20260816-2126-gen5-s4-handling-retry2 exhausted its three attempts and
missed only the 0.80 training goal-rate floor, at 0.7731. Every
evaluation gate passed: 65-22-13 versus promoted/easy.json, 87% non-draw
against an 80% floor, 12.6% physical-side imbalance against a 20%
ceiling, and both handling telemetry floors clear. The round improved the
goal rate monotonically across attempts (0.537 -> 0.683 -> 0.773) and the
checkpoint plays well by hand, so close Stage 4 by human override.
Promote it to Game/bots/promoted/medium.json. Medium and Hard both point
at the new policy: Hard stays a label-only duplicate until a stronger one
earns hard.json, which keeps the tiers monotonic rather than leaving Hard
weaker than Medium.
generation5_state.json flips that log entry to "pass" with a
decision_override block preserving the original verdict and reasoning,
and advances to Stage 5 attempt 1. This is what passing_entry() needs to
resolve Stage 5's resume checkpoint and evaluation reference, and what
league_pool() will need at Stage 6; --skip-to-next-stage would advance
the stage without marking anything as passing and die immediately.
generation5.sh now pulls before launching. Each stage ends in
commit_progress()'s push, which fails and kills the run hours in if the
box is behind origin.
Six rounds of reward shaping (~700M steps) failed to produce upright ground
driving. A critical review of the simulation rather than the reward found
why:
1. The hull was a 1x1x4 box with inertia (1,1,1) and no restoring torque
anywhere, so belly-down and rolled-90 were geometrically identical
resting states. "Upright" was not a physically distinguished state at
all - the reward was paying for a property the simulation did not have.
2. ~65% of episodes spawned ships via _random_position, which samples Y
uniformly over the full 18m volume (mean ~8.7m). The measured
airborne_fraction ~0.44 was largely that spawn distribution, and every
ground-handling term fades out above 3m, so the shaping being tuned
barely ever applied.
3. air_drill_chance 0.20 spawned deliberately unreachable-without-climbing
states in the stage meant to teach ground driving, and its own
air_touch_fraction (0.0002) shows the drills were never solved.
Fixes land in the physics and the task distribution, not the reward:
- ship.tscn: hull 1x1x4 -> 1.6x0.6x4 so it has one stable resting face;
inertia (1,1,1) -> (7,1,7), physically correct for the hull, making
tumbling reluctant while keeping yaw snappy.
- ship.gd: new altitude-faded righting torque (spring-damper toward
belly-down, faded out by 3m so aerials keep full attitude freedom).
This is the grav-plating analogue of Rocket League's auto-righting and
helps human pilots land cleanly too.
- training_mode.gd: new ground_start_chance branch spawning ships level and
resting on the floor with a floor-level ball - the state the handling
stage's rewards are actually written for.
- generation5.py: ground-start-chance 0.50, air-drill-chance 0.20 -> 0.0.
Reward terms are left exactly as they were; they should finally pull in a
direction the ship can go.
Round 4 changed three things at once and two of them cut upright pressure:
grounded_upright_reward went to 0 and ground_tilt_penalty was cut 2.5x,
while the new uprightness multiplier only pays below GROUND_HANDLING_HEIGHT
*and* while moving forward *and* facing the ball - a far narrower slice of
ticks than the penalty it was meant to replace. Net pressure fell and
upright_fraction fell with it (0.268 -> 0.239 -> 0.238, the lowest of any
round). Restore ground_tilt_penalty to 0.05 and change nothing else, so
this is a genuine single-variable test of multiplier plus full tilt
pressure.
The conjunctive mechanism itself held up: forward_motion_fraction reached
its best sustained value (0.242) without goal_rate sagging, ep_rew_mean
turned positive for the first time (+0.28), and eval win rate hit 49% with
no reward hacking.
Also adds grounded_upright_fraction: a diagnostic, deliberately ungated
metric measuring uprightness over real floor-contact ticks instead of
sub-3m ticks. upright_fraction has never exceeded 0.331 across four rounds
and ~560M steps without cheating, and its denominator is dominated by
ballistic transit (airborne_fraction ~0.45, mean_altitude ~4.4m) where
attitude is not meaningfully controllable - so it likely cannot measure
what the 0.45 floor was meant to capture. Re-baseline that floor from what
this reports rather than from another round of reshaping.
