Three angled boards turned to flat outlines by a bimanual robot

Three tilted boards, each turned and set exactly inside its outline

Role

The controller of a simulated bimanual robot in a workshop of wooden boards, acting only through a local HTTP command service, with nobody giving movement advice.

Request

Pick up the three angled boards, turn each to its target orientation and let go so every corner rests inside its outline; once they settle, confirm success, stop the engine and save a short summary.

Deliverable

The final physical state, from the engine's last observation: three boards lifted, turned, released and settled. A short summary file is saved too; its claims don't count.

What the reward depends on

All eight corners of every board inside an outline of its own size once settled, after a real lift, turn and release. One point or none: coming close earns nothing.

Where the requirements are

Stated in the request and one interface document, with the runtime's source open to read; each observation reports the targets and progress. Nothing is hidden: the difficulty is physical control.

A contact‑only physics workshop, read as exact state and driven by typed commands

Interface

A Codex chat subagent running GPT‑6.1 Sol at its default reasoning setting, sending JSON to a local HTTP control server through its shell tools: read the state, send a step or a sequence of velocity segments, and post short progress notes.

Physics

Rigid‑body physics (cannon‑es) in its own headless Edge process at 180 Hz, apart from any viewer. Simulated time advances only during commands and pauses while the agent plans.

The robot

Two seven‑joint manipulators on bases that slide across the floor, each with a palm and five fingers of three joints, with spread and thumb opposition. Motion is rate‑limited and stopped by collisions, so a target isn't always reached.

Grip

No attachment constraint: a hold is opposed, force‑bearing finger contact, so a board can slip, turn or drop while the hand stays closed, and over‑curling can push it out.

Observations

Exact structured state, not images: board poses, sizes and velocities; hand and finger poses, contact force and slip; collision and joint‑limit feedback; each target's pose and progress.

The build

Three boards in angled stock cradles, each to be lifted clear, turned about up to three axes and laid flat in its own outline. No tools or screws.

Recording

Every command result and sampled pose is saved; four fixed cameras render review videos afterwards, at simulation speed. The saved observations, not the videos, are the record.

Network

Internet on, for the pinned physics libraries; web tools off by instruction, not by a network sandbox. Nothing the job needs is online.

Limits

One session of 20 minutes of wall time, planning included, plus the scenario's own action and simulation‑time budgets. The request rules out resets, teleporting, direct state changes and earlier trajectories.

One all‑or‑nothing check on the final physical state

Structure

One check, worth one point: the whole build complete. Boards placed and each board's worst corner protrusion are reported as diagnostics, never scored.

Placement

Each board must be lifted, released and left to settle with all eight corners inside a target outline of its own size, allowing only a numerical margin. Holding a board in place doesn't count.

Who decides

A script reads the final observation and transforms every board's eight corners into its target's frame; it also needs the engine's own success flag for lift, release, settling and any joints. No model judge, and the agent's claims don't count.

A stricter check

This attempt ran under an earlier engine check that accepted boards centimetres off, as its documentation described, and its request lacked the corner rule. It was regraded on the strict check, with no new attempt.

Engine reward

After each action the engine also returns a shaped reward from the boards' distances to their targets, with a bonus on success and a penalty for a lost board; the grader ignores it.

Formula

Reward = 1 if the build is complete, else 0. An untouched scene, a partial build or a missing final state scores 0.

All three boards came to rest outside their outlines

GPT‑6.1 Sol, the only model tested on the builds, scored 0%. Its three boards came to rest 16.2, 22.7 and 43.6 mm outside their outlines at the worst corner; it stopped after 7.6 minutes, when the engine's earlier check reported success.

Each dot is one attempt; the line marks the median.

One attempt, by GPT‑6.1 Sol at seed 1; no other model has been tested on the ten builds.

Made on 29 October 2026, before the corner rule joined the request, under the engine's earlier check, which its documentation described; regraded the same day on the strict one, with no new attempt.

Where the attempt fell short

One attempt, by GPT‑6.1 Sol, made under the engine's earlier check and graded on the strict one.

1. Turned, not inside. All three boards came to rest outside their outlines, by 16.2, 22.7 and 43.6 mm at the worst corner.

2. Stopped on the old check. The engine's earlier check, described in its documentation, accepted boards centimetres off; when it reported success, the episode ended, as the request expected.

3. Most of the time unused. It stopped after 7.6 of its 20 minutes, having sent 17 motion commands covering 85 simulated seconds.

What an attempt costs

Length

The attempt took 7.6 minutes of wall time, with 17 motion commands and 85 simulated seconds. Model calls, tool calls and tokens weren't recorded; a motion command isn't a model call.

Cost

Not recorded: the attempt ran on a subscription sign‑in, which logs no cost or tokens, so we give no figure.

What you license

With a licence

Revision 2: the frozen runtime with this scenario at seed 1, the request and interface document, the grader and its self‑test, and the attempt's commands, observations, pose trajectory and four review videos, as baseline results.

Format

Runs on Node.js, with the physics in headless Microsoft Edge driven by Playwright, and a Python grader; today on a Windows PC, outside a container. Other packaging on request.

Fresh variants

Other builds and layouts in the same design, on request. They're never published.

Public or not

The AI Agent Trials, our public benchmark, sums up the job and shows each model's score, so hold out fresh variants for evaluation. The files, grader and answer key aren't published.

What we haven't measured, and what's uneven

• One attempt, by one model, at one seed, so we haven't measured how much attempts vary or how other models do.
• This attempt's request lacked the corner rule and its documentation described the earlier, looser check, so it can't show how a model does when told the strict rule.
• No reference solution: no attempt, scripted or by an agent, has completed a build under the strict check, so it isn't shown that one can be; the self‑tests check only the scoring.
• Each outline is its board's own size, so a board must settle almost exactly at its target pose; whether the contact solver allows that hasn't been tested.
• The grader trusts the engine's own flag for lift, release, settling and joints, and checks only the geometry; the lift flag can register for a board grasped in raised stock without upward travel.
• The agent reads exact state and may read the runtime's source, which reveals the thresholds: this tests control from state, not from vision.
• Simplified physics: rigid contacts, simplified fingers and idealised stock fixtures, so results don't speak to real robots.
• Up to three attempts shared one PC at a time, so wall time may include contention.