Skip to content

Lab Challenges and Duels

inksong edited this page Nov 8, 2026 · 1 revision

Lab, Challenges and Duels

The Lab workspace (⌘4 / Ctrl+4) is where programs are written, tested and judged, like a programming exercise site for flight. Programs and examples are on the left with the challenges; the editor is in the middle (every save is a revision with its author and a note; problems are marked inline); Params, Tests, Run, Judge, Duel and History are on the right.

Unit tests

Tests feed fixed observations and check the sticks, like the sample cases of an exercise. They run in the program host, so they see exactly what a deployment sees.

Challenges

Ten deterministic scenarios run in the headless evaluator, faster than real time, on visible and hidden seeds whose airframes differ, so a program tuned to one airframe does not pass. Scores run from 0 to 100.

id Scenario Pass condition
liftoff Payload puts the hover throttle near 0.62 Off the ground within 5 s, 1.0 ± 0.2 m held for 3 s
hover Forward centre of gravity and level error 20 s with drift ≤ 0.8 m and height 1.0 ± 0.25 m
heading Unequal motors make the yaw drift Heading error ≤ 10° for 15 s
square UWB-style position at 10 Hz Four corners in order; faster scores higher
rings The ring course with solid rings Every ring, timed; touches are counted
pursuit A target on a scripted path Share of 60 s within 2 m
duel_easy, duel_medium, duel_hard A built-in hunter Shoot it down and survive
ram_easy A hovering target, guns off Crash it by collision within 60 s

Duels and the Elo ladder

The Duel tab takes 2 to 8 programs, a number of rounds and a seed. The result moves an Elo ladder (rank, Elo, change, wins, losses, draws) and lists every round with a replay. The same scenario, seed and revisions always give the same result.

Replays

Every judged seed and duel round leaves a 10 Hz trace. Opening it switches to the Fly stage, which replays every aircraft, every round in flight, every hit, collision and explosion, with play, pause, 1x, 4x and 16x and a scrubber. Effects are a pure function of the replay time, so scrubbing back shows them again.

Agents in the Lab

Everything in the Lab is available to agents through the API, the SDK, the application MCP server and Copilot: program_create, program_update, program_validate, program_test, challenge_submit, job_status, duel_run, ladder and the rest. Copilot needs no approval to write programs; deploying one on a real aircraft waits for the operator.

Clone this wiki locally