Rounds 2 and 3 showed that tuning grounded_upright_reward's magnitude only
slides along a tradeoff instead of resolving it: at 0.015 upright_fraction
climbed to 0.331 while goal_rate sagged to 0.542 (then farmed outright at
0.696/0.366), and at 0.004 goal_rate climbed 0.569->0.604 while
upright_fraction went flat at ~0.26. An additive uprightness bonus is an
alternative to playing well, so the policy just picks whichever is cheaper
and no magnitude buys both behaviours.
Change the mechanism rather than the number: grounded_upright_reward drops
to 0, and uprightness becomes a multiplier inside the nose-led approach
term, which already requires moving forward at the ball. Parked-and-upright
and fast-but-sideways now both pay zero; only upright, forward, nose-on to
the ball pays full. forward-velocity-to-ball rises 0.06 -> 0.15 to offset
the ~2-3x expected-value cut from the new factor, and ground-tilt-penalty
drops 0.05 -> 0.02 now that uprightness is paid positively during play.
Delete the three blocked attempts and reset state to restart from the
Stage-3 foundation.
grounded_upright_reward at 0.015 overshot: four force-retries pushed
upright_fraction from 0.265 to a plateauing 0.331, then the fifth jumped it
to 0.696 (55% over the 0.45 floor) while goal_rate collapsed 0.542->0.366
and forward_motion_fraction fell 0.244->0.184 (vertical_thrust_mean went
negative) - the policy learned to sit pinned upright and farm the bonus
instead of chasing the ball. It was sized "comparable to
time_penalty/ball_distance_penalty" but at 0.015/tick it was actually above
ball_distance_penalty's 0.01/tick worst case, so idling near the ball beat
playing. Cut to 0.004/tick (episode ceiling ~7.2, below
ball_distance_penalty's ~18 worst case). Delete the five blocked attempts
and reset generation5_state.json so the next run starts fresh from the
Stage-3 foundation rather than continuing from the farming checkpoint.
Adversarial review of the previous stage-4 retune found two problems:
non_forward_speed used planar_speed - forward_component, which under-charges
diagonal motion relative to true lateral speed (e.g. ~29% penalty at 45
degrees off the nose instead of the correct ~71%); fixed to the Pythagorean
magnitude for forward-facing angles, full speed for backward-facing ones.
Also, ground_tilt_penalty and non_forward_penalty only ever cost reward near
the floor with nothing offsetting them above it, which could teach a policy
that's still bad at ground handling to just avoid the floor rather than get
better at it. Added grounded_upright_reward (ship_ai_controller.gd) plus a
new ShipObservations.is_floor_contact helper for genuine belly-on-floor
contact detection, so grounding well while upright is the locally profitable
choice, not just the least-punished one.
Stage 4's upright/forward-motion telemetry plateaued flat across all three
blocked attempts because ground_tilt_penalty (0.003) was too weak to matter
and nothing penalized sideways/reverse motion at all. Raise
ground_tilt_penalty to 0.05 and add a new non_forward_penalty term
(ship_ai_controller.gd) that directly costs non-forward planar velocity near
the floor, independent of the ball. Delete the three blocked attempts'
checkpoints/logs/exports and reset generation5_state.json so the next run
starts fresh from the Stage-3 foundation checkpoint instead of continuing
from the drifted retry2 weights.
Extends ShipObservations beyond the old self+1-opponent layout to padded
teammate/opponent arrays (MAX_TEAMMATES=4, MAX_OPPONENTS=5, SIZE=83),
zero-filling slots past the real roster size the same way the old single-
opponent slot was zero-filled when absent.
