STRAFE movement: body pinned perpendicular to the threat, reversals by sign flip

New engine movements/strafe.nim, selected by TR_MOVEMENT=strafe (default stays
tfil, byte-identical — test_tfil_commit_env.nim's 30 checks still pass).

Design (the owner's):
- AXIS = incoming bullet's direction when a bullet is in flight, else the
  perpendicular of the enemy bearing. The body heading is kept inside a band
  (TR_STRAFE_BAND, default 20 deg) around the perpendicular LINE; it turns only
  when outside the band, and never turns to face a movement target.
- Candidate tiles on the perpendicular line through our position, both forward
  and backward, within TR_STRAFE_REACH px, with a perpendicular jitter of
  +/- TR_STRAFE_SPREAD tiles. A tile is acceptable when its path max heat is
  <= PathDangerThreshold, the SAME safety rule TFIL uses.
- Move by SIGN only: setForward(+/-MaxSpeed>). Dwell is re-picked after a random
  number of ticks in [TR_STRAFE_DWELL_MIN, TR_STRAFE_DWELL_MAX], on arrival, or
  on a serious threat spike.
- Heat machinery is REUSED from the shipped mover, not re-implemented: the
  exported heatDecay()/bulletMagScale() (j105 time-indexed model) and the
  PillarHotness/PillarRadiance globals (j106 pillar-free default). The heat
  shape is overridable via TR_STRAFE_CORRIDOR_HEAT/WALL_HOTNESS/WALL_RADIANCE
  (defaults = the shipped TFIL field).
- GUI overlay: strafe line, threat axis, candidate tiles (safe/unsafe), chosen
  target, sign-coloured movement ray, and the heading band.

Gates (offline, recorded DrussGT fixture, 20026 ticks):
- A TILE AVAILABILITY: shipped heat field -> a safe tile exists on only 36.6%
  of picks (63.4% fall back to the least-hot tile); the ring retune
  (corridor 5, wall 10/5) raises it to 91.9%.
- B PREDICTABILITY: reversal-interval entropy 5.84 bits vs TFIL 5.09; direction
  entropy 1.00 both; long-lag autocorrelation ~0 for both (no periodic
  component). Fewer reversals (710 vs 1453) and more full-speed ticks.
  measurements: common_libs/tests/measure_strafe_gates.nim

Also registers TR_STRAFE_* in the boot env report (ModularBot_garage/src/
env_report.nim) and wires the engine into ModularBot.nim (hold -> strafe,
ram trigger -> rammer).
This commit is contained in:
2026-09-25 22:38:52 +02:00
parent 99cf9e5c82
commit a50c0125d5
4 changed files with 1103 additions and 9 deletions
+17 -6
View File
@@ -33,6 +33,7 @@ import movements/rammer
import movements/ram_decision import movements/ram_decision
import movements/the_floor_is_lava import movements/the_floor_is_lava
import movements/the_floor_is_lava_ring import movements/the_floor_is_lava_ring
import movements/strafe
import movement_harness/virtual_bodies as mvb import movement_harness/virtual_bodies as mvb
import movement_harness/bullet_shadows import movement_harness/bullet_shadows
import targeting/enemy_tracker import targeting/enemy_tracker
@@ -193,6 +194,7 @@ type
bitbrain: BitBrainGun bitbrain: BitBrainGun
mover: TFILModule mover: TFILModule
ringMover: TFILRingModule ringMover: TFILRingModule
strafeMover: StrafeModule
rammer: RammerModule rammer: RammerModule
isRamming: bool isRamming: bool
ramDurationTicks: int ramDurationTicks: int
@@ -317,7 +319,9 @@ proc printConfig(bot: ModularBot, forceAll: bool = false) =
let tc = if bot.currentTargetId != bot.prevTarget or forceAll: CLR_CHANGE else: "" let tc = if bot.currentTargetId != bot.prevTarget or forceAll: CLR_CHANGE else: ""
let radarName = if bot.radarMode == 0: "radar_lock" else: "adaptive_melee" let radarName = if bot.radarMode == 0: "radar_lock" else: "adaptive_melee"
let moveName = let moveName =
if MovementName == "tfil_ring": if MovementName == "strafe":
if bot.isRamming: "strafe(ram)" else: "strafe"
elif MovementName == "tfil_ring":
if bot.isRamming: "tfil_ring(ram)" else: "tfil_ring" if bot.isRamming: "tfil_ring(ram)" else: "tfil_ring"
elif bot.isRamming: "rammer" elif bot.isRamming: "rammer"
else: "tfil" else: "tfil"
@@ -567,6 +571,7 @@ method onBulletHitWall*(bot: ModularBot, e: BulletHitWallEvent) =
bot.moveTracker.shadows.removeBulletNear(e.bullet.x, e.bullet.y) bot.moveTracker.shadows.removeBulletNear(e.bullet.x, e.bullet.y)
bot.mover.removeBulletNear(e.bullet.x, e.bullet.y) bot.mover.removeBulletNear(e.bullet.x, e.bullet.y)
bot.ringMover.removeBulletNear(e.bullet.x, e.bullet.y) bot.ringMover.removeBulletNear(e.bullet.x, e.bullet.y)
bot.strafeMover.removeBulletNear(e.bullet.x, e.bullet.y)
method onBulletHitBullet*(bot: ModularBot, e: BulletHitBulletEvent) = method onBulletHitBullet*(bot: ModularBot, e: BulletHitBulletEvent) =
discard bot.resolveOwnBullet(e.bullet.bulletId) # bullet-vs-bullet: free the slot discard bot.resolveOwnBullet(e.bullet.bulletId) # bullet-vs-bullet: free the slot
@@ -574,6 +579,7 @@ method onBulletHitBullet*(bot: ModularBot, e: BulletHitBulletEvent) =
bot.moveTracker.shadows.removeBulletNear(e.bullet.x, e.bullet.y) bot.moveTracker.shadows.removeBulletNear(e.bullet.x, e.bullet.y)
bot.mover.removeBulletNear(e.bullet.x, e.bullet.y) bot.mover.removeBulletNear(e.bullet.x, e.bullet.y)
bot.ringMover.removeBulletNear(e.bullet.x, e.bullet.y) bot.ringMover.removeBulletNear(e.bullet.x, e.bullet.y)
bot.strafeMover.removeBulletNear(e.bullet.x, e.bullet.y)
method onHitByBullet*(bot: ModularBot, e: HitByBulletEvent) = method onHitByBullet*(bot: ModularBot, e: HitByBulletEvent) =
# Feeds the ram bullet-rain abort window. Accumulate REAL ENERGY (the server's # Feeds the ram bullet-rain abort window. Accumulate REAL ENERGY (the server's
@@ -735,6 +741,7 @@ method onRoundStarted*(bot: ModularBot, e: RoundStartedEvent) =
bot.meleeRadar.init() bot.meleeRadar.init()
bot.mover.resetRound() bot.mover.resetRound()
bot.ringMover.resetRound() bot.ringMover.resetRound()
bot.strafeMover.resetRound()
bot.enemyTracker.resetRound() bot.enemyTracker.resetRound()
# The TM pattern gun is cold every round (no cross-battle persistence): wipe # The TM pattern gun is cold every round (no cross-battle persistence): wipe
# its clause teams and motion history explicitly, so a same-id opponent in the # its clause teams and motion history explicitly, so a same-id opponent in the
@@ -971,12 +978,15 @@ method run*(bot: ModularBot) =
if bot.isRamming: if bot.isRamming:
bot.logRamChange(true, ramReason, ramDist, ws, ramDmgRate) bot.logRamChange(true, ramReason, ramDist, ws, ramDmgRate)
# Movement dispatch. `TR_MOVEMENT=tfil_ring` routes BOTH holding and # Movement dispatch. `TR_MOVEMENT=strafe` (this module) routes holding
# ramming through the single ring engine (ram is just band [0, 50]); # through the perpendicular-strafe engine and still lets the rammer take
# `shouldRam` already accounts for the bullet-rain abort. The default `tfil` # over on a ram trigger; `tfil_ring` routes BOTH holding and ramming
# keeps the old two-engine behaviour (tfil when holding, rammer when ramming). # through the single ring engine; the default `tfil` keeps the old
