Why it matters: Plenty of off-the-shelf drones can tail a cyclist down a trail these days, but most of them still lean on GPS to keep track of where they are. That's a problem in a forest, where a thick canopy can make the signal pretty unreliable. A lab at Purdue University has now shown off a drone that chases a person through the woods with no GPS at all.
Purdue computer science professor Aniket Bera posted the 44-second video on LinkedIn last Friday. It shows a quadcopter from his IDEAS Lab weaving between trees to keep a person in frame. And the drone works it all out on its own computer, without a map of the forest to rely on. Bera says skipping the cloud matters because a drone banking hard around a thin branch can't afford to wait on a server somewhere.
The person also keeps disappearing behind trunks, of course. For that, Bera's post lists an occlusion prediction step. It has the drone guess where its target is headed until they come back into view.
Navigating without GPS is also exactly the kind of autonomy the US military has been pouring money into. Last November, Purdue announced that the Army Research Laboratory had signed a five-year, $1.5 million deal with Bera's lab to build drones and ground robots that can find their way through hostile terrain without satellites. That said, the LinkedIn post doesn't say whether that money paid for the forest flights.
In the same post, Bera wrote that his team built every layer of the drone from scratch, from the 3D-printed carbon fiber frame to the software on the chip. As DroneXL noticed, though, the charts overlaid on the video label the drone's motion planner "SUPER" and its altitude estimate "FAST-LIO."
Those names belong to two open-source projects from the University of Hong Kong's MaRS Lab. SUPER is the lab's own LiDAR-guided drone. In a Science Robotics paper early last year, the team had it flying through unfamiliar terrain at more than 20 meters per second, or about 45 mph. A few weeks after that, the lab put the frame's design files online for free, including designs for 3D-printed parts and carbon fiber plates.
The Hong Kong lab doesn't get a mention in Bera's post, even though SUPER's code page asks anyone who builds on it to cite the team's papers. There's no paper covering the forest demo on the IDEAS Lab's publications page yet, either.
For now, anyone who wants to try building a similar drone can grab HKU's SUPER code as a free download. The lab's hardware kit lists every part needed, too, from the Livox Mid-360 LiDAR on top to the Intel NUC mini PC that runs the software.