Slot stability across ticks requires a persistent identity: Ship gains
spawn_index (set once by GameMode.spawn_ship, never reassigned — there's no
despawn path anywhere in this codebase, so a roster is fixed for the whole
episode/match). ai_ship_controller.gd's opponent discovery is rewritten from
"first non-self ship" to classify every other ship by team and sort by
spawn_index; training_mode.gd/ship_ai_controller.gd carry the equivalent
sorted lists through the training path so both agree on slot assignment for
the same roster.
training_mode.gd and match_mode.gd both gain a team_size export (default 1,
so every existing curriculum script and match keeps today's 1v1 behaviour
unchanged). This is plumbing only: no 2v2+ curriculum or reward design, and
no match-mode UI to pick team size, has been done yet. The two checkpoints
in Game/bots/promoted/ are fitted to the old 35-float layout and are not
migrated — expected to go stale until the next training run.
Retraining from scratch removes the constraint that blocked these: the
existing checkpoints in Game/bots/promoted/ no longer need byte-identical
geometry.
- Fix the goal aperture constants (were 1.85/1.65, now match
objects/goal.tscn's actual sensor exactly at 1.75/1.5) so a ball crossing
the visible edge can't fail to score.
- Cut a real navigable hole in each end wall's collision
(_build_end_wall_colliders), replacing the previous solid box — the ball
now genuinely enters the net instead of triggering the sensor a hair
before hitting a solid wall. Correct for both FLOOR and ELEVATED goal
modes via _goal_surround_bounds. A separate, slightly wider
GOAL_VISUAL_APERTURE_* pair keeps the collision hole exact while still
giving goal.gd's bezel/rim frame clearance to be seen against the hull cut.
- Resize the play volume 1.5x (INNER_HALF_X/Z/HEIGHT 12/18/12 -> 18/27/18) to
comfortably fit a 5v5 roster: updates every hand-authored literal in
arena_boundary.tscn/arena_base.tscn/the elevated arena variants that
doesn't derive from those constants, adds 5 spawn markers per team, and
rescales ship_observations.gd's normalization scales and arena_02's
cosmetic decoration/nebula-dust shader uniforms to match.
- Bake the ~170 runtime-generated CollisionShape3D nodes into the scene via
a new bake_colliders()/tools/bake_arena_boundary.gd instead of rebuilding
them on every load — a real load-time cost repeated in every parallel
headless training env. Colliders must be direct children of the
StaticBody3D to register at all, so bake_colliders() parents them onto
self and tags them with a group for idempotent re-baking, rather than
grouping them under an intermediate container node. _ready() self-heals:
it skips regenerating only when an existing bake's goal_mode metadata
matches the current one, so the FLOOR-mode bake in the shared scene is
never silently reused by an ELEVATED arena variant.
_place_ships_random/_place_air_drill sampled each ship's randomized episode-
start position independently, so a team_size > 1 roster could spawn
interpenetrating (ships are ~1x1x4). Both now resample (up to 20 attempts,
matching the existing corner/fillet rejection-sampling pattern) against
every ship already placed that reset, rejecting anything within
MIN_SHIP_SEPARATION (4.5m, matching the arena spawn-marker spacing) of one.
The goal mouths are now a real navigable hole in the end walls, sized to the
ball rather than the ship — a ship's 1x1 cross-section fits through it, and
there's nothing behind the net to stop it. The escape failsafe previously
only existed in TrainingMode (where a physics regression just wastes
training time); now that any ship can genuinely fly out through an open
goal, every mode needs it or a stray ship/ball falls into the void with no
way back short of quitting. Moved up to GameMode as the shared default
_physics_process, removing TrainingMode's now-duplicate copy.
drag_coefficient/angular_drag/the idle angular-drag multiplier were applied
once per physics tick with no delta scaling, correct only because
project.godot never pins physics/common/physics_ticks_per_second and
Godot's default happens to be 60. _tick_scaled(k, state.step) makes the
decay rate invariant to tick rate instead. Also promotes the previously
hardcoded 0.9 idle angular-drag literal to an export, matching its sibling.
VecMonitor's info_keywords does a bare info[key] lookup on a completed
episode's terminal info dict and raises KeyError -- crashing the whole
training run -- if a key is ever absent. get_info() was only including
airborne_fraction/mean_altitude/vertical_thrust_mean when _telemetry_ticks
was nonzero and air_touch_fraction when _touches was nonzero; an episode
with zero ball touches (common, especially early in training) crashed on
the very first rollout in a smoke-test run. All four now always default to
0.0 rather than being conditionally present.