# two-engine behaviour (tfil when holding, rammer when ramming).
var spd, tr: float var spd, tr: float
if MovementName == "tfil_ring": if MovementName == "strafe" and not shouldRam:
(spd, tr) = bot.strafeMover.computeMove(ws)
elif MovementName == "tfil_ring":
bot.ringMover.band = bot.ringMover.band =
if shouldRam: (lo: 0.0, hi: 50.0) if shouldRam: (lo: 0.0, hi: 50.0)
else: (lo: RangeLo, hi: RangeHi) else: (lo: RangeLo, hi: RangeHi)
@@ -1280,6 +1290,7 @@ when isMainModule:
meleeRadar: initAdaptiveMeleeRadar(), meleeRadar: initAdaptiveMeleeRadar(),
mover: TFILModule(debugGraphics: true), mover: TFILModule(debugGraphics: true),
ringMover: TFILRingModule(debugGraphics: true), ringMover: TFILRingModule(debugGraphics: true),
strafeMover: StrafeModule(debugGraphics: true),
rammer: initRammer(), rammer: initRammer(),
moveTracker: mvb.initVirtualBodyTracker(1), moveTracker: mvb.initVirtualBodyTracker(1),
currentGun: -1, currentGun: -1,
+24 -3
View File
@@ -25,6 +25,7 @@ import gun_harness/selector
import movements/ram_decision import movements/ram_decision
import movements/the_floor_is_lava import movements/the_floor_is_lava
import movements/the_floor_is_lava_ring import movements/the_floor_is_lava_ring
import movements/strafe
import guns/tm_horizon import guns/tm_horizon
import guns/bitbrain_gun import guns/bitbrain_gun
import guns/pattern_matcher import guns/pattern_matcher
@@ -174,9 +175,12 @@ proc printEffectiveValues(ctx: EnvReportContext) =
echo "[env] --- B. effective values (resolved at boot) ---" echo "[env] --- B. effective values (resolved at boot) ---"
# ── movement / process-level ModularBot knobs ───────────────────────────── # ── movement / process-level ModularBot knobs ─────────────────────────────
# Only `tfil_ring` selects the ring engine; every other value runs `tfil`, so # Only `tfil_ring` and `strafe` select a non-default engine; every other value
# report the RESOLVED engine rather than the raw string. # runs `tfil`, so report the RESOLVED engine rather than the raw string.
let effectiveMovement = if ctx.movementName == "tfil_ring": "tfil_ring" else: "tfil" let effectiveMovement =
if ctx.movementName == "tfil_ring": "tfil_ring"
elif ctx.movementName == "strafe": "strafe"
else: "tfil"
emit("TR_MOVEMENT", effectiveMovement, sourceOf("TR_MOVEMENT")) emit("TR_MOVEMENT", effectiveMovement, sourceOf("TR_MOVEMENT"))
emit("TR_MOVEMENT_LOG", onOff(MovementLog), sourceOfPresence("TR_MOVEMENT_LOG")) emit("TR_MOVEMENT_LOG", onOff(MovementLog), sourceOfPresence("TR_MOVEMENT_LOG"))
emit("TR_VBULLET_ADMIT_ONLY", onOff(ctx.vBulletAdmitOnly), emit("TR_VBULLET_ADMIT_ONLY", onOff(ctx.vBulletAdmitOnly),
@@ -259,6 +263,19 @@ proc printEffectiveValues(ctx: EnvReportContext) =
emit("TR_TFIL_PILLAR_ON", onOff(PillarHotness > 0.0), emit("TR_TFIL_PILLAR_ON", onOff(PillarHotness > 0.0),
sourceOf("TR_TFIL_PILLAR_ON")) sourceOf("TR_TFIL_PILLAR_ON"))
# ── movement (STRAFE) ─────────────────────────────────────────────────────
emit("TR_STRAFE_BAND", $StrafeBand, sourceOf("TR_STRAFE_BAND"))
emit("TR_STRAFE_SPREAD", $StrafeSpread, sourceOf("TR_STRAFE_SPREAD"))
emit("TR_STRAFE_REACH", $StrafeReach, sourceOf("TR_STRAFE_REACH"))
emit("TR_STRAFE_DWELL_MIN", $StrafeDwellMin, sourceOf("TR_STRAFE_DWELL_MIN"))
emit("TR_STRAFE_DWELL_MAX", $StrafeDwellMax, sourceOf("TR_STRAFE_DWELL_MAX"))
emit("TR_STRAFE_LOG", onOff(StrafeLog), sourceOfPresence("TR_STRAFE_LOG"))
# STRAFE's heat shape is override-able (default = the shipped TFIL field) so
# the shipped field and the ring retune can be A/B'd on one binary at boot.
emit("TR_STRAFE_CORRIDOR_HEAT", $StrafeCorridorHeat, sourceOf("TR_STRAFE_CORRIDOR_HEAT"))
emit("TR_STRAFE_WALL_HOTNESS", $StrafeWallHotness, sourceOf("TR_STRAFE_WALL_HOTNESS"))
emit("TR_STRAFE_WALL_RADIANCE", $StrafeWallRadiance, sourceOf("TR_STRAFE_WALL_RADIANCE"))
# ── ramming ─────────────────────────────────────────────────────────────── # ── ramming ───────────────────────────────────────────────────────────────
emit("TR_RAM_OPPORTUNITY", onOff(RamOppEnabled), sourceOf("TR_RAM_OPPORTUNITY")) emit("TR_RAM_OPPORTUNITY", onOff(RamOppEnabled), sourceOf("TR_RAM_OPPORTUNITY"))
emit("TR_RAM_OPP_DIST", $RamOppDist, sourceOf("TR_RAM_OPP_DIST")) emit("TR_RAM_OPP_DIST", $RamOppDist, sourceOf("TR_RAM_OPP_DIST"))
@@ -409,6 +426,10 @@ proc knownEnvNames*(): seq[string] =
"TR_TFIL_COMMIT_LOG", "TR_TFIL_COMMIT_LOG",
"TR_TFIL_HEAT_TIME", "TR_TFIL_HEAT_TAU", "TR_TFIL_HEAT_POWER_GAIN", "TR_TFIL_HEAT_TIME", "TR_TFIL_HEAT_TAU", "TR_TFIL_HEAT_POWER_GAIN",
"TR_TFIL_PILLAR_ON", "TR_TFIL_PILLAR_ON",
"TR_STRAFE_BAND", "TR_STRAFE_SPREAD", "TR_STRAFE_REACH",
"TR_STRAFE_DWELL_MIN", "TR_STRAFE_DWELL_MAX", "TR_STRAFE_LOG",
"TR_STRAFE_CORRIDOR_HEAT", "TR_STRAFE_WALL_HOTNESS",
"TR_STRAFE_WALL_RADIANCE",
# harness vars (read by the test framework, inherited by the bot, so they # harness vars (read by the test framework, inherited by the bot, so they
# must NOT be reported as typos) # must NOT be reported as typos)
"TR_SERVER_JAR", "TR_BATTLE_RUNNER", "TR_BATTLE_RUNNER_DIR", "TR_SERVER_JAR", "TR_BATTLE_RUNNER", "TR_BATTLE_RUNNER_DIR",
