Small unmanned aerial vehicles (UAVs) will soon navigate without using GPS or human assistance as demonstrated by several successful test runs last month as part of the Fast Lightweight Autonomy (FLA) program.
DARPA recently announced the breakthrough, after four days of testing in central Florida. The test runs marked progress toward development of small, quadcopter drones that are able to fly through obstacle-ridden environments without the guidance of a human operator or GPS navigation. Advanced software algorithms and sensors enabled the drones to navigate obstacle courses to find target objects, according to a DARPA press release.
The drones would be able to help troops remotely assess operational areas that are underground or in buildings where GPS can’t reach, or when the enemy is engaging in electronic signal jamming. It would either collect full motion video or still images that would be delivered back to the operators, according to a DARPA public affairs official.
“Small, low-cost unmanned aircraft rely heavily on tele-operators and GPS, not only for knowing the vehicle’s position precisely, but also for correcting errors in the estimated altitude and velocity of the air vehicle,” explained JC Ledé, Program Manager for FLA at DARPA. “In FLA, the aircraft has to figure out all of that out on its own with sufficient accuracy to avoid obstacles and complete its mission.”
Several different approaches and technologies for navigation in GPS-denied environments were demonstrated during the test period.
One approach, described by Dr. Camillo J. Taylor, Professor of Computer and Information Science at the University of Pennsylvania, uses Stereo Visual Inertial Odometry and Light Detection and Ranging (LIDAR) technology.
Stereo Visual Inertial Odometry requires two stereo camera sensors attached to the front of the quadcopter, according to Taylor. A stereo camera has multiple lenses and separate image sensors for each lens, which means that it can use triangulation algorithms to imitate the human eye’s ability to perceive distance and three-dimensionality.
“We prefer to use the stereo approach because it resolves the scale ambiguity,” said Taylor.
“Basically with two ‘eyes’ we can triangulate… and that helps us know how far metrically we’ve gone.”
Inertial measurement navigation can also make use of cold-atom interferometry technology, which uses software algorithms to monitor the acceleration and rotation of a compacted mass of atoms contained within a sensor, according to DARPA.
The quadcopter features a “nodding” LIDAR sensor, so called because it scans back and forth like a nodding head, explained Taylor. The LIDAR system uses a special, near-infrared laser that emits electromagnetic pulses and measures the return wavelengths to calculate the distance and 3D shape of objects in its path, according to a previous Defense Systems report.
Another approach explored at the testing site took inspiration from the way that people give each other directions.
“One of the big challenges we face when we can’t operate with GPS and don’t have any RC link to base station, is that we have a very limited knowledge prior to launch of what the environment will look like,” said Dr. Andrew Browning, Principal Investigator, Scientific Systems Company/AeroVironment.
To address that issue, another drone prototype technology pre-programs geographic or object cues into the drone’s software so that it knows to turn left when it senses a dumpster, for example. The other navigation relies on real-time sensors and decision-making, said Dr. Nicholas Roy, Principal Investigator at the Draper/ Massachusetts Institute of Technology. Eventually artificial intelligence learning technology may be incorporated, as well.
The FLA program that sponsored the autonomous quadcopter projects wants its drones to use lightweight, off-the-shelf technology and fly at speeds of up to 45 mph, according to DARPA’s press release.
“The goal of FLA is to develop advanced algorithms to allow unmanned air or ground vehicles to operate without the guidance of a human tele-operator, GPS, or any datalinks going to or coming from the vehicle,” said Ledé.
The challenge is to develop state-of-the-art software programs that offer a more complex solution than simply increasing the computing power of the technology, as computing hardware adds unwanted weight to the quadcopter. Ideally, the algorithms for the autonomous drone would run on a single computing board similar to that of a smart phone, said Ledé.
Now that this round of testing is complete, small autonomous drones that don’t rely on GPS will continue to be developed in Phase 2 of the program
- Jul 10, 2017