Found via TRAINING.md's Generation 4 validation ladder (rung 2, a 60k-step
smoke run) -- confirmed fixed by rerunning the same smoke run clean, then
export/evaluate parity (rungs 2-3) against Game/bots/promoted/easy.json.
Three curriculum generations (2026-07-21 through 2026-08-04) all tried
gating *when* the policy could use vertical thrust/pitch-roll on top of a
continuous Gaussian action space, and all three failed the same way: PPO's
action-distribution std collapsed within ~10% of steps and never recovered,
landing at a 15-32% win rate vs the grounded reference regardless of
mechanism (hard mask, then a gradual ramp). Generation 3's final attempt
just landed at 24% — the worst of the three.
Root cause, verified against this project's own physics: hovering this ship
requires *holding* thrust.y ~= 0.408 continuously (mass 5.0, vertical_thrust
120, gravity 9.8). A collapsed near-zero-mean Gaussian can brush that value
but never sustain it long enough to earn the reward gradient that would
move the mean — no amount of gating *when* the axis acts fixes a problem in
*how* the policy represents a decision on it. This also independently found
and fixes a real bug: godot_rl never marks an episode timeout as a
truncation, so PPO was bootstrapping V(s)=0 on every 30s draw in every
generation to date.
- Game/scripts/ship_action_codec.gd (new): single source of truth for a
per-axis MultiDiscrete action space (7 heads, nvec [5,5,5,5,5,5,2]) shared
by training and in-game inference, replacing the continuous Gaussian.
thrust_y's bins are deliberately asymmetric so a random policy drifts
through the volume instead of floor-pinning. Legacy continuous decode
(ai_ship_controller.gd's old logic) preserved verbatim so every
pre-generation-4 export (e.g. Game/bots/promoted/easy.json) keeps working
unchanged via an optional "action_space" JSON field.
- ship_observations.gd: append own contact state (SIZE 31 -> 35, append-only)
so the value function can see what wall_contact_penalty fires on.
- ship_ai_controller.gd: action space/decode via the codec; drop the
vertical_ramp/pitch_roll_ramp mechanism entirely; tilt_penalty default
lowered 4x (aerial approaches require pitching); flight telemetry
(airborne_fraction, mean_altitude, air_touch_fraction, vertical_thrust_mean)
and truncation-snapshot fields on get_info().
- training_mode.gd: new air_drill_chance state-setter branch (ball spawned
high, ships low, kept clear of walls) so aerial practice is forced by the
environment instead of relying on reward-driven exploration alone; snapshot
terminal observations before a timeout reset for the truncation fix.
- cosmic_env.py: remap ShipAIController's truncated/terminal_obs info into
SB3's TimeLimit.truncated/terminal_observation keys.
- train.py: --reset-logits (+ --reset-logits-heads) replaces the
now-meaningless --reset-std; new EntropyFloorCallback (a persistent
per-rollout ent_coef controller replacing the one-shot std-reset shock)
and per-head entropy logging; FlightTelemetryCallback; --air-drill-chance/
--tilt-penalty flags; optional AbortIfCallback kill-criterion.
- export_policy.py: writes the action_space block for MultiDiscrete models;
index-level parity check (argmax per head) instead of comparing floats.
- curriculum.py: full rewrite — 3 stages (bootstrap/selfplay/gauntlet), no
grounded stage, full action space live from step 1; deletes generation
1-3's checkpoint-lineage machinery (nothing to resume from); final report
evaluates against both promoted/easy.json and the new
promoted/reference-grounded.json (a copy of curric-s5-aggression, the
strongest grounded-era artifact, kept as a fixed yardstick).
- run_training.sh/.gitignore: commit only final.zip, not the ~2400
intermediate checkpoint files a single stage was writing (~500MB ->
~0.2MB per run); requirements.txt pinned (behaviour here now depends on
specific library internals, not just public APIs).
- test_action_space.py (new): offline rung-0 check catching a head-order
mismatch before it silently corrupts 24h of training.
Validated: GDScript compiles clean (Godot --headless --import + script
validation), free_play.tscn and training.tscn both boot headless without
errors, offline action-space assertions pass. Not yet run: the actual
smoke-training/A-B validation ladder steps in TRAINING.md's "Generation 4"
section, before committing to the full ~32h curriculum.
