Cyphorto
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Cyphorto is an autonomous drone pilot built with Meta Muse, taken from a simulator toward the real world. What it does: it flies the 60-gate Roadtrip race in the Uncrashed FPV simulator on its own. It watches only the game screen, finds each gate by vision (no memorised route, no waypoints, no GPS), and flies through it by driving a virtual Xbox controller. It detects crashes, stalls and lost gates, and recovers. Results: on Oct 24, before the hackathon, it got through 3 gates at about 6 km/h. During the hackathon it finished the full course for the first time on Oct 26 (9:29.874), then 8:27.545, 7:32.523, and 7:22.401 on Oct 28, all 60 gates, confirmed on the game's Results screen. How: every flight is logged. Muse reads the log, finds the root cause of a failure and changes exactly one thing, frozen as a new version. 30 frozen, hash-verified versions and 166 regression tests. Real world: Muse designed a 5 ft gate; I built it from PVC and flew my own drone through it by hand to capture footage. The simulator's colour vision failed on the real white gate (it locked onto house windows), so Muse built a new shape-based detector that finds the white gate in the drone's footage and outputs the same gate format the pilot already flies from. It runs on recorded footage today; running it live on board is the next step. Why: drones that can find their own way through gaps could one day deliver medicine to places roads can't reach. Until that's proven, a human stays in control of every real flight. Notes: most of the code was built with Muse; I also used other AI assistants for planning, checks and video editing. Video narration voiced with ElevenLabs.