feat(*): Add a distance-to-ball penalty so idling scores negative, boost ball-seeking rewards, and halve the wall-contact penalty

Game/scripts/ship_ai_controller.gd

@@ -15,15 +15,23 @@ extends AIController3D
 # Reward shaping weights. Dense terms accrue per physics tick (60 sim-ticks
 # per sim-second); event terms fire once. Exported so tuning needs no code
 # edits. Goal rewards are added by TrainingMode, which owns goal events.
-@export var ball_touch_reward := 0.25
-@export var velocity_to_ball_weight := 0.002
+@export var ball_touch_reward := 1.0
+@export var velocity_to_ball_weight := 0.02
 @export var ball_velocity_to_goal_weight := 0.004
+# Per-tick penalty scaled by distance to the ball (full value at the arena's
+# far diagonal, 0 on top of the ball). Run04 lesson: with idling worth a flat
+# 0, camping in a corner strictly dominated risking the wall/tilt penalties
+# to chase the ball — this makes "do nothing far from the ball" the worst
+# option instead of the safest. A penalty, not a proximity bonus, so orbiting
+# the ball farms nothing.
+@export var ball_distance_penalty := 0.002
 # Per-tick penalty while pressed against a side wall, end wall, or the
 # ceiling — NOT the floor (run03 lesson: taxing floor contact punishes the
 # ship's natural low flight and drowns every other signal). At 60 ticks per
-# sim-second this is -0.3/s: parked on a wall for a full 30 s episode loses
-# ~9 — comparable to conceding — while a brief graze costs almost nothing.
-@export var wall_contact_penalty := 0.005
+# sim-second this is -0.15/s. Halved for run05: the ball lives near walls,
+# and the old -0.3/s made the productive region of the pitch aversive
+# relative to the (then far weaker) ball-seeking shaping.
+@export var wall_contact_penalty := 0.0025
 # Per-tick penalty for not being upright, scaled by tilt: 0 when flat, full
 # value (-0.12/s) when inverted. A penalty rather than an upright bonus so a
 # flat, idle ship farms nothing.
@@ -33,6 +41,14 @@ extends AIController3D
 # penalty); below it they read as wall (sideways) or ceiling (downward).
 const FLOOR_NORMAL_MIN_Y := 0.7
 
+# Longest possible ship-to-ball separation: the enclosure's interior diagonal.
+# Normalizes ball_distance_penalty so its export is the worst-case per-tick cost.
+const MAX_BALL_DISTANCE := sqrt(
+	(2.0 * ArenaBoundary.INNER_HALF_X) ** 2
+	+ (2.0 * ArenaBoundary.INNER_HALF_Z) ** 2
+	+ ArenaBoundary.INNER_HEIGHT ** 2
+)
+
 var ship: Ship
 var rl_controller: RLShipController
 var ball: RigidBody3D
@@ -92,6 +108,11 @@ func _physics_process(delta):
 		var closing_speed := ship.linear_velocity.dot(to_ball.normalized())
 		reward += velocity_to_ball_weight * closing_speed / ship.max_speed
 
+	# Dense penalty: distance to the ball, so idling far away bleeds reward
+	# instead of scoring a safe zero (see ball_distance_penalty).
+	if ball_distance_penalty > 0.0:
+		reward -= ball_distance_penalty * to_ball.length() / MAX_BALL_DISTANCE
+
 	# Dense shaping: ball velocity toward the goal we attack
 	var ball_to_goal := attack_goal_position - ball.global_position
 	if ball_to_goal.length_squared() > 0.