See TRAINING.md's "Generation 4" section for the full design writeup.
ship.gd's controllerless path and player_ship_controller.gd each
allocated a fresh ShipAction every physics tick; ai_ship_controller.gd
and rl_ship_controller.gd already avoid this via a persistent member.
Convert both to reuse a member instance, matching the existing
full-field-overwrite convention (rather than +=/-= off a fresh zero).
Also drop the completed items from TODO.md.
HUDController found its ship via get_first_node_in_group("ship"), a group
that has 2+ members once a match has an AI opponent — it only worked
because the player ship happened to spawn first. spawn_camera_rig now
wires the HUD's ship the same way it already wires the camera rig's
target.
match_mode.gd fired the signal every frame while only the once-a-second
value is displayed, forcing the HUD to re-format and re-shape the label
each frame. Gate emission on the whole-second value changing, matching
ship.gd's threshold-gated telemetry discipline.
The coroutine stalls mid-countdown while the tree is paused for the
results screen, but _end_match unpauses before the deferred scene
change actually tears things down — letting it resume for a frame and
re-emit kickoff_countdown / unfreeze bodies in the dying scene.
ship.gd, HUDController.gd, and goal.gd each declared their own TEAM_COLORS,
arena_boundary.gd/arena_deck.gdshader had a third pair, and HUD.tscn baked
in a fourth (hardcoded "BLUE"/"ORANGE" labels) — nose, goal rim, end zone,
and scoreboard all rendered different blues. New scripts/team_colors.gd
(class_name TeamColors) is now the single source every one of those reads
from, and team identity moves to purple/green.
WorldEnvironment's Environment sub-resource was shared across every
instantiate() of a cached arena PackedScene, so glow/brightness/sky
tweaks in Arena._ready() compounded further each time a player
re-entered an arena instead of applying fresh.
Hull/Canopy/EngineGlowL/EngineGlowR are runtime-baked into one ArrayMesh
in _ready via SurfaceTool.append_from, dropping 4 MeshInstance3D children
to 1 (Nose/TailFin stay separate, they're retinted per-team). Skipped in
headless mode like the goal/arena_boundary visual builds, since physics
only cares about CollisionShape3D.
All 6 source surfaces (hull, canopy, engine_l x2, engine_r x2) carry
distinct materials, so this doesn't literally cut draw calls 6 to 3 as
TODO.md assumed — Godot still issues one draw call per surface regardless
of node count. The real win is scene-tree/transform overhead, not batching.
_apply_team_color() allocated a fresh StandardMaterial3D on every call, and
ran at least twice per ship (once from _ready at the default team, once from
the team setter when the game mode assigns the real team). Cache one
StandardMaterial3D per team in a static dict on Ship and reuse it across
every ship on that team.
Goal._build_visuals() built 10 MeshInstance3D nodes across 4 materials
(1 pocket + 1 net + 4 bezel-ring + 4 rim-ring boxes) per goal. Replaced
with a single MeshInstance3D wrapping one ArrayMesh with 4 SurfaceTool-
committed surfaces (pocket, net, bezel, rim), one material per surface
via surface_set_material — 10 nodes down to 1, same 4 materials.
Kept 4 surfaces rather than collapsing further: the rim is a tuned
team-tinted emitter, the net carries its own discard shader, and the
pocket/bezel differ in albedo/metallic/roughness. Merging those into a
shared material would be a visible regression, not a free win.
Added a local box-to-SurfaceTool helper (6 quads via arena_boundary.gd's
_add_quad/_add_tri winding-correction trick, copied in rather than
shared since that file's geometry is collision-adjacent). The pocket's
old cull_mode = CULL_FRONT trick is replaced by emitting its geometry
with inverted winding; the net's cull_front stays material-driven since
goal_net.gdshader's own render_mode depends on that winding convention.
Verified in the editor: both team-tinted goals render an intact pocket,
net, bezel and glowing rim with no backface/winding artifacts, and the
headless free_play smoke test still runs clean.
