TOPO_Gun with GF targeting #84

New Issue

2026-08-25T19:11:53+02:00

SirStone commented

2026-08-25 19:11:53 +02:00

Parent

What to build

Wire ANN, GA, sliding window, replay tape, and normalization into a working gun module that replaces head-on firing with guess-factor targeting.

Gun interface: feed(state) / aim() → (angle, power).

Power formula: power = clamp(3.0 - (distance - 150) / 400, 0.5, 3.0) — applied first.
Aiming: bullet_speed = 20 - 3 × power, MEA = arcsin(8.0 / bullet_speed), aim_angle = bearing_to_enemy + gf × MEA.

GA runs on a parallel thread, evolving against the replay tape. Champion is pushed to inference when it beats current. Evolution never blocks the tick loop.

First adaptive shot: bot fires aimed bullets at OscillatorBot using ANN-predicted guess factors.

Acceptance criteria

Gun module exposes feed(state) / aim() → (angle, power) interface
Power formula produces values clamped to [0.5, 3.0] based on distance
ANN predicts GF, converted to aim angle via MEA geometry
GA evolves on a parallel thread without blocking the tick loop
Fitness evaluated as virtual bullet hits on sampled replay tape ticks
Champion promoted to inference when it beats current champion
Bot fires adaptively at OscillatorBot — visibly better than head-on over time
Power formula self-check: when isMainModule with doAssert validates distance→power mapping

Blocked by

#82
#83

## Parent https://git.fossellini.top/SirStone/SirRoboGarage/issues/80 ## What to build Wire ANN, GA, sliding window, replay tape, and normalization into a working gun module that replaces head-on firing with guess-factor targeting. Gun interface: `feed(state)` / `aim() → (angle, power)`. Power formula: `power = clamp(3.0 - (distance - 150) / 400, 0.5, 3.0)` — applied first. Aiming: `bullet_speed = 20 - 3 × power`, `MEA = arcsin(8.0 / bullet_speed)`, `aim_angle = bearing_to_enemy + gf × MEA`. GA runs on a parallel thread, evolving against the replay tape. Champion is pushed to inference when it beats current. Evolution never blocks the tick loop. First adaptive shot: bot fires aimed bullets at OscillatorBot using ANN-predicted guess factors. ## Acceptance criteria - [ ] Gun module exposes `feed(state)` / `aim() → (angle, power)` interface - [ ] Power formula produces values clamped to [0.5, 3.0] based on distance - [ ] ANN predicts GF, converted to aim angle via MEA geometry - [ ] GA evolves on a parallel thread without blocking the tick loop - [ ] Fitness evaluated as virtual bullet hits on sampled replay tape ticks - [ ] Champion promoted to inference when it beats current champion - [ ] Bot fires adaptively at OscillatorBot — visibly better than head-on over time - [ ] Power formula self-check: `when isMainModule` with `doAssert` validates distance→power mapping ## Blocked by - https://git.fossellini.top/SirStone/SirRoboGarage/issues/82 - https://git.fossellini.top/SirStone/SirRoboGarage/issues/83

SirStone added the ready-for-agent label 2026-08-25 19:11:53 +02:00

SirStone referenced this issue

2026-08-25 19:12:04 +02:00

Weight persistence and hybrid seeding #85

SirStone referenced this issue

2026-08-25 19:12:11 +02:00

Virtual Guns system #86

SirStone closed this issue

2026-08-25 19:49:01 +02:00

Sign in to join this conversation.