You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
Deadlink is the software layer that makes a fleet of autonomous drones work as one coordinated swarm. Today one person flies one drone by hand. Deadlink lets one operator launch a whole swarm that shares a single mission, divides the work between the drones, and acts together to get it done, then keeps going when one drone drops mid-flight and produces a complete replayable record of everything they did. It's drone-agnostic and runs above any flight stack. Drones are phase one: the same layer runs every autonomous fleet that comes after, from delivery and inspection to defense and search-and-rescue. The core works in simulation now; this grant puts it on real hardware. I'm proving it first in search-and-rescue, because it's the hardest test there is and a silent failure costs lives.
The goal is to prove a fleet of drones can run a real mission as one self-coordinating swarm, on physical hardware, and still finish when one of them fails. In simulation this already works: three drones share one search mission, split the area between them, and fly it together; I kill one mid-flight and the swarm detects the dead link, takes back its authority, reassigns its unfinished area, and the rest complete the entire mission, with every decision written to a replayable log. Next I'll port this onto Crazyflie drones, fly the same coordinated-swarm-with-a-failure scenario in the air, and publish an honest demo plus the log. The milestone: a real swarm flying one mission together and finishing it even after a drone is dropped mid-flight.
Entirely on getting Deadlink from simulation onto hardware. The minimum ($1,500) buys a small fleet of Crazyflie drones plus the radios and positioning to fly a coordinated swarm and record the demo. The full goal ($15,000) adds a first real-world pilot: more drones and flight hours, a larger test environment, and validating the coordination and recovery against a realistic search mission.
Me, solo. I'm Saketh, 17, building Deadlink full-time in SF (The Residency + the Founders Inc cohort). Self-taught: first-author ML paper at 16 with no lab or advisor, USACO Platinum, BlueDot AI-safety fellow, and a 1517 Fund Medici grant for an earlier version of this work. I wrote and tested the entire Deadlink runtime myself: mission planning and validation, deterministic multi-drone execution, task assignment, fault injection, dead-link detection, authority revocation, stale-input rejection, event logging, and the full test suite. I pressure-tested the problem with ~120 search-and-rescue and public-safety teams.
The most likely failure is that the hardware port surfaces real-world problems simulation hides: radio dropouts, positioning noise, or timing that makes the swarm coordinate less cleanly in the air than in sim. If that happens it's still net-positive: I learn exactly where the model meets reality, publish it honestly, and harden the core, because the whole point of Deadlink is software that fails loudly and shows its work instead of failing silently. The grant is small and the worst case is a slower, better-understood path to the same milestone, not a dead end. No safety risk to people, money, or third parties: these are small indoor research drones.
A 1517 Fund Medici grant ($1,000, no strings) for an earlier version of this work. No other capital. I've built Deadlink with effectively no money so far.