+740
View File
@@ -0,0 +1,740 @@
## STRAFE — body pinned perpendicular to the threat; reversals by SIGN FLIP.
##
## ── The idea (the owner's design) ───────────────────────────────────────────
## TFIL changes its left<->right direction by TURNING the body. Job j85 measured
## the cost of that turn: the speed at the reversal tick drops to ~0.6 px/tick
## (from ~6.4), takes 7-8 ticks to recover, and the hit rate on reversal ticks
## is 1.40x baseline (peaking 1.84x at 6-10 ticks) — the shipped mover also has
## the LOWEST mean speed of the arms tested (4.69 vs 5.37 px/tick).
##
## In Tank Royale the backward speed EQUALS the forward speed (measured
## +8.000 / -8.000 over 75,518 ticks), so a reversal is free if it is done by
## flipping the sign of `setForward` instead of turning the hull around. STRAFE
## exploits exactly that:
##
## * keep the BODY pinned ~perpendicular to the threat (`threat axis` below),
## so "move left" and "move right" are both along the body axis;
## * pick a random safe tile on the perpendicular line (forward or backward)
## and move to it with `setForward(±8)`;
## * NEVER turn to face a movement target — the only turns are small
## corrections that keep the heading inside a band around the perpendicular.
##
## Bonus physics: the bot is a 36 px square. Projected width is 36 px side-on
## or head-on but 50.9 px at 45 degrees, so staying near 90 degrees avoids the
## worst orientation (up to 29% smaller target).
##
## ── Threat axis (point 1 of the spec) ───────────────────────────────────────
## When a bullet is in flight the axis is the INCOMING BULLET's direction: a
## bullet comes from where the enemy WAS when it fired, which at long range
## differs from its current position by 100+ px. When the sky is clear the axis
## falls back to the PERPENDICULAR of the enemy bearing.
##
## ── Heading band, not an exact pin (point 2) ────────────────────────────────
## The body heading is kept inside `TR_STRAFE_BAND` degrees of the perpendicular
## line (`lineDir`); the band offset is re-randomised on every pick so the
## heading is not a constant. Only turns when OUTSIDE the band, and turns the
## short way (the folded deviation is in [-90, 90]).
##
## ── Candidate tiles (point 3) ───────────────────────────────────────────────
## Tiles on the perpendicular line through the current position, both forward
## and backward, within `TR_STRAFE_REACH` px, inside the arena, with a small
## PERPENDICULAR JITTER of `±TR_STRAFE_SPREAD` tiles (the owner's "spread a
## little"). A tile is acceptable when the max heat ON THE STRAIGHT-LINE PATH
## from the bot is <= `PathDangerThreshold` (10.0) — the SAME safety rule TFIL
## uses. The heat field itself is the SHIPPED TFIL field (jobs j105/j106: the
## time-indexed bullet model + the pillar-free default) — see the reuse note
## below.
##
## ── Move by sign only (point 5) ─────────────────────────────────────────────
## `speed = MaxSpeed * sign`, where `sign` is +1 when the chosen tile lies along
## the current heading and -1 when it lies opposite. There is NO turn-to-target
## anywhere in this mover.
##
## ── Randomised dwell (point 6) ──────────────────────────────────────────────
## The target is re-picked after `rand(TR_STRAFE_DWELL_MIN .. TR_STRAFE_DWELL_MAX)`
## ticks, on arrival, or immediately when the chosen tile's heat spikes by
## `DangerReplanThreshold` (a serious threat). This is the PRIMARY anti-pattern
## defence: a periodic reversal is trivially learnable by DrussGT's pattern gun,
## so the reversal TIMING is randomised and the offline entropy gate (B) checks
## it.
##
## ── Reuse of the j105/j106 heat machinery ───────────────────────────────────
## This module does NOT re-implement the time-indexed bullet model. It imports
## `movements/the_floor_is_lava` and calls its EXPORTED `heatDecay(dt)` and
## `bulletMagScale(power)`, and reads its exported `PillarHotness` /
## `PillarRadiance` (0/0 = the shipped pillar-free default) and `TfilHeatTime`
## (for the debug corridor reach). So `TR_TFIL_HEAT_TIME` / `TR_TFIL_HEAT_TAU` /
## `TR_TFIL_HEAT_POWER_GAIN` / `TR_TFIL_PILLAR_ON` drive the STRAFE field exactly
## as they drive TFIL's. The bullet tracking, heat painting and path sampling are
## copied from the shipped mover (the same pattern `the_floor_is_lava_ring.nim`
## uses) because the shipped file must stay byte-identical and its private
## constants are not exported.
##
## ── Env knobs (all read at module init) ─────────────────────────────────────
## TR_MOVEMENT = strafe (selects this engine; default stays tfil)
## TR_STRAFE_BAND 20.0 heading band half-width (deg)
## TR_STRAFE_SPREAD 1 perpendicular jitter (tiles, ±)
## TR_STRAFE_REACH 144.0 along-line reach (px)
## TR_STRAFE_DWELL_MIN 6 min ticks before a re-pick
## TR_STRAFE_DWELL_MAX 20 max ticks before a re-pick
## TR_STRAFE_LOG off presence-based: echo one line per pick
## TR_STRAFE_CORRIDOR_HEAT 20.0 lava per corridor tile (shipped TFIL value)
## TR_STRAFE_WALL_HOTNESS 30.0 peak wall radiance (shipped TFIL value)
## TR_STRAFE_WALL_RADIANCE 10.0 wall radiance falloff (shipped TFIL value)
##
## The default `TR_MOVEMENT=tfil` path is untouched; this module is only ever
## called when the bot explicitly selects `strafe`.
import std/[math, random, os]
from std/strutils import parseFloat, parseInt, strip
import std/strformat
import gun_harness/gun_interface
import movement_harness/movement_interface
import robocode_tankroyale_botapi/graphics
import robocode_tankroyale_botapi/color
# j105/j106 reuse: the exported time-indexed heat helpers + pillar globals.
import movements/the_floor_is_lava
const GridSize = 36.0
const MaxSpeed = 8.0
const BulletCoreRadiusMin = 9.0 ## core radius at power 0.1
const BulletCoreRadiusMax = 54.0 ## core radius at power 3.0
const BulletAuraExtMin = 36.0 ## aura extension at power 3.0 (slow)
const BulletAuraExtMax = 54.0 ## aura extension at power 0.1 (fast)
const BulletCore = 10.0 ## lava per bullet-overlapping tile (shipped default)
const BulletAura = 5.0 ## lava for aura ring tiles (shipped default)
const EnemyCoreRadius = 18.0 ## half of 36px body
const EnemyAuraRadius = 54.0 ## 18 + 36
const EnemyCore = 40.0 ## lava per tile overlapping enemy body
const EnemyAura = 10.0 ## lava per tile in enemy aura ring
const CorridorHeatDefault = 20.0 ## shipped TFIL corridor heat
const WallHotnessDefault = 30.0 ## shipped TFIL wall radiance peak
const WallRadianceDefault = 10.0 ## shipped TFIL wall radiance falloff
## Heat shape is override-able so the two candidate fields can be compared on
## one binary (the shipped field vs the ring variant's retune). Default = the
## SHIPPED TFIL field, so the strafe mover is judged on the same field as the
## baseline. `TR_STRAFE_CORRIDOR_HEAT` / `TR_STRAFE_WALL_HOTNESS` /
## `TR_STRAFE_WALL_RADIANCE` are strafe-specific (the ring mover's own field is
## controlled by its separate `TR_TFIL_*` names).
var
StrafeCorridorHeat* = CorridorHeatDefault
StrafeWallHotness* = WallHotnessDefault
StrafeWallRadiance* = WallRadianceDefault
const PathSampleStep = 18.0 ## ~half a tile
const PathDangerThreshold = 10.0 ## max lava on path; above this = unsafe
const DangerReplanThreshold = 25.0 ## serious-threat replan (bullet core)
const MaxTrackedBullets = 20 ## hard cap on tracked bullets
# ── Env knobs ────────────────────────────────────────────────────────────────
proc getEnvFloat(name: string, default: float): float =
let s = getEnv(name, "")
if s.len == 0: return default
try: result = parseFloat(s.strip())
except ValueError: result = default
proc getEnvInt(name: string, default: int): int =
let s = getEnv(name, "")
if s.len == 0: return default
try: result = parseInt(s.strip())
