Experimental walk controller lab

Synthetic test geometry. No campus records or verified routes. Software-VM measurements are not universal performance claims. Native movement supports flat ground only; stairs and slopes are unsupported. Back to the campus walkthrough.

Loading…

WASD move · Q/E turn · drag to look · Escape clears input. The timed route applies the same prescribed velocity to each controller.

Measured result

Choose a controller and run a path. Frame intervals measure browser callbacks; CPU measurements are separate. No GPU timer is available on this VM.

Measured VM comparison and configuration

WIN11 VM / Edge / SwiftShader: paired six-second runs measured callback median 31.2-31.3 ms, p95 31.3 ms and p99 31.3-31.4 ms for all three. No sustained frame-rate gain. At 631 boxes (one draw / 7,572 triangles), mean simulation CPU per 120 Hz step: native 0.016 ms, Rapier 0.355 ms, Havok 0.366 ms. Synthetic results do not predict complex assets or hardware GPU performance.

Both capsule libraries passed thin-wall, corner, stairs and slope fixtures. Native supports flat ground only. Rapier contact tuning leaves about 3 mm flat-ground variation and 0.83% path shortfall here; Havok flat grounding is steadier. No jump. Callback and CPU timing do not measure GPU execution.

Shared metre-scale boxes, tiled floor, capsule radius 0.7 m / height 1.7 m, fixed 120 Hz simulation, 35 cm step cap, 45° slope cap, no jump. Native collision remains a flat horizontal section and has no stairs/slopes support. Libraries are tuned against their official character-controller examples.

Rapier controller guide · Babylon controller guide · Official Babylon example