University student competition · Skytag Open
One flies away.
One follows.
Can you catch it?
Systems is a university student competition for autonomous navigation algorithms. In the main game, one agent flies away while another follows and tries to catch it. Then they swap roles. Compete in checkpoint races too, using flight-controller SITL and MAVLink in the same simulation sandbox.
Research affiliation: Department of Computer Information Technologies (KIT), Kyiv Aviation Institute.
Student competition for: École Nationale de l’Aviation Civile (ENAC) students.
Several games. One simulation sandbox.
Skytag Open is the main event: tag, follow and catch. Checkpoint Race is a separate game for route planning and obstacle avoidance. Each game has its own objectives and standings.
Fly away
The runner chooses its own route through the arena and tries to stay uncaught for 90 seconds. It gets a 5-second head start before the scored clock begins.
Follow and catch
The chaser observes the runner, follows it and closes the gap. Hold within the 2 m virtual tag zone for one continuous second to catch it. Contact is a collision, not a tag.
Swap and replay
A catch ends the heat. Reset both controllers and the world, swap roles and spawn sides, then play again. Every algorithm must be able to run and chase.
Main game — Tag, Follow & Catch
chaser = 100 × (110 − catch_time) / 110 runner = 100 × elapsed_time / 90
- The chaser scores only on a valid catch; the runner earns more by remaining uncaught longer.
- No catch in 90 seconds: chaser 0, runner 100. Invalid episodes score 0.
- Free navigation: there are no mandatory checkpoints in this game. Following and catching the other agent is the objective.
- Rank by mean score across equal numbers of runner and chaser heats. Report catch rate, catch time and escape rate alongside points.
Another game — Checkpoint Race
score = 100 × valid × (0.80 C + 0.20 E)
- Two agents navigate five ordered checkpoints in a shared arena. Swap spawn sides for the return heat.
- C: fraction of checkpoints completed. E: reference route / traveled distance, capped at 1 and awarded only on completion.
- Each agent's route time and distance stop at completion; the heat ends when both finish or after 90 seconds. Earlier completion breaks equal-score ties.
- No tag objective in this game. Keep its results separate from the main game.
Both games use a 100 × 60 × 30 m virtual arena and equal vehicle capabilities within each flight-controller track. Collisions, arena exits and agent contract failures invalidate the affected episodes. Navigation, state estimation and reliable MAVLink control matter in every game.
Flight-controller SITL is part of the challenge
The algorithm chooses where to go. The simulated flight controller estimates state and controls the virtual vehicle. MAVLink commands, setpoints and telemetry connect the two.
A required interface, not an optional export
- Each scored agent must control its own SITL instance through MAVLink. Directly changing simulator positions bypasses the competition contract.
- Lifecycle commands use
COMMAND_LONGorCOMMAND_INT; inspectCOMMAND_ACKand verify actual controller state. - Navigation uses supported position or velocity messages such as
SET_POSITION_TARGET_LOCAL_NED, with explicit frames, units and type masks. - Streamed setpoints and command acknowledgements have different semantics. Heartbeat traffic alone does not satisfy PX4 Offboard control requirements.
What the benchmark should test
- Autonomous navigation with partial observations, noisy estimates and delayed telemetry.
- Replanning around static obstacles and other software agents.
- Controller tracking limits, mode transitions and recovery from stale observations.
- Equal compute budgets, isolated connections and replayable referee logs.
Guidance means route following here; targeting means choosing virtual game objectives. The emphasis is on navigation quality and reliable software behavior.
Interface references: PX4 simulation, ArduPilot SITL, MAVLink commands, PX4 Offboard, and MAVLink setpoints.
2025 season — confirmed standings
The original eight-team results for the main game, Tag, Follow & Catch. These standings have been confirmed by the competition organizer.
Archived season points are preserved as reported. The current normalized heat scores and the separate Checkpoint Race leaderboard use their own scoring rules.
| # | Team | Virtual vehicle class | Tags | Fastest tag | Survival | Season points |
|---|---|---|---|---|---|---|
| 1 | Vortex Hounds | Multirotor | 7 / 7 | 8.2 s | 528 s | 1130 |
| 2 | Paper Falcon | Fixed-wing | 6 / 7 | 11.6 s | 561 s | 1035 |
| 3 | Redline Swarm | Multirotor | 6 / 7 | 14.9 s | 512 s | 944 |
| 4 | Tinwing Collective | Fixed-wing | 5 / 7 | 13.4 s | 498 s | 873 |
| 5 | Magpie Works | Multirotor | 5 / 7 | 19.7 s | 471 s | 801 |
| 6 | Bramble FPV | Multirotor | 4 / 7 | 16.3 s | 455 s | 731 |
| 7 | Stall Warning | Fixed-wing | 3 / 7 | 22.8 s | 402 s | 591 |
| 8 | Groundhog Dynamics | Ground vehicle | 2 / 7 | 31.5 s | 366 s | 470 |
Game-specific results
Catch and escape rates for the main game; completion and efficiency for Checkpoint Race. Keep both leaderboards separate.
Timing and recovery
Observation age, controller state, command acceptance and recovery after telemetry becomes stale.
Paired experiments
Matched seeds, swapped roles, pinned stack versions and an archive of score components and traces.
Competition paper · 10 October 2026
Systems: A Sandbox for Autonomous Navigation Algorithms
Research affiliated with the Department of Computer Information Technologies (KIT), Kyiv Aviation Institute, for a university student competition serving École Nationale de l’Aviation Civile (ENAC) students. The paper covers the architecture, game rules, evaluation methodology and season results.
Inside the paper
Why the competition requires SITL. How MAVLink separates commands from navigation setpoints. Why chasing and escaping define the main game, while route completion defines Checkpoint Race. How paired matches, hidden maps and controlled observation delays can support meaningful comparisons.
The plots cover the shared architecture, confirmed season standings and an analytical timing budget. The archive also includes the main-game scoring curves. The archive includes the PDF, readable report, data, vector plots and reproducible builder.
Status: competition rules, architecture and evaluation methodology.
Prepare your student team entry
ENAC students: choose your game, describe your policy and select its SITL track. Download a local entry manifest to develop alongside your code. Online registration and execution are not connected on this site.