The contest

The competition asks one question. A policy is shown 1 of a task and none, then has to do that task. What the demonstration contains and where the policy is then put are the two facts that decide what a score means.

One-arm Franka, one demonstration

The policy is shown one demonstration of a task done with one arm of a dual Franka Panda - the frames, the action trajectory and the proprioception - and must do the task from the identical initial scene.

The policy is shown the frames, the action trajectory and the proprioception — everything the robot saw and everything it did — and is then scored from the identical scene. Its skills are Pick and place (one Franka arm), Stacking (one Franka arm), Press and push (one Franka arm), and a duel runs 3 per side.

Replaying the demonstration

Can the demonstration be replayed? yes — the actions are shown and the scene is the same; a replay policy winning is not a bug. The rollout starts in the demonstration's own scene with its actions visible, so a policy that plays those actions back can solve the unit. The contest says so rather than pretending otherwise: what it measures is how well a policy uses one demonstration of the scene it is in, and a replay is the floor of that, not a loophole.

Honest measurement sets out what that does and does not let a score tell you.

What is fixed

One demonstration, no language, both sides of a duel on an identical unit list with identical demonstration bytes, the crown rule at +3.0 pts, and every record published in full — per-unit outcomes, both sides, and the clips.

Entering

A submission is a Hugging Face repository at a commit sha, holding icil.yaml and the policy code it names, queued against the contest and run in a sandbox with no network — the same policy on every skill. See What you submit.