STEWART / LEARN
On this page

Watch, then try

These silent 12-second recordings use real controls and the fixed-step solver. Captions explain cause and effect. Playback is edited and is not a runtime performance benchmark. Play Ball Lab for the representative experiment.

1. A pose becomes lengths; different lengths become a pose

Watch: first a changed target drives six changing lengths. Next a different, valid six-length set is entered. FK reconstructs its ghost pose. Applying the result starts a second visible motion of the physical plate. FK does not teleport the mechanism or produce motor forces.

Try IK · Try FK

2. Same force, different stiffness

Watch: the same +16 N, −16 N, then zero-force sequence acts on two independently simulated platforms. K = 800 N/m predicts ±20 mm settled displacement; K = 200 N/m predicts ±80 mm. The video shows the measured transient values, not a fictitious enlarged response. Release is part of the experiment. Static F/K predictions do not describe every transient sample.

Try compliance

3. Motors or springs?

Watch: an impulse moves the gravity-compensated platform without a position-return command. A passive spring configuration responds differently. Stiffness and damping remain physical parameters; motor-off is not locked.

Try passive joints

Reproduce

Run python tools/record_demos.py, python tools/build_preview.py, then python build.py --site. The recorder writes the source-HTML hash and sampled states to artifacts/recordings.json. WebM and MP4 playback are provided by the generated site. The new hero is the ball scenario: balance, target movement, circle, push and reset. Written explanations remain available when video is unavailable.