A developer going by elodin-sys dropped a Show HN post Friday morning releasing AI Grand Prix Playground, an open-source practice simulator designed to help contestants prep for Anduril's autonomous drone-racing competition before the official tooling ships. The project runs on macOS and Linux with a stated ~5 minute setup time using uv, git, and git lfs.

Technical Architecture

The simulation stacks Elodin's deterministic 6-DOF physics engine (GPU-rendered, multi-rate sensors) against a real Betaflight SITL flight controller running in lockstep. The flight stack communicates over UDP using standard MAVLink-style RC and PWM commands—a choice that lets teams work with production-grade firmware rather than a simplified model. The FPV camera sensor matches the AI Grand Prix tech specs exactly: 640×360 resolution, focal lengths fx=fy=320, principal point cx=320, cy=180, plus a +20° up-tilt at 30 Hz refresh rate.

Competition Context

Anduril's AI Grand Prix dangles $500,000 in prize money for autonomous drone racers capable of navigating gate courses using perception, planning, and control algorithms. The official Virtual Qualifier 1 simulator hasn't shipped yet, leaving would-be competitors in a holding pattern. Elodin-sys built the practice rig specifically to fill that gap—letting teams iterate on their solver code today rather than waiting on Anduril's timeline. The project includes a 3-gate forward course in Elodin's ENU frame (+X/East) with automatic pass-time tracking for lap timing.

Developer Experience

Contestants drop their autopilot logic into the solver/ directory, implementing a single function contract: autopilot(update: SensorUpdate) -> RCCommand. A baseline solver ships out of the box so teams can verify the stack works before writing custom code. The simulation exports run data to numbered betaflight_db### directories containing full state logs; developers can dump component telemetry to flattened CSVs for regression testing or offline analysis, or render FPV footage to video for visual inspection. The test suite runs 36 tests across packet round-trips, course geometry, camera intrinsics, and the baseline solver—all executing in under a second without requiring the Elodin runtime.

Key Takeaways

  • Real Betaflight SITL firmware in lockstep with physics—not a simplified sim model—means flight controller tuning transfers directly to competition
  • Camera intrinsics match official AI Grand Prix specs exactly, so perception pipelines developed here should work on the real qualifier
  • Deterministic simulation enables reproducible bug reports and regression diffs across solver iterations
  • Apache 2.0 license means teams can ship derivative work without licensing friction

The Bottom Line

This is hacker culture doing what it does best: filling gaps that slow down interesting work. Elodin-sys didn't wait for Anduril to ship tooling—they shipped their own, and now the community has a head start on one of the more technically demanding autonomous racing competitions out there. If you're serious about competing, clone this repo today.