autonomous navigation from laboratory settings to real-world applications,” the researchers wrote in a paper in Science Robotics introducing the new drone.

According to the authors, the inspiration for the project came from birds’ ability to nimbly navigate cluttered forest environments. To replicate this capability, they first designed a drone just 11 inches across with a thrust-to-weight ratio of more than five, which allowed it to carry out aggressive high-speed maneuvers.

They then fitted it with a lightweight lidar device capable of detecting obstacles at up to 70 meters. Given they were targeting high-speed flight, the researchers say they were keen to avoid the kind of motion blur that camera-based systems suffer from.

Most important though, was the navigation system they designed for the drone. At each route-planning cycle, SUPER’s flight controller generates two flight trajectories towards its goal. The first is designed to be a high-speed route and assumes that some of the areas ahead with limited lidar data are free of obstacles. The second is a back-up trajectory that focuses on safety, only passing through areas known to be free of obstacles.

The drone starts by following the high-speed trajectory but switches to the backup if the real-time lidar data detects anything in the way. To test out the approach, the researchers pitted it against two other research drones and a commercial drone in a series of trials, which involved flying at high speed, dodging thin electrical wires, navigating a dense forest, and flying at night.

The SUPER drone achieved a nearly perfect success rate of 99.63 percent across all the trials, which is nearly 36 times better than the best alternative the researchers tested. This was all while achieving faster flight speeds and significantly reduced planning times.

The drone also demonstrated excellent object tracking, successfully tailing someone jogging through dense forest. In contrast, the commercial drone, which used vision-based sensors, ultimately lost track of the target.

The researchers suggest that the development of smaller, lighter lidar systems and aerodynamic optimizations could enable even higher speeds. Imbuing SUPER with the ability to detect moving objects and predict their motion could also improve its ability to operate in highly dynamic environments.

Given its already impressive performance though, it seems like it won’t be long before fast, agile drones are buzzing over our heads in all kinds of places.