mirror of
https://github.com/VioletTides/KnockThru.git
synced 2026-10-29 04:03:02 +01:00
Robin Anderson
5f694c5d9b
A knock needs the controller to move first. While every tracked controller stays under 0.1 m/s, the pose thread reads every 10 ms. The first read that sees a controller faster than that switches back to 1 kHz, which holds for 0.5 s after the last movement. KNOCKTHRU_TEST_RATE now takes moving_speed, hold_s, still_ms, and idle_ms for sweeps. Every recording gives the same actions as the 1 kHz replay, calibration included. Margins from the sweep: - still_ms 20 and 30 give the same result. - moving_speed 0.2 with hold_s 0.5 gives the same result. - moving_speed 0.3 with hold_s 0.5 loses one headset double. - moving_speed 0.3 with hold_s 0.3 gives a wrong action in mixed. Reads in the recordings, which are knock sessions with little rest, drop from 414,710 to 378,881 (9%). straps and leg-doubles drop 20%. play and noise do not change. scripts/measure.sh 60 on the Frame in standby, twice: 0.23-0.28% CPU and 70 wakeups/s for the whole app, the same as before this change. The still rate only applies with a controller on, which needs the headset worn, so the Frame has no measurement of it yet.