Apex GP

Devlog

A Rocket Car, a boost button, and one electric voice

A new car class took two days of measurement to drive properly — grip, brakes, a boost button under the thumb, and an electric bank that was playing two voices.

Sixty-one commits across the worktrees; most were one car.

The Rocket Car

A new class arrived: imported from a CC-BY model, de-branded, 50,933 triangles. Single-speed electric, 1 MW and 1,400 Nm, plus a compressed-air thruster worth 60,864 N for 1.2 seconds, once a lap.

The first measurements said it was not good. Lateral grip came out at 0.742/1.682/1.812 g at 40/80/160 km/h, against the modern car's 1.178/1.857/2.600. The fan load was there — 2,200 kg of suction — but only 6.91 kN of its 21.58 kN reached the steering axle, saturating the front tyres at 1.8 g; a neutral 10.79 kN per axle, and lock opened to 19.8 degrees at the road wheel, fixed it. The brakes were capped by the shared 24 kN hydraulic force, not the tyres: a per-class brake scale, defaulting to 1.0 so no other class moves, took 100-0 from 33.4 m to 26.6 m.

Two electric tones

I could hear two sounds: a bad one at the race start, a good one under acceleration. Both were real. The mixer blended an on loop and an off loop by throttle, and they were different syntheses — at 800 rpm the off loop sat 7.4 dB louder. The bank is a single layer now, one synthesis pitch-shifted across nine anchors, and a checker rejects a reintroduced second layer.

The boost button

The charge strip sat bottom-centre, nowhere near a thumb. It is a round button by the throttle pedal now: 80 px cap on a 90 px target, after a first attempt left 3 px under the assist caption. A tap also decayed in 160 ms under a held finger; a 0.6 s window keeps it pressed. The thruster plume had the same shape of bug — only the first 5% of each particle's life was drawn, so it vanished when emission stopped; it is 420 particles with a 1 s life now.

Multiplayer

A clock correction ignores errors under 10 ms now, so playout stops wobbling on transport noise, and heading converges on its own 0.10 s constant instead of the position's 71 ms, so a yaw correction cannot drag the body sideways. Extrapolating past the newest sample was tried and rejected: overshoot became a permanent 0.158 m offset.

Next: the Rocket Car in a full lobby, and the boost felt on a phone rather than in a shot harness.

← All devlog posts · RSS