except ValueError: result = default
const
DefaultStrafeBand = 20.0
DefaultStrafeSpread = 1
DefaultStrafeReach = 144.0
DefaultStrafeDwellMin = 6
DefaultStrafeDwellMax = 20
var
StrafeBand* = DefaultStrafeBand
StrafeSpread* = DefaultStrafeSpread
StrafeReach* = DefaultStrafeReach
StrafeDwellMin* = DefaultStrafeDwellMin
StrafeDwellMax* = DefaultStrafeDwellMax
StrafeLog* = false
proc loadStrafeHeatEnv*() =
## Re-read the heat-shape overrides. Exposed so a gate can restore the shipped
## field after temporarily retuning it on the SAME process.
StrafeCorridorHeat = getEnvFloat("TR_STRAFE_CORRIDOR_HEAT", CorridorHeatDefault)
StrafeWallHotness = getEnvFloat("TR_STRAFE_WALL_HOTNESS", WallHotnessDefault)
StrafeWallRadiance = getEnvFloat("TR_STRAFE_WALL_RADIANCE", WallRadianceDefault)
proc loadStrafeEnv*() =
## Read the strafe knobs. Called once at module init; callable again after
## `putEnv` so a gate script can exercise the arms in one process.
StrafeBand = max(0.0, min(90.0, getEnvFloat("TR_STRAFE_BAND", DefaultStrafeBand)))
StrafeSpread = max(0, getEnvInt("TR_STRAFE_SPREAD", DefaultStrafeSpread))
StrafeReach = max(GridSize, getEnvFloat("TR_STRAFE_REACH", DefaultStrafeReach))
StrafeDwellMin = max(1, getEnvInt("TR_STRAFE_DWELL_MIN", DefaultStrafeDwellMin))
StrafeDwellMax = max(StrafeDwellMin, getEnvInt("TR_STRAFE_DWELL_MAX", DefaultStrafeDwellMax))
StrafeLog = existsEnv("TR_STRAFE_LOG")
loadStrafeHeatEnv()
loadStrafeEnv()
# ── small angle helpers ──────────────────────────────────────────────────────
proc wrap180(d: float): float {.inline.} =
result = d
while result > 180.0: result -= 360.0
while result < -180.0: result += 360.0
const DegToRad = PI / 180.0
# ── module types ─────────────────────────────────────────────────────────────
type
TrackedBullet = object
originX, originY: float
x, y: float
velX, velY: float ## speed * cos(heading), speed * sin(heading)
power: float
alive: bool
age: int
StrafeModule* = object
debugGraphics*: bool
cols*, rows*: int
marginX*, marginY*: float
arenaWidth*, arenaHeight*: float
lava: seq[float]
bullets: seq[TrackedBullet]
prevEnergy: seq[tuple[id: int, energy: float]]
# ── decision state ──
targetX*, targetY*: float ## chosen tile centre (world coords)
targetValid*: bool
targetLava: float ## heat at the chosen tile when picked
dwell*: int ## ticks remaining on the current target
dir*: float ## commanded sign: +1 forward, -1 backward
bandOffset: float ## random in [-band, band], re-rolled per pick
lineDir*: float ## undirected strafe line bearing (deg)
# ── diagnostics (gate B + GUI) ──
callCount*: int
picks*: int
lastPickCall*: int ## callCount at the last pick (interval source)
lastCandCount*: int ## candidates generated at the last pick
lastSafeCount*: int ## of those, path-safe at the last pick
fallbackPicks*: int ## picks where the safe pool was empty
reversals*: int ## sign flips of the commanded direction
lineDirFlips*: int ## times the axis orientation flipped
lastCmdSign*: float
lastLineForward: float
lastBotX, lastBotY: float
lastTileCol, lastTileRow: int
proc initStrafe*(): StrafeModule =
StrafeModule(debugGraphics: false)
proc removeBulletNear*(m: var StrafeModule, x, y: float) =
## Mark the tracked bullet closest to (x,y) within GridSize tolerance as dead.
var bestIdx = -1
var bestD2 = GridSize * GridSize
for i, b in m.bullets:
let d2 = (b.x - x)*(b.x - x) + (b.y - y)*(b.y - y)
if d2 < bestD2:
bestD2 = d2
bestIdx = i
if bestIdx >= 0:
m.bullets.del(bestIdx)
proc prevEnergyGet(m: StrafeModule, id: int): float =
for e in m.prevEnergy:
if e.id == id: return e.energy
100.0
proc prevEnergySet(m: var StrafeModule, id: int, energy: float) =
for i in 0..<m.prevEnergy.len:
if m.prevEnergy[i].id == id:
m.prevEnergy[i].energy = energy
return
m.prevEnergy.add((id: id, energy: energy))
proc clearGraphics*(m: var StrafeModule) =
## No-op: the SVG buffer is a module-level global cleared by the framework
## after every go(). Exists so callers can signal "STRAFE is inactive".
discard
proc resetRound*(m: var StrafeModule) =
m.bullets = @[]
m.prevEnergy = @[]
m.targetValid = false
m.targetLava = 0.0
m.dwell = 0
m.dir = 1.0
m.bandOffset = 0.0
m.lineDir = 0.0
m.callCount = 0
m.picks = 0
m.lastPickCall = 0
m.lastCandCount = 0
m.lastSafeCount = 0
m.fallbackPicks = 0
m.reversals = 0
m.lineDirFlips = 0
m.lastCmdSign = 0.0
m.lastLineForward = 0.0
m.lastBotX = 0.0
m.lastBotY = 0.0
m.lastTileCol = 0
m.lastTileRow = 0
proc initGrid(m: var StrafeModule, arenaWidth, arenaHeight: float) =
m.cols = int(arenaWidth / GridSize)
m.rows = int(arenaHeight / GridSize)
m.marginX = (arenaWidth - m.cols.float * GridSize) / 2.0
m.marginY = (arenaHeight - m.rows.float * GridSize) / 2.0
m.arenaWidth = arenaWidth
m.arenaHeight = arenaHeight
m.lava = newSeq[float](m.cols * m.rows)
proc lavaAt(m: StrafeModule, col, row: int): float {.inline.} =
m.lava[row * m.cols + col]
proc tileAt(m: StrafeModule, wx, wy: float): tuple[col, row: int] =
(col: clamp(int((wx - m.marginX) / GridSize), 0, m.cols - 1),
row: clamp(int((wy - m.marginY) / GridSize), 0, m.rows - 1))
proc bulletRadii(power: float): tuple[core, aura: float] =
let t = (power - 0.1) / 2.9
let core = BulletCoreRadiusMin + t * (BulletCoreRadiusMax - BulletCoreRadiusMin)
let auraExt = BulletAuraExtMax - t * (BulletAuraExtMax - BulletAuraExtMin)
(core, core + auraExt)
# ── bullet tracking (copied from the shipped TFIL mover) ─────────────────────
proc detectFires(m: var StrafeModule, ws: WorldState) =
for ei in ws.enemies:
let prev = m.prevEnergyGet(ei.id)
let drop = prev - ei.energy
m.prevEnergySet(ei.id, ei.energy)
if drop >= 0.09 and drop <= 3.01:
let speed = 20.0 - 3.0 * drop
let dist = sqrt((ws.selfX - ei.x)^2 + (ws.selfY - ei.y)^2)
let travelTime = dist / speed
let predX = ws.selfX + ws.selfSpeed * cos(ws.selfHeading * DegToRad) * travelTime
let predY = ws.selfY + ws.selfSpeed * sin(ws.selfHeading * DegToRad) * travelTime
let heading = arctan2(predY - ei.y, predX - ei.x)
if m.bullets.len >= MaxTrackedBullets:
m.bullets.del(0)
m.bullets.add TrackedBullet(
originX: ei.x, originY: ei.y,
x: ei.x, y: ei.y,
velX: speed * cos(heading),
velY: speed * sin(heading),
power: drop, alive: true, age: 0)
proc advanceBullets(m: var StrafeModule, selfX, selfY: float) =
var i = 0
while i < m.bullets.len:
var b = m.bullets[i]
b.x += b.velX
b.y += b.velY
b.age += 1
let dx = selfX - b.x
let dy = selfY - b.y
let dot = b.velX * dx + b.velY * dy
let outOfBounds = b.x < 0.0 or b.x > m.arenaWidth or b.y < 0.0 or b.y > m.arenaHeight
if dot < 0.0 or outOfBounds or b.age > 200:
b.alive = false
m.bullets[i] = b
if b.alive: inc i
else: m.bullets.del(i)
