3D sandbox / 28 September 2026

Drive it.

V1-PROOF at full scale on the physical test fixtures. The physics and balance controller are the same code as the saved study, not a look-alike.

Loading three.js and Rapier…

Start lane

Camera

Time

Leg angle

Drive speed

Shove

Dimensions & CoM Size references

Drive W A S D or arrows · Shift for the drive-speed slider's top (0.5 m/s is unqualified) · Legs Q lower / E raise, or 1 tall, 2 pinned 30°, 3 low · Shove F forward, B back, Z left, X right · R reset · K kill · C camera · P pause · Mouse wheel zoom. A gamepad works: left stick drives, right stick moves the legs, A shoves, B resets.

Pitch from trim, last 10 s

Motor current, last 10 s

Speed vs command, last 10 s

Study parameters

Break it on purpose.

Same parameter names and sweep ranges as tools/v1-proof/run_sim.js. The controller keeps its nominal gains, as in the study. Live knobs apply at once; rebuild knobs reset the robot at the current lane.

What this is and is not

Same plant and controller. The page loads tools/v1-proof/sim.js itself: Rapier rigid bodies at 1 kHz, encoder-quantized odometry, delayed noisy attitude, the 500 Hz discrete LQR, voltage- and speed-limited motor torque, drive lag, deadband and optional backlash.
Fixtures. The lanes are the physical protocol's fixtures: 3° grade, 3° cross-slope, 5 mm bump (both wheels and one wheel), 3 mm seam, plus the 20 mm threshold the study records as a failure. Fixture ramps have kinks and transitions; the study tilted the whole floor.
Leg height is a preview. Pinned 15°, 30° and 45° poses were in the study and pass. Moving between them re-places the CoM and pitch trim at 10°/s; leg inertia, servo torque and its reaction on the body are not modelled. Powered height motion is unvalidated.
Size. Body, links, wheels and motors are drawn from model.json envelopes: 120 × 110 × 150 mm body, 110 mm links, 100 mm wheels at 230 mm track. Packaging, belts and wiring are not detailed. The dashed box is the parked STAIR-V1 envelope for scale.
Not modelled. Raw IMU fusion, firmware timing, thermal limits, tire compliance, regeneration, servo dynamics. A good run here is a candidate for a bench test, not evidence the robot works.

Headless check: node tools/v1-proof/test_sandbox.js. Saved evidence: Simulations.