Moves fast
Autonomous racing drones have completed timed races. KAIST also reports a crash during a faster attempt; race times do not establish a defense’s response time.
Last reviewed
What we are looking for
Is rapid physical movement demonstrated, and is its scope clear?
Where this applies
Swift, Crazyflie, MonoRace and KAIST · separate racing configurations
What has been observed
- Swift recorded the fastest race time in the reported contest.UZH / Nature · Aug 30, 2023
- A Crazyflie study reports 27 completed physical races.Sanchez Roncero et al. / arXiv · Sep 29, 2026
- MonoRace recorded the fastest completion time in the reported human-versus-AI race.Bahnam et al. / Science Robotics · Oct 14, 2026
- KAIST reports completing the multi-drone race in 37.2 seconds.Azhari et al. / KAIST / arXiv · Dec 23, 2025
What limits this finding
- The paper identifies changed appearance as a failure risk for Swift.UZH / Nature · Aug 30, 2023
- Swift was not trained to recover after a crash.UZH / Nature · Aug 30, 2023
- MonoRace lacked drone-to-drone collision avoidance and collided during the multi-drone race.Bahnam et al. / Science Robotics · Oct 14, 2026
- KAIST’s faster attempt crashed before completing the race.Azhari et al. / KAIST / arXiv · Dec 23, 2025
What we still need to know
Compatibility with the other properties in the same configuration.
What would change our view
Comparable timing and conditions on a jointly assessed platform.
What would be enough to confirm it
Timed physical flights or comparative races. A defense-specific response window requires a separate matched test.
How this finding has changed
Autonomous racing drones have completed timed races. KAIST also reports a crash during a faster attempt; race times do not establish a defense’s response time.
Updated assessment. Added a different team’s race result and limitations from the same A2RL event. Ratings and confidence remain unchanged.
MonoRace adds a timed competition to earlier physical racing results. These different courses do not establish how long a defense has to respond.
Checked again; no change. Reviewed; rating and confidence unchanged. The inspected material does not resolve the remaining gap.
MonoRace adds a timed competition to earlier physical racing results. These different courses do not establish how long a defense has to respond.
Updated assessment. New scoped evidence added; rating and confidence unchanged. Different platforms and limits remain separate.
Swift and Crazyflie studies report timed flights on separate laboratory courses. They do not establish how long a defense has to respond.
Checked again; no change. Reviewed; rating and confidence unchanged. The inspected material does not close the remaining gap.
Swift and Crazyflie studies report timed flights on separate laboratory courses. They do not establish how long a defense has to respond.
Updated assessment. Added a separate laboratory flight study; rating unchanged.
Swift demonstrated competitive speed on a set course. This does not tell us how much time a defense would have to respond.
First review. Initial assessment; no previous cycle for comparison.