_process() called get_viewport().get_camera_3d() every frame to drive
the containment field's camera-side fade. Cache it the same way
ship_camera.gd caches the ball, revalidating with is_instance_valid
since exactly one camera rig is spawned per game-mode run today.
hud_gauge, hud_attitude_indicator, and hud_heading_tape now skip
queue_redraw() when the newly-lerped value hasn't moved past a small
epsilon, instead of redrawing every frame forever. Angle-wrapping
values (heading, attitude roll) use a new shared angle_delta_deg
helper on HudInstrument so the wrap boundary doesn't read as a false
jump.
HUD.tscn's Instruments node now sets process_mode = 1 (PAUSABLE),
overriding the inherited ALWAYS mode from the HUD root so instruments
stop processing during the post-match pause freeze, while sibling
ResultOverlay keeps running its win-screen tween.
Ship._emit_telemetry_data() ran get_euler()+trig every physics tick
for every ship regardless of whether a HUD was watching, wasting work
on AI ships and every headless training instance. Disable
_physics_process outright when headless, and skip emission the rest
of the time unless a signal actually has a listener.
Arenas inherit from a new arena_base.tscn instead of restating ~40 shared
lines each; only sky/ambient/glow/tint/decoration vary, exposed via new
Arena exports since nested Environment properties aren't overridable
through scene inheritance. Bot construction and score-keeping move onto
GameMode, shared by match and spectate modes while preserving their
differing GameSettings-override behavior and the HUD's score-row
duck-typing. HUD instruments share a HudInstrument base for the
smoothing-weight calc and angle-lerp helper. Also dedupes
MAIN_MENU_SCENE_PATH into ScenePaths and documents why DIFFICULTIES
tiers share one checkpoint.
Each of the three arenas gains an ELEVATED sibling scene that inherits the
base arena and overrides the boundary's goal_mode, the two goal transforms
and the ship spawns, so the goal sits at mid-wall height instead of flush
with the deck. Registered in ArenaRegistry alongside the floor-level arenas,
plus a training_elevated scene for self-play on them.
TrainingMode reads the arena's goal_mode once in _start() and widens the
ball-placement height range to match the goal's real position; on FLOOR
arenas the bound is a no-op, so floor-level training is unchanged. It is read
in _start() rather than _ready() because TrainingMode has no _ready()
override and GameMode._ready() is what discovers the arena first.
Policies trained against floor-level goals are not expected to score on an
elevated one, so the two are kept as separate arenas rather than a variant of
the same entry.
The goal was a single flat translucent slab. Because the chase camera sits
behind the goal line at kickoff it rendered as a large blue rectangle across
the lower screen, and the hull beside the mouth was translucent field panel,
so the pocket behind it showed straight through the wall and the net appeared
a second time alongside the frame.
The boundary now carries an opaque bulkhead around each mouth, which occludes
properly. On top of that the goal becomes a pocket sunk into the wall: a dark
machined bezel lining the opening, a thin team-coloured emissive rim flush
with the wall face, and netting from a single box viewed inside-out, giving a
five-sided pocket instead of a flat panel across the back.
The net is cut with discard rather than alpha blending so it still writes
depth and sorts against the frame and the containment field like solid
geometry; a blended net would join the transparent queue and sort per object
against the boundary shell, which is the artefact this whole pass removed.
goal.tscn drops from 18 node/sub-resource blocks to 3 — it is now just the
Area3D, its CollisionShape3D and the script. All geometry is built in code
from two loops, and the mouth is measured off the collision shape rather than
restated, so the frame cannot drift from the volume that actually scores.
The sensor, its collision shape and the goal_scored signal are unchanged, and
visuals are skipped under --headless.
Every visible surface of the enclosure shared one alpha-blended material and
was drawn as several overlapping layers: floor, ceiling and four walls as
solid BoxMeshes, plus corner curves and fillets generated on top of the box
faces they eased into. Alpha-blended surfaces don't write depth and sort per
object, so the perimeter composited that tint twice and the corners three
times, giving hard-edged trapezoidal patches that re-sorted as the camera
moved. The floor box also overhung the walls by 1 m and showed through them.
