MIT's SANDO planner guarantees collision-free drone paths as long as obstacle speeds have a known ceiling

A safety case for a drone in a crowded street or a collapsed building eventually has to say what the aircraft does when something moves that was not on the plan. Planners that dodge moving obstacles often do it without a formal guarantee of staying clear, according to Interesting Engineering's account of the MIT work ^[12]. SANDO's answer is to give up on predicting exactly where each object will be ^[5].

Here is what teams will tell themselves: a planner with a proof attached ^[2] retires the custom safety review before a delivery or inspection job. Here is what the operator actually does with it on Monday. They pick a top speed for everything that might move near the drone and sign off on it ^[4]. The account does not mention any regulator or certification body reviewing SANDO's proof. I'd expect the site-by-site safety work to change subject. The argument moves away from how the planner behaves and toward two other questions: whether the ceiling fits the site, and whether the onboard cameras and sensors that detect, group and track obstacles ^[6] see everything that moves.

The use cases on offer are wildfire response, search and rescue in collapsed buildings, mine exploration and package delivery in crowded areas ^[15]. The hardware evidence is 12 flights on a real drone with dynamic obstacles, replanning on its own computer and sensors ^[10]. Fei Gao, an associate professor at Zhejiang University who was not involved in the research, said in remarks reported by MIT News: "Its combination of spatiotemporal planning, formal safety analysis, and hardware validation provides a practical approach to autonomous flight in complex dynamic environments" ^[11]. The researchers list cutting the system's computing requirements and pairing it with machine-learning models as possible next steps ^[16].

For a team weighing a pilot, I'd sort candidate sites on two axes. The first is whether a speed ceiling for everything that moves there can be written down and defended to whoever signs off. The second is density: how much of the airspace the worst-case spheres and heat-map detours ^[7] would take up. My recommendation is to start where the ceiling is defensible and the space is open. The tradeoff is that a more cautious ceiling buys margin by leaving the drone less room to fly ^[13]. A defensible ceiling at a dense site keeps the guarantee but pays for it in detours, so the pilot should log time to destination on every trip next to the collision count. Without a defensible ceiling, in open space or dense, the proof rests on a guess, because anything faster than the cap falls outside it ^[14]. In that box SANDO is an onboard planner running on top of the safety case you already have.