Drone technology sure is promising plenty, but before the public can really warm to the idea of unmanned vehicles zipping around in all directions they want to feel pretty confident that they won't crash into things. Among the many computer scientists working on this problem is a team of researchers from MIT, who have developed route-planning software for drones that allows them to make intricate turns to autonomously navigate tight spaces.
Scientists at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) made some headway in this area last year, when they showed off a fixed-wing drone capable of zooming through trees at 30 mph (48 km/h). But while this and other crash-avoidance systems aim to guide an aircraft through a busy environment by steering it away from obstacles, the new solution instead guides them toward more favorable airspace, resulting in a drone better equipped to handle denser environments, albeit at slower speeds.
"Rather than plan paths based on the number of obstacles in the environment, it's much more manageable to look at the inverse: the segments of space that are 'free' for the drone to travel through," says Benoit Landry, lead author on a the new research paper. "Using free-space segments is a more 'glass-half-full' approach that works far better for drones in small, cluttered spaces."
To achieve this, Landry's team fitted a quadcopter with motion-capture optical sensors and an onboard inertial measurement unit to monitor the exact position of obstacles. They then devised an algorithm that detects free spaces within the environment, links them together and assigns a chain of flight maneuvers to culminate in a complete flight plan.
The technique was then demonstrated within a simulated forest, with the custom quadcopter darting in and around an obstacle course constructed from PVC pipe and strings. Measuring 3.5 in (8.9 cm) from rotor to rotor, the drone as capable of zipping through 10-square-foot gaps at speeds of up to one meter per second.
The technology offers an exciting glimpse of how drones may one day autonomously navigate everything from collapsed buildings to thick forests, but in its present form it is unable to plan its path in real time, requiring an average of ten minutes to chart its route prior to take off. But Landry says this preparation time can be reduced with a few modifications.
"For example, you could define 'free-space regions' more broadly as links between areas where two or more free-space regions overlap," he says. "That would let you solve for a general motion-plan through those links, and then fill in the details with specific paths inside of the chosen regions. Currently we solve both problems at the same time to lower energy consumption, but if we wanted to run plans faster that would be a good option."
Meanwhile, a separate project also carried out at CSAIL demonstrated a new approach to crash avoidance using a fixed-wing plane. While the obstacle course was a little less challenging, with only a single set of obstacles to fly through before hitting a safety net on the other side, the aircraft was able to chart its path in real-time, without any prior knowledge of the obstacles.
The researchers approached this by loading the plane with a set of 40 to 50 trajectories that it could fly along, which they describe as funnels. When the aircraft is fired from the launcher, it sifts through this preloaded catalogue in around 0.02 seconds to stitch together funnels that are free of obstacles and determine a safe route through.
Both drones can be seen in action in the videos below.