Replace all of it with a single generated mesh covering the inner surface
exactly once, every piece cut to meet its neighbours edge-on and only
inward-facing triangles emitted. It carries two surfaces: an opaque hull
(deck, base fillets, goal bulkheads) and the translucent containment field
(walls, corners, ceiling fillets, ceiling).
The field shader is additive and unshaded rather than alpha-blended: additive
cannot double-darken and composites order-independently, so overlap is
structurally invisible, and being unshaded it no longer picks up per-arena
light and GI gradients across a 28x40 m panel. Fresnel replaces the old
StandardMaterial3D rim, which was a lit effect and the wrong tool. The deck
shader draws plating and field markings procedurally from the boundary's own
constants, so markings cannot drift from the collision geometry.
Per-face MeshInstance3D visibility toggling is gone; the field shader fades
facets the camera has crossed outside of per-pixel from one uniform, which is
what allows a single merged mesh.
Also adds ELEVATED goal mode and the ceiling fillets, and skips the mesh build
entirely under --headless, where training spawns instances that never render.
Collision is untouched: 166 collider shapes in FLOOR mode and 158 in ELEVATED,
verified byte-identical to before, so trained policies in Game/bots/ are
unaffected.
Enable glow/bloom, color adjustments, and MSAA+FXAA across all three
arenas so emissive ship/station accents actually bleed light and edges
read cleanly. Expose the player-facing knobs (anti-aliasing mode, glow
intensity, brightness) through a new Settings screen off the main menu,
persisted via a VideoSettings autoload that scales each arena's own
tuned Environment values on load rather than overwriting them.
Introduce arena_02 (Nebula) and arena_03 (Asteroid Field) alongside
arena_01, listed in a new ArenaRegistry (scripts/arena_registry.gd) as
the single source of truth for available arenas. Free Play lets the
player pick an arena from the main menu; Match/Spectate each pick one
at random per session; Training keeps its own fixed arena_01.
Nebula gets an RL-style visual treatment: a near-invisible glass
boundary material shared by all arenas, a baked equirect nebula sky
texture, a drifting-dust particle system, and a Blender-modeled
station/debris/planet decoration set. GameMode gains a
_get_arena_scene_path() hook so modes can instantiate their arena in
code instead of hardcoding it in the scene.
Replace the raw checkpoint dropdown with curated Easy/Medium/Hard presets
that drive GameSettings' bot model/reaction_ticks/action_noise overrides.
Move raw-checkpoint testing and Spectate mode into a dev-only section
hidden via OS.is_debug_build() so they disappear from release exports.
Generation 2's single "unmask" stage (flip vertical/pitch-roll locomotion
from grounded-only to full 3D in one step) failed 3 independent 240M-step
attempts, landing at a stable 32% / 28% / 31% win rate vs curric-s5-aggression
each time -- not noise, and not fixable by more training time (attempts 2-3
each continued the same checkpoint lineage for another full 240M steps with
zero improvement). Every attempt shows train/std collapsing from ~0.30 to
~0.13-0.15 within the first ~10% of steps and never recovering: the policy
locks the newly-opened axes back down before ever meaningfully exploring
them.
Replaces the boolean allow_vertical/allow_pitch_roll mask on ShipAIController
with float vertical_ramp/pitch_roll_ramp multipliers (0.0-1.0), scaling axis
effect in set_action() instead of gating it outright -- the action space
never changes shape, so checkpoints stay resumable across ramp values. The
single unmask stage in curriculum.py becomes 4: three ungated warmup stages
(25%/50%/75% authority, airborne_penalty ramping in step) that train,
checkpoint, and always advance with no eval gate, then the measured stage at
full authority -- same reference, opponent mode, and 240M budget as the 3
failed attempts, for a direct comparison. Adds a "gated" flag/branch to
main()'s loop for the ungated stages.
This is generation 3 of the curriculum; generation 2's state is archived to
curriculum_state_gen2.json (mirroring the earlier gen1 -> gen2 archival) and
curriculum_state.json resets fresh, since its stage 0 no longer means what it
used to. See TRAINING.md's "Generation 3" section for the full postmortem,
stage table, and the open question about whether scaling action effect in
Godot (which PPO's own entropy/exploration math never sees) actually
addresses the collapse.
Josh Creek