MIT’s SANDO Keeps Drones on Safe Paths by Planning Around Moving Obstacles in Unknown Spaces

Drones flying search and rescue missions or package runs through crowded neighborhoods regularly encounter objects that move without any advance notice. A new planning system from MIT called SANDO addresses that problem by generating flight paths that stay collision free even when the area has no prior map and obstacles can shift position at any moment.

The concept relies on using the drone’s built-in cameras and sensors to locate and group moving things, then wraps each one in what’s effectively a 3D bubble that stretches out according to the fastest speed the object may possible attain. As you expand the planning window, the bubbles grow, indicating where every possible obstacle could end up before the drone arrives. SANDO then stitches together a trail of connected 3D regions which produces a kind of safety tunnel the drone may fly through without colliding into any of those bubbles.

Inside the tunnel, a heat map indicates the places with the most obstacles and directs the drone to the less congested pathways rather than attempting to force it through the densely packed clusters. Once the tunnel is built up, the system calculates the shortest possible route that will fit inside it, all while keeping things streamlined to keep the computer from becoming bogged down. This entire process occurs on the drone’s own hardware, so you receive updates quickly enough to respond when a new impediment appears. As it travels forward, SANDO continuously refreshes both the corridor and the path, adjusting for new barriers that appear or items that deviate from the initial plan. All that’s required from the outside is a suitable upper speed restriction, while the rest is provided by the drone’s onboard gear.

The system was tested through 12 flights in a variety of situations, including several moving objects, and it avoided every single one of them using its own onboard processors and sensors. When compared to other planners in simulations, it achieved its targets faster while recording zero constraint breaches throughout static woods, obstacle-heavy environments, and fully dynamic landscapes. According to lead researcher Kota Kondo, the strategy is safe as long as you know the max possible speed of those obstacles. The team, including senior author Jonathan How, Jesús Tordesillas, Juan Rached, Lili Sun, and Yixuan Jia, contributed to the research published in IEEE Transactions on Robotics.

Real-world uses include checking buildings after an earthquake, mapping mine tunnels, and capturing data above a wildfire as it burns and debris moves about. Package delivery via pedestrian and traffic-heavy regions also fits the description, and who knows, perhaps future versions will be able to learn and follow plain language directions about preferred routes. The current lot demonstrates that you can maintain a drone flying securely in locations that are changing quicker than any static map can keep up with, as long as the planner considers the full reach of every impediment it detects.
[Source]