# ── heat field (SHIPPED TFIL field; time model + pillar via the import) ──────
proc buildHeat(m: var StrafeModule, ws: WorldState) =
for i in 0..<m.lava.len: m.lava[i] = 0.0
# Bullet cores / auras (j105 time model: heat is a function of time-to-arrive).
for b in m.bullets:
let bx = b.x
let by = b.y
let (coreR, auraR) = bulletRadii(b.power)
let bSpeed = sqrt(b.velX * b.velX + b.velY * b.velY)
let bUx = if bSpeed > 0.0: b.velX / bSpeed else: 0.0
let bUy = if bSpeed > 0.0: b.velY / bSpeed else: 0.0
let bMag = bulletMagScale(b.power)
let colMin = max(0, int(floor((bx - auraR - m.marginX) / GridSize)))
let colMax = min(m.cols-1, int(floor((bx + auraR - m.marginX) / GridSize)))
let rowMin = max(0, int(floor((by - auraR - m.marginY) / GridSize)))
let rowMax = min(m.rows-1, int(floor((by + auraR - m.marginY) / GridSize)))
for row in rowMin..rowMax:
for col in colMin..colMax:
let x0 = m.marginX + col.float * GridSize
let y0 = m.marginY + row.float * GridSize
let nearX = clamp(bx, x0, x0 + GridSize)
let nearY = clamp(by, y0, y0 + GridSize)
let dx = nearX - bx
let dy = nearY - by
let d2 = dx*dx + dy*dy
if d2 <= coreR * coreR:
let along = dx * bUx + dy * bUy
m.lava[row * m.cols + col] += BulletCore * bMag * heatDecay(along / bSpeed)
elif d2 <= auraR * auraR:
let along = dx * bUx + dy * bUy
m.lava[row * m.cols + col] += BulletAura * bMag * heatDecay(along / bSpeed)
# Corridors (rotated rectangle from the bullet to the wall, auraR wide).
for b in m.bullets:
let speed = sqrt(b.velX * b.velX + b.velY * b.velY)
if speed < 0.001: continue
let dx = b.velX / speed
let dy = b.velY / speed
let px = -dy
let py = dx
var tMin = Inf
if dx > 0.0: tMin = min(tMin, (m.arenaWidth - b.x) / dx)
elif dx < 0.0: tMin = min(tMin, (0.0 - b.x) / dx)
if dy > 0.0: tMin = min(tMin, (m.arenaHeight - b.y) / dy)
elif dy < 0.0: tMin = min(tMin, (0.0 - b.y) / dy)
if tMin == 0.0: continue
let (_, auraR) = bulletRadii(b.power)
let bMag = bulletMagScale(b.power)
let wx = b.x + dx * tMin
let wy = b.y + dy * tMin
let c0x = b.x + px * auraR; let c0y = b.y + py * auraR
let c1x = b.x - px * auraR; let c1y = b.y - py * auraR
let c2x = wx - px * auraR; let c2y = wy - py * auraR
let c3x = wx + px * auraR; let c3y = wy + py * auraR
let xMin = min(min(c0x, c1x), min(c2x, c3x))
let xMax = max(max(c0x, c1x), max(c2x, c3x))
let yMin = min(min(c0y, c1y), min(c2y, c3y))
let yMax = max(max(c0y, c1y), max(c2y, c3y))
let colMin = max(0, int(floor((xMin - m.marginX) / GridSize)))
let colMax = min(m.cols-1, int(floor((xMax - m.marginX) / GridSize)))
let rowMin = max(0, int(floor((yMin - m.marginY) / GridSize)))
let rowMax = min(m.rows-1, int(floor((yMax - m.marginY) / GridSize)))
for row in rowMin..rowMax:
for col in colMin..colMax:
let cx = m.marginX + (col.float + 0.5) * GridSize
let cy = m.marginY + (row.float + 0.5) * GridSize
let relX = cx - b.x
let relY = cy - b.y
let along = relX * dx + relY * dy
let perp = relX * px + relY * py
if along >= 0.0 and along <= tMin and perp >= -auraR and perp <= auraR:
m.lava[row * m.cols + col] += StrafeCorridorHeat * bMag * heatDecay(along / speed)
# Enemy auras
for ei in ws.enemies:
let ex = ei.x
let ey = ei.y
let colMin = max(0, int(floor((ex - EnemyAuraRadius - m.marginX) / GridSize)))
let colMax = min(m.cols-1, int(floor((ex + EnemyAuraRadius - m.marginX) / GridSize)))
let rowMin = max(0, int(floor((ey - EnemyAuraRadius - m.marginY) / GridSize)))
let rowMax = min(m.rows-1, int(floor((ey + EnemyAuraRadius - m.marginY) / GridSize)))
for row in rowMin..rowMax:
for col in colMin..colMax:
let x0 = m.marginX + col.float * GridSize
let y0 = m.marginY + row.float * GridSize
let nearX = clamp(ex, x0, x0 + GridSize)
let nearY = clamp(ey, y0, y0 + GridSize)
let dx = nearX - ex
let dy = nearY - ey
let d2 = dx*dx + dy*dy
if d2 <= EnemyCoreRadius * EnemyCoreRadius:
m.lava[row * m.cols + col] += EnemyCore
elif d2 <= EnemyAuraRadius * EnemyAuraRadius:
m.lava[row * m.cols + col] += EnemyAura
# Wall radiance
for row in 0..<m.rows:
for col in 0..<m.cols:
let heat = max(0.0, StrafeWallHotness - col.float * StrafeWallRadiance) +
max(0.0, StrafeWallHotness - (m.cols-1-col).float * StrafeWallRadiance) +
max(0.0, StrafeWallHotness - row.float * StrafeWallRadiance) +
max(0.0, StrafeWallHotness - (m.rows-1-row).float * StrafeWallRadiance)
m.lava[row * m.cols + col] += heat
# Pillar radiance (imported globals; 0/0 = the shipped pillar-free default).
let pc0 = if m.cols mod 2 == 1: m.cols div 2 else: m.cols div 2 - 1
let pc1 = m.cols div 2
let pr0 = if m.rows mod 2 == 1: m.rows div 2 else: m.rows div 2 - 1
let pr1 = m.rows div 2
for row in 0..<m.rows:
for col in 0..<m.cols:
var minDist = int.high
for pcol in pc0..pc1:
for prow in pr0..pr1:
let d = max(abs(col - pcol), abs(row - prow))
if d < minDist: minDist = d
m.lava[row * m.cols + col] += max(0.0, PillarHotness - minDist.float * PillarRadiance)
proc pathMaxHeat(m: StrafeModule, x0, y0, x1, y1: float): float =
let dx = x1 - x0
let dy = y1 - y0
let dist = sqrt(dx*dx + dy*dy)
if dist < 0.1: return 0.0
let steps = max(1, int(dist / PathSampleStep))
for si in 0..steps:
let f = si.float / steps.float
let (c, r) = m.tileAt(x0 + dx * f, y0 + dy * f)
result = max(result, m.lavaAt(c, r))
# ── the threat axis ──────────────────────────────────────────────────────────
proc threatBearing(m: StrafeModule, ws: WorldState): float =
## Incoming bullet direction when a bullet is in flight (nearest one), else
## the enemy bearing.
var bestD = Inf
var found = false
for b in m.bullets:
let d = hypot(b.x - ws.selfX, b.y - ws.selfY)
if d < bestD:
bestD = d
result = arctan2(b.velY, b.velX) * 180.0 / PI
found = true
if found: return result
if ws.enemyX != 0.0 or ws.enemyY != 0.0:
return arctan2(ws.enemyY - ws.selfY, ws.enemyX - ws.selfX) * 180.0 / PI
for ei in ws.enemies:
return arctan2(ei.y - ws.selfY, ei.x - ws.selfX) * 180.0 / PI
return 0.0
# ── target picking ───────────────────────────────────────────────────────────
type
Cand = object
col, row: int
x, y: float
pathHeat: float
along: float ## signed offset along the line (+ = forward of lineForward)
proc pickTarget(m: var StrafeModule, ws: WorldState, lineForward: float) =
let ux = cos(lineForward * DegToRad)
let uy = sin(lineForward * DegToRad)
let px = -uy
let py = ux
let kmax = max(1, int(StrafeReach / GridSize))
let spread = max(0, StrafeSpread)
let (bc, br) = m.tileAt(ws.selfX, ws.selfY)
var cands: seq[Cand]
for k in 1..kmax:
for s in [-1.0, 1.0]:
let along = s * k.float * GridSize
for j in -spread..spread:
let wx = ws.selfX + ux * along + px * (j.float * GridSize)
let wy = ws.selfY + uy * along + py * (j.float * GridSize)
if wx < 0.0 or wx >= m.arenaWidth or wy < 0.0 or wy >= m.arenaHeight: continue
let (c, r) = m.tileAt(wx, wy)
if c == bc and r == br: continue
var dup = false
for e in cands:
if e.col == c and e.row == r: dup = true; break
if dup: continue
let cx = m.marginX + (c.float + 0.5) * GridSize
let cy = m.marginY + (r.float + 0.5) * GridSize
cands.add Cand(col: c, row: r, x: cx, y: cy,
pathHeat: m.pathMaxHeat(ws.selfX, ws.selfY, cx, cy),
along: along)
var safe: seq[Cand]
for c in cands:
if c.pathHeat <= PathDangerThreshold: safe.add c
m.lastCandCount = cands.len
m.lastSafeCount = safe.len
var pool: seq[Cand]
if safe.len > 0:
pool = safe
elif cands.len > 0:
