Target: reflex gain is set by the drive, so will can suppress a reflex ahead of time at a visible cost (the Gom Jabbar case)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
parent
409b56c556
commit
9e0052ddfc
1 changed files with 12 additions and 6 deletions
|
|
@ -34,9 +34,10 @@ Signal flows one way per tick; each layer has one owner.
|
|||
1. **Perception.** Contacts, impulses, temperature, proximity and named game
|
||||
stimuli become typed stimuli on body regions. A poke that barely moves the
|
||||
body is still a stimulus.
|
||||
2. **Reflex.** Fast, local, involuntary responses to stimuli (withdrawal,
|
||||
startle, bracing), each with a latency, overlaid on whatever the body is
|
||||
doing. Physiology sets the scale: withdrawal roughly 60-150 ms, voluntary
|
||||
2. **Reflex.** Fast, local responses to stimuli (withdrawal, startle,
|
||||
bracing), each with a latency, overlaid on whatever the body is doing.
|
||||
Their gain is set from above by the drive, the way descending pathways
|
||||
modulate spinal reflexes: the Gom Jabbar case. Physiology sets the scale: withdrawal roughly 60-150 ms, voluntary
|
||||
reaction about 200 ms (figures `[M]` in the research doc; verify before
|
||||
they become constants).
|
||||
3. **Drive.** Per-character state that evolves every tick: pain accumulates
|
||||
|
|
@ -67,9 +68,14 @@ through the engine's runner (Unity Inference Engine, ONNX Runtime).
|
|||
directly; swapping the PD ragdoll for a learned tracker changes no layer above.
|
||||
- `one-physics-world`: the body simulates in the host's physics world. No
|
||||
second world that the game's objects cannot touch.
|
||||
- `reflex-below-voluntary`: a reflex acts on its own shorter latency and
|
||||
overlays the running behaviour; it never waits for behaviour selection, and
|
||||
behaviour selection never cancels a reflex already in flight.
|
||||
- `reflex-gain-from-drive`: a reflex fires on its own shorter latency and
|
||||
overlays the running behaviour; it never waits for behaviour selection. Will
|
||||
acts on reflexes ahead of time, not after: drive state (resolve, the pull of
|
||||
the objective) sets each reflex's gain, so a resolved character can hold its
|
||||
hand in the box. Suppression is never free: the stimulus still feeds pain,
|
||||
and holding a reflex down shows as visible strain until resolve breaks.
|
||||
Deliberate behaviour takes over a fired reflex only after its own, longer
|
||||
latency. (`motion-pipeline:ruling:lodur-reflexes-suppressible-by-will`.)
|
||||
- `drive-is-state`: what a hurt character does is decided by drive state
|
||||
evolving over time, never by a per-wound lookup.
|
||||
- `bounded-tick`: per-character work per tick is bounded and allocates nothing
|
||||
|
|
|
|||
Loading…
Reference in a new issue