Drones Can Now Spot Collisions Early — Without Wasting Computing Power
This could make delivery drones and drone light shows safer and cheaper to run.
Researchers have a smarter way for drones to avoid crashing into each other. In distributed model predictive control, a fixed prediction horizon forces a compromise: a short horizon is inexpensive but reacts late to approaching neighbors, while a long one anticipates conflicts but costs more per step, growing superlinearly with its length. The proposed conflict-predictive variable horizon is set locally by each drone, leaving the control itself unchanged. Each drone extrapolates its neighbors' flight lines, tests them against its own with confidence funnels, and computes each time to conflict in closed form. The horizon collapses to its minimum in clear airspace and grows only when a conflict lies ahead. In simulation on dense antipodal-swap benchmarks, it reduced per-step solver cost and total computation well below a long fixed horizon, and maintained separation in every run, which a short fixed horizon of comparable per-step cost did not.
- Each drone decides for itself how far ahead to plan, based on whether a neighbor is on a collision course.
- In dense simulations, it avoided every crash while using far less computing than the standard always-plan-far-ahead method.
- A short-sighted drone at the same computing cost failed to keep its distance — showing the 'look further when needed' trick is doing the real work.
Why It Matters
Safer, longer-flying drones could mean cheaper deliveries, quieter skies, and fewer mid-air accidents as drones multiply.