# Fallback (never freeze): the two coolest tiles on the line, over
# threshold but still the least dangerous direction to move.
var sorted = cands
for i in 1..<sorted.len:
let key = sorted[i]
var j = i - 1
while j >= 0 and sorted[j].pathHeat > key.pathHeat:
sorted[j + 1] = sorted[j]
dec j
sorted[j + 1] = key
let take = min(2, sorted.len)
for i in 0..<take: pool.add sorted[i]
inc m.fallbackPicks
if pool.len == 0:
m.targetValid = false
return
let chosen = rand(pool.high)
m.targetX = pool[chosen].x
m.targetY = pool[chosen].y
m.targetLava = m.lavaAt(pool[chosen].col, pool[chosen].row)
m.targetValid = true
# sign is frozen for the whole dwell: reversal timing == dwell timing.
let hx = cos(ws.selfHeading * DegToRad)
let hy = sin(ws.selfHeading * DegToRad)
let alongDot = (m.targetX - ws.selfX) * hx + (m.targetY - ws.selfY) * hy
m.dir = if alongDot >= 0.0: 1.0 else: -1.0
if m.lastCmdSign == 0.0: m.lastCmdSign = m.dir
# Randomised dwell + randomisation inside the band (point 2 / point 6).
m.dwell = rand(StrafeDwellMin..StrafeDwellMax)
m.bandOffset = rand(2.0 * StrafeBand) - StrafeBand
m.lastPickCall = m.callCount
inc m.picks
if StrafeLog:
let reason = if safe.len == 0: "fallback" else: "pick"
echo fmt"[strafe] {reason} n={cands.len} safe={safe.len} " &
fmt"along={pool[chosen].along.int} dir={m.dir.int} " &
fmt"dwell={m.dwell} line={lineForward.int} band={m.bandOffset:.0f}"
if safe.len == 0:
echo fmt"[strafe] WARNING: no SAFE tile on the line " &
fmt"(cands={cands.len}, least-hot={pool[chosen].pathHeat:.1f})"
# ── main entry point ─────────────────────────────────────────────────────────
proc computeMove*(m: var StrafeModule, ws: WorldState): MoveCommand =
if m.cols == 0:
m.initGrid(ws.arenaWidth, ws.arenaHeight)
# Soft reset on a position jump (a ram teleport moved us).
let jumpDist = sqrt((ws.selfX - m.lastBotX)^2 + (ws.selfY - m.lastBotY)^2)
let jumped = (m.callCount > 0) and (jumpDist > 12.0)
if jumped:
m.dwell = 0
m.targetValid = false
m.bullets = @[]
m.prevEnergy = @[]
for ei in ws.enemies:
m.prevEnergySet(ei.id, ei.energy)
m.advanceBullets(ws.selfX, ws.selfY)
m.detectFires(ws)
m.buildHeat(ws)
# ── threat axis -> perpendicular line ──
let threat = m.threatBearing(ws)
let lineAngle = threat + 90.0
m.lineDir = lineAngle
# Forward orientation = the half of the (undirected) line nearest our heading.
var lineForward = lineAngle
let hRel = wrap180(ws.selfHeading - lineAngle)
if hRel > 90.0 or hRel < -90.0:
lineForward = lineAngle + 180.0
if m.picks > 0:
let fl = abs(wrap180(lineForward - m.lastLineForward))
if fl > 90.0: inc m.lineDirFlips
m.lastLineForward = lineForward
# ── target (re)pick ──
var serious = false
if m.targetValid:
let (tc, tr) = m.tileAt(m.targetX, m.targetY)
if m.lavaAt(tc, tr) > m.targetLava + DangerReplanThreshold:
serious = true
if hypot(m.targetX - ws.selfX, m.targetY - ws.selfY) < GridSize * 0.75:
m.dwell = 0
if (not m.targetValid) or m.dwell <= 0 or serious:
m.pickTarget(ws, lineForward)
else:
dec m.dwell
# ── heading band: turn ONLY to stay perpendicular, never to the target ──
var turnRate = 0.0
let mtr = 10.0 - 0.75 * abs(ws.selfSpeed)
let dev = wrap180(ws.selfHeading - lineForward)
if abs(dev) > StrafeBand:
let targetHeading = lineForward + m.bandOffset
turnRate = clamp(wrap180(targetHeading - ws.selfHeading), -mtr, mtr)
# ── move by sign only ──
# Never freeze: if there is no valid target (the rare all-candidates-outside
# case) keep driving on the last sign. The normal path picks a target every
# dwell, and the fallback pool is non-empty whenever any candidate tile exists.
if m.targetValid and m.dir != m.lastCmdSign:
inc m.reversals
m.lastCmdSign = m.dir
result = (speed: MaxSpeed * (if m.dir != 0.0: m.dir else: 1.0), turnRate: turnRate)
# ── GUI overlay (the user watches this) ──
if m.debugGraphics:
let ux = cos(lineForward * DegToRad)
let uy = sin(lineForward * DegToRad)
# Strafe line across the arena (cyan).
setStrokeColor(fromHex("#00FFFF"))
setStrokeWidth(1.0)
drawLine(ws.selfX - ux * 600.0, ws.selfY - uy * 600.0,
ws.selfX + ux * 600.0, ws.selfY + uy * 600.0)
# Threat axis (enemy/bullet bearing) as a faint grey line through us.
let tx = cos(threat * DegToRad)
let ty = sin(threat * DegToRad)
setStrokeColor(fromHex("#888888"))
drawLine(ws.selfX - tx * 600.0, ws.selfY - ty * 600.0,
ws.selfX + tx * 600.0, ws.selfY + ty * 600.0)
# Candidate line reach up to kmax tiles: safe tiles bright, unsafe dim.
let kmax = max(1, int(StrafeReach / GridSize))
let spread = max(0, StrafeSpread)
let px = -uy
let py = ux
for k in 1..kmax:
for s in [-1.0, 1.0]:
let along = s * k.float * GridSize
for j in -spread..spread:
let wx = ws.selfX + ux * along + px * (j.float * GridSize)
let wy = ws.selfY + uy * along + py * (j.float * GridSize)
if wx < 0.0 or wx >= m.arenaWidth or wy < 0.0 or wy >= m.arenaHeight: continue
let (c, r) = m.tileAt(wx, wy)
let x0 = m.marginX + c.float * GridSize
let y0 = m.marginY + r.float * GridSize
if m.lavaAt(c, r) <= PathDangerThreshold:
setStrokeColor(fromHex("#00FF00"))
setStrokeWidth(1.0)
else:
setStrokeColor(fromHex("#804000"))
setStrokeWidth(1.0)
drawRectangle(x0, y0, GridSize, GridSize)
