MIT’s SANDO planner gives drones a time-sensitive route around moving obstacles

MIT SANDO, a new drone trajectory planner described by the Cambridge institution October 7, tackles a problem that a static map cannot solve: a clear flight path may stop being clear before the aircraft reaches it. The system builds a safety corridor that changes with time, allowing for moving obstacles in surroundings the drone has not mapped in advance.

MIT reports that the method avoided dynamic obstacles in 12 real test flights as well as in simulations. The researchers also provide a mathematical safety argument. That argument depends on the system’s model and assumptions, including a bound on how fast obstacles can move; it is not a promise that every drone carrying the software is incapable of crashing.

MIT SANDO plans for where an obstacle could go

The system’s name expands to Safe Autonomous Trajectory Planning for Dynamic Unknown Environments. Rather than assume an obstacle will follow one predicted path, the planner uses its maximum speed to calculate the region it could reach over a given period.

It represents that possible movement with a sphere and builds the drone’s safe corridor around the reachable area. As time passes and the aircraft gathers information, the corridor and trajectory are updated.

The MIT research account explains why the time dimension matters. A region known to be empty now can be occupied later. Planning only around an object’s current position leaves that future movement out of the problem.

SANDO also uses a heat-map approach to steer away from regions containing many obstacles. Within the resulting safe corridor, it optimizes a trajectory toward the goal. The researchers sought to make that computation fast enough for the drone’s onboard computer to revise its route as conditions change.

A proof and a flight test answer different questions

The work appears in IEEE Transactions on Robotics, according to MIT. Lead author Kota Kondo recently completed his MIT doctorate; senior author Jonathan How is an aeronautics and astronautics professor. The team also includes MIT graduate researchers and Jesús Tordesillas of Comillas Pontifical University in Madrid.

In simulations, MIT says SANDO reached its destination faster than several comparison systems while avoiding collisions. The 12 physical flights used onboard computing and sensors to replan around dynamic obstacles.

These results supply different kinds of evidence. A formal analysis addresses whether the planned trajectory satisfies specified conditions. Physical tests show that the implemented system worked in the situations actually tested. Neither makes the other unnecessary, and a limited flight series cannot establish performance in every environment.

Rescue missions remain an application, not the test report

MIT describes possible uses in collapsed buildings, mines, disaster response and delivery. These are proposed settings where unfamiliar surroundings and moving hazards can matter. The account does not establish an operating rescue service or report that SANDO has already flown through a real wildfire.

Further work could improve computational efficiency and combine the planner with machine-learning systems that accept ordinary-language instructions. Those are future directions, not features that should be assumed available today.

The near-term contribution is more specific and more useful than a claim of an uncrashable drone: a planning method that reasons about how empty space can change while a robot is moving through it. Whether it is ready for a particular high-stakes deployment requires evidence beyond the reported demonstrations.