This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Early reaction (treating closing neighbours as nearer than predicted) fed three controller parts: the push away from neighbours, a last-second minimum-distance limit, and an acceleration budget shared between reaching the goal and keeping apart. Which part matters, and does stronger help?
What was tested
A switch limited early reaction to steering away (the push and the last-second limit). Both versions flew five strengths, 0.5 to 4, on the wing crossing: 81 runs, three layouts each.
What was found then
Up to strength 2, stronger mostly meant fewer losses. Beyond that, the all-parts version backfired at 22 m/s: 113 lost at strength 4, worse than the old controller's 95. Steering-only at strength 3 was the best yet, 9 lost at 22 m/s, and deep approaches (over halfway into the old bubble) fell from 2.38 to 0.5 per craft-hour. But long stays inside the old bubble lasted 1.5–3.7× longer. Caveat: three calm runs per speed, and strength 3 was picked on them.
What we learned later
Fresh layouts reproduced the strength-3 result (1 to 7 lost per run across the three speeds).
research/data/oos/tau_m2k3.md
Combined with removing the extra room, it failed the storm test and was not adopted.
research/data/oos/tau_adopt.md
What changed
nothing (a steering-only switch was added, off by default)
Numbers and source references
- 0.5, 1, 2, 3, 4
- charge strengths flown (multiples of the craft's response delay)research/data/oos/tau_k_hi.md
- 81
- new runs across cycles 127–129 (9 + 36 + 36)research/reviews/brief-c128.md; research/reviews/brief-c129.md:36; research/data/oos/tau_k_hi.md
- 26.7 / 29.3 / 21.3
- wing lost at 22 m/s, full charge, strength 0.5 / 1 / 2research/data/oos/tau_k.md
- 21 / 38 / 113
- wing lost at 22 m/s, full charge, strength 2 / 3 / 4research/data/oos/tau_k_hi.md
- 95
- today's controller, wing lost at 22 m/sresearch/data/oos/tau_k_hi.md
- 1 / 4 / 9
- wing lost per run, avoidance-only charge at strength 3 (1.3 / 4 / 8.7)research/data/oos/tau_k_hi.md
- 10 / 51 / 95
- today's controller at 8 / 16 / 22 m/sresearch/data/oos/tau_m2k3.md
- 0.07 / 0.16 / 0.5 vs 0.13 / 0.85 / 2.38
- deep approaches per craft-hour, avoidance-only strength 3 vs today'sresearch/data/oos/tau_m2k3.md (a)
- 1.5–3.7×
- averaged run-level 90th-percentile stay inside the old spacing vs today's (5.4 / 10.7 / 15.1 s vs 3.5 / 4.1 / 4.1 s)research/data/oos/tau_m2k3.md (a)