# Chosen target tile (magenta fill) + the sign-coloured movement ray.
if m.targetValid:
let (cc, cr) = m.tileAt(m.targetX, m.targetY)
let gx0 = m.marginX + cc.float * GridSize
let gy0 = m.marginY + cr.float * GridSize
setStrokeColor(fromHex("#FF00FF"))
setStrokeWidth(2.5)
drawRectangle(gx0, gy0, GridSize, GridSize)
setFillColor(fromHex("#FF00FF"))
fillCircle(m.targetX, m.targetY, 5.0)
# movement ray: green forward, red backward (the sign flip is the point)
setStrokeColor(if m.dir >= 0.0: fromHex("#00FF00") else: fromHex("#FF2222"))
setStrokeWidth(2.0)
drawLine(ws.selfX, ws.selfY,
ws.selfX + ux * m.dir * StrafeReach, ws.selfY + uy * m.dir * StrafeReach)
# Heading band: the two ±band boundary rays (yellow) around the line.
setStrokeColor(fromHex("#FFFF00"))
setStrokeWidth(1.0)
let bl = (lineForward - StrafeBand) * DegToRad
let br2 = (lineForward + StrafeBand) * DegToRad
drawLine(ws.selfX, ws.selfY,
ws.selfX + cos(bl) * 70.0, ws.selfY + sin(bl) * 70.0)
drawLine(ws.selfX, ws.selfY,
ws.selfX + cos(br2) * 70.0, ws.selfY + sin(br2) * 70.0)
# Current heading ray (blue).
setStrokeColor(fromHex("#3399FF"))
setStrokeWidth(2.0)
drawLine(ws.selfX, ws.selfY,
ws.selfX + cos(ws.selfHeading * DegToRad) * 60.0,
ws.selfY + sin(ws.selfHeading * DegToRad) * 60.0)
# snapshot for the next call
m.lastBotX = ws.selfX
m.lastBotY = ws.selfY
m.lastTileCol = clamp(int((ws.selfX - m.marginX) / GridSize), 0, m.cols - 1)
m.lastTileRow = clamp(int((ws.selfY - m.marginY) / GridSize), 0, m.rows - 1)
m.callCount += 1
+322
View File
@@ -0,0 +1,322 @@
## STRAFE gates — the two CHEAP GATES the spec demands, run OFFLINE on the
## recorded DrussGT fixture (no Java, no server, no battle).
##
## nim c -r --nimcache:/tmp/nc_j108 --path:common_libs \
## common_libs/tests/measure_strafe_gates.nim
##
## GATE A TILE AVAILABILITY: on every STRAFE pick, how many of the tiles on the
## perpendicular line are SAFE (path max heat <= PathDangerThreshold),
## and how often the safe pool is EMPTY (the mover then falls back to
## the least-hot tile). If the line is usually blocked the design fails
## by construction and this prints it.
##
## GATE B PREDICTABILITY: the main risk. A constant heading running back and
## forth on one line is exactly what a pattern-matching gun exploits.
## Both movers are replayed over the SAME fixture with the SAME seed and
## compared on:
## * reversal-interval entropy (bits) — higher = less periodic
## * direction entropy (bits) — higher = less biased
## * max |autocorrelation| of the signed direction over lags 20..60
## * mean |turnRate| and the fraction of ticks with no turn at all
## * mean |speed| around a reversal (the j85 speed-collapse test)
##
## The fixture was recorded with DrussGT as the SUBJECT and ModularBot as the
## adversary (see the file's meta line), so `sx/sy/sh` are OUR recorded states
## and `ex/ey/ee` are DrussGT's. Replaying it drives each mover open-loop; that
## is the standard repo practice for movement gates.
import std/[os, json, math, random, strformat, strutils, algorithm, tables, sequtils]
import gun_harness/gun_interface
import movements/the_floor_is_lava
import movements/strafe
const repoRoot = currentSourcePath().parentDir.parentDir.parentDir
const fixtureRel = "tr_drussgt_vs_modularbot.jsonl"
const Seed = 20250923
const ArenaW = 800.0
const ArenaH = 600.0
proc loadStates(): seq[WorldState] =
let path = repoRoot / "tools" / "fixtures" / fixtureRel
for rawLine in lines(path):
let line = rawLine.strip()
if line.len == 0: continue
let n = parseJson(line)
if n.hasKey("meta") or n.hasKey("end"): continue
let ex = n["ex"].getFloat()
let ey = n["ey"].getFloat()
result.add WorldState(
enemyX: ex, enemyY: ey,
enemyHeading: n["eh"].getFloat(), enemySpeed: n["es"].getFloat(),
enemyEnergy: n["ee"].getFloat(),
selfX: n["sx"].getFloat(), selfY: n["sy"].getFloat(),
selfHeading: n["sh"].getFloat(), selfSpeed: n["ss"].getFloat(),
selfEnergy: n["se"].getFloat(),
arenaWidth: ArenaW, arenaHeight: ArenaH,
tick: n["tick"].getInt(),
enemies: @[EnemyInfo(id: 1, x: ex, y: ey,
heading: n["eh"].getFloat(), speed: n["es"].getFloat(),
energy: n["ee"].getFloat())])
proc loadRoundStarts(): seq[int] =
let side = repoRoot / "tools" / "fixtures" / "drussgt_meta" /
(fixtureRel & ".rounds.json")
if not fileExists(side): return
for r in parseFile(side)["rounds"]:
result.add r["startTick"].getInt()
# ── stats helpers ────────────────────────────────────────────────────────────
proc entropy(vals: openArray[int]): float =
## Empirical Shannon entropy (bits) over the distinct values in `vals`.
if vals.len == 0: return 0.0
var counts = initCountTable[int]()
for v in vals: counts.inc(v)
for c in counts.values:
let p = c.float / vals.len.float
result -= p * log2(p)
proc meanF(s: openArray[float]): float =
if s.len == 0: return 0.0
var t = 0.0
for v in s: t += v
t / s.len.float
proc autocorrAt(x: seq[float], k: int): float =
let n = x.len
if n < k + 2: return 0.0
let mu = meanF(x)
var denom = 0.0
for v in x: denom += (v - mu) * (v - mu)
if denom <= 0.0: return 0.0
var num = 0.0
for i in 0..<(n - k):
num += (x[i] - mu) * (x[i + k] - mu)
num / denom
proc autocorrPeak(x: seq[float], loLag, hiLag: int): tuple[r: float, lag: int] =
## Max |autocorr| over a LONG-lag window, with the lag that produced it.
## Long lags beat persistence: a periodic reversal shows a peak here, while a
## random telegraph has already decayed to ~0.
for k in loLag..hiLag:
let r = autocorrAt(x, k)
if abs(r) > abs(result.r): result = (r, k)
proc coeffVar(vals: openArray[int]): float =
## Coefficient of variation (sd/mean) of the reversal intervals.
if vals.len == 0: return 0.0
let mu = meanF(vals.mapIt(it.float))
if mu <= 0.0: return 0.0
var acc = 0.0
for v in vals: acc += (v.float - mu) * (v.float - mu)
sqrt(acc / vals.len.float) / mu
proc reversalIntervals(signs: seq[float]): seq[int] =
## Intervals (in ticks) between successive sign flips of the commanded speed.
var last = 0.0
var lastIdx = -1
for i, s in signs:
if s == 0.0: continue
if last != 0.0 and s != last:
if lastIdx >= 0: result.add i - lastIdx
lastIdx = i
last = s
proc directionEntropy(signs: seq[float]): float =
## Bernoulli entropy of the commanded direction (forward vs backward).
var fwd, bwd: int
for s in signs:
if s > 0.0: inc fwd
elif s < 0.0: inc bwd
if fwd + bwd == 0: return 0.0
entropy(@[fwd, bwd])
# ── replays ──────────────────────────────────────────────────────────────────
type
Replay = object
signs: seq[float] ## commanded speed sign per tick
turn: seq[float] ## turnRate per tick
absSpeed: seq[float] ## |speed| per tick
revIdx: seq[int] ## tick indices of the reversals
# STRAFE-only
picks: int
pickIntervals: seq[int]
safeCounts: seq[int]
candCounts: seq[int]
fallbackPicks: int
lineDirFlips: int
proc replayTfil(states: seq[WorldState], starts: seq[int]): Replay =
randomize(Seed)
var m = initTFIL()
var prevSign = 0.0
for i in 0..<states.len:
if i == 0 or i in starts: m.resetRound()
let cmd = m.computeMove(states[i])
let sign = if cmd.speed > 0.0: 1.0 elif cmd.speed < 0.0: -1.0 else: 0.0
result.signs.add sign
result.turn.add cmd.turnRate
result.absSpeed.add abs(cmd.speed)
if sign != 0.0 and prevSign != 0.0 and sign != prevSign:
result.revIdx.add i
if sign != 0.0: prevSign = sign
proc replayStrafe(states: seq[WorldState], starts: seq[int]): Replay =
randomize(Seed)
var m = initStrafe()
var prevSign = 0.0
var prevPicks = 0
var prevLastPick = 0
var prevFlips = 0
for i in 0..<states.len:
if i == 0 or i in starts:
m.resetRound()
prevPicks = 0
prevLastPick = i
prevFlips = 0
let cmd = m.computeMove(states[i])
let sign = if cmd.speed > 0.0: 1.0 elif cmd.speed < 0.0: -1.0 else: 0.0
result.signs.add sign
result.turn.add cmd.turnRate
result.absSpeed.add abs(cmd.speed)
if sign != 0.0 and prevSign != 0.0 and sign != prevSign:
result.revIdx.add i
if sign != 0.0: prevSign = sign
if m.picks > prevPicks:
inc result.picks
result.safeCounts.add m.lastSafeCount
result.candCounts.add m.lastCandCount
if m.lastSafeCount == 0: inc result.fallbackPicks
result.pickIntervals.add i - prevLastPick
prevLastPick = i
prevPicks = m.picks
if m.lineDirFlips > prevFlips:
result.lineDirFlips += m.lineDirFlips - prevFlips
prevFlips = m.lineDirFlips
proc reversalSpeedProfile(rep: Replay, offsets: seq[int]): seq[float] =
result = newSeq[float](offsets.len)
var n = newSeq[int](offsets.len)
for i in rep.revIdx:
for oi, off in offsets:
let j = i + off
if j >= 0 and j < rep.absSpeed.len:
result[oi] += rep.absSpeed[j]
inc n[oi]
for oi in 0..<offsets.len:
if n[oi] > 0: result[oi] /= n[oi].float
else: result[oi] = NaN
proc fmtF(x: float, d = 3): string =
if x.classify == fcNan: "-"
else: formatFloat(x, ffDecimal, d)
# ── driver ───────────────────────────────────────────────────────────────────
let states = loadStates()
let starts = loadRoundStarts()
let nTicks = states.len
let tfil = replayTfil(states, starts)
let sf = replayStrafe(states, starts)
# Field variant: the ring mover's retune, where no SINGLE soft source can poison
# a path (corridor below the threshold, wall radiance == the threshold). This is
# the same field shape the user's ring experiment used; we only test whether it
# unblocks the strafe line.
StrafeCorridorHeat = 5.0
StrafeWallHotness = 10.0
StrafeWallRadiance = 5.0
let sfRetune = replayStrafe(states, starts)
loadStrafeHeatEnv() # restore the shipped field for any later use
echo "STRAFE gates — offline fixture ", fixtureRel, " (", nTicks, " ticks, ",
starts.len, " rounds), seed=", Seed
echo ""
# ── GATE A ───────────────────────────────────────────────────────────────────
echo "=== GATE A — TILE AVAILABILITY (every STRAFE pick) ==="
proc gateA(tag: string, rep: Replay) =
let picks = rep.picks
echo "--- field: ", tag, " ---"
echo " picks total : ", picks
if picks == 0: return
echo " picks with ZERO safe tile: ", rep.fallbackPicks, " (",
fmtF(100.0 * rep.fallbackPicks.float / picks.float, 1), "%)"
echo " mean candidates on line : ",
fmtF(meanF(rep.candCounts.mapIt(it.float)), 2)
echo " mean SAFE candidates : ",
fmtF(meanF(rep.safeCounts.mapIt(it.float)), 2)
var anySafe = 0
for s in rep.safeCounts:
if s > 0: inc anySafe
echo " line has a SAFE tile at : ",
fmtF(100.0 * anySafe.float / picks.float, 1), "% of picks"
var hist = initCountTable[int]()
for s in rep.safeCounts: hist.inc(min(s, 12))
var line = ""
for k in 0..12:
line.add &"{k}:{hist.getOrDefault(k,0)} "
echo " safe-count dist (0..12+) : ", line
gateA("shipped TFIL heat (corridor 20, wall 30/10)", sf)
gateA("ring retune (corridor 5, wall 10/5)", sfRetune)
echo ""
# ── GATE B ───────────────────────────────────────────────────────────────────
echo "=== GATE B — PREDICTABILITY (STRAFE vs shipped TFIL) ==="
let Offsets = @[-6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
proc reportMover(name: string, rep: Replay, strafeExtra: bool) =
let revInt = reversalIntervals(rep.signs)
let revEnt = entropy(revInt)
let dirEnt = directionEntropy(rep.signs)
var nzTurn = 0
for t in rep.turn:
if abs(t) < 0.5: inc nzTurn
let prof = reversalSpeedProfile(rep, Offsets)
echo "--- ", name, " ---"
echo " reversals : ", rep.revIdx.len
echo " mean reversal interval : ", fmtF(meanF(revInt.mapIt(it.float)), 2), " ticks"
echo " reversal-interval CV : ", fmtF(coeffVar(revInt), 3)
echo " REVERSAL-INTERVAL ENTROPY: ", fmtF(revEnt, 3), " bits"
echo " DIRECTION ENTROPY : ", fmtF(dirEnt, 3), " bits"
echo " autocorr lag-1 : ", fmtF(autocorrAt(rep.signs,1), 3), " (persistence)"
let pk = autocorrPeak(rep.signs, 20, 60)
echo " max |autocorr| (20..60) : ", fmtF(pk.r, 3), " at lag ", pk.lag, " (periodicity)"
echo " mean |turnRate| : ", fmtF(meanF(rep.turn.mapIt(abs(it))), 2), " deg/tick"
echo " ticks with no turn (<0.5): ", fmtF(100.0 * nzTurn.float / rep.turn.len.float, 1), "%"
echo " mean |speed| : ", fmtF(meanF(rep.absSpeed), 2), " px/tick"
if strafeExtra:
echo " pick interval entropy : ", fmtF(entropy(rep.pickIntervals), 3), " bits"
echo " mean pick interval : ", fmtF(meanF(rep.pickIntervals.mapIt(it.float)), 2), " ticks"
echo " line-axis orientation flips: ", rep.lineDirFlips
var cells = ""
for i in 0..<Offsets.len:
if i > 0: cells.add " "
cells.add &"{Offsets[i]:+d}:{fmtF(prof[i],1)}"
echo " |speed| around reversal : ", cells
echo ""
reportMover("STRAFE", sf, true)
reportMover("TFIL (shipped default)", tfil, false)
# ── honest read ──────────────────────────────────────────────────────────────
echo "=== HONEST READ ==="
let revIntS = reversalIntervals(sf.signs)
let revIntT = reversalIntervals(tfil.signs)
let revEntS = entropy(revIntS)
let revEntT = entropy(revIntT)
let dirEntS = directionEntropy(sf.signs)
let dirEntT = directionEntropy(tfil.signs)
echo "reversal-interval entropy: STRAFE ", fmtF(revEntS,2), " vs TFIL ", fmtF(revEntT,2), " bits"
echo "direction entropy : STRAFE ", fmtF(dirEntS,2), " vs TFIL ", fmtF(dirEntT,2), " bits"
echo "max |autocorr| (20..60) : STRAFE ", fmtF(autocorrPeak(sf.signs,20,60).r,2),
" at lag ", autocorrPeak(sf.signs,20,60).lag,
" vs TFIL ", fmtF(autocorrPeak(tfil.signs,20,60).r,2),
" at lag ", autocorrPeak(tfil.signs,20,60).lag
echo "mean |turnRate| : STRAFE ", fmtF(meanF(sf.turn.mapIt(abs(it))),2),
" vs TFIL ", fmtF(meanF(tfil.turn.mapIt(abs(it))),2), " deg/tick"
echo "mean |speed| : STRAFE ", fmtF(meanF(sf.absSpeed),2),
" vs TFIL ", fmtF(meanF(tfil.absSpeed),2), " px/tick"