This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
If smaller extra room only adds closeness in gusty traffic, does early reaction (steering away sooner) with the full room keep the crash cut on the crossing at cruise speed?
What was tested
Multirotor and eVTOL flew all four combinations on the crossing at three speeds, and the storm and standard mixed test sets on five sets of layouts, comparing the old controller, the new default and the full-room candidate.
What was found then
Multirotor at 16 m/s: 31 crashes for the old controller, none for the new default, 3 for the candidate, keeping 90.3 % of the cut, just above the 90 % line required. Smaller room alone gave no crossing gain. In storms the candidate had 168 crashes against 190, with time close back at the old level. The eVTOL was similar, but its candidate crashed slightly more in storms (235 against 222) while spending 29 % less time close. Both met the restated rule (the adoption rule at cruise speed only); each was a trade, sent to the researcher.
What we learned later
Under a decision rule fixed in advance, the eVTOL kept the three-quarter room (the full room lost slightly more craft on both test sets); the multirotor switched at c182 after a check on fresh layouts.
research/data/oos/bubble_decision.md; research/design/tau_mr_lagk_record.md
What changed
nothing
Numbers and source references
- four
- combinations (neither / smaller bubble / charge / both)research/data/oos/mr_cx_2x2.md
- 8 / 16 / 22 m/s; three
- crossing speeds; crossing seedsresearch/data/oos/mr_cx_2x2.md; research/data/oos/evtol_2x2.md
- five
- benchmark seed offsetsresearch/data/oos/mr_c5_gate.md; research/data/oos/evtol_c5_gate.md
- 31 / 0 / 3
- multirotor crossing crashes at 16 m/s: old / adopted / full-bubble candidateresearch/data/oos/mr_cx_2x2.md
- 90.3 %
- share of the adopted crash cut the candidate keeps (0.9032)research/data/oos/mr_cx_2x2.md
- 90 %
- required retained share (screen)research/data/oos/mr_cx_2x2.md
- 168 / 190 / 330
- multirotor storm-set crashes, candidate / adopted / old, five seedsresearch/data/oos/mr_c5_gate.md
- 235 / 222 / 449
- eVTOL storm-set crashes, candidate / adopted / old, five seedsresearch/data/oos/evtol_c5_gate.md
- 29 %
- eVTOL candidate's lower time at depth vs adopted (0.145 vs 0.204)research/data/oos/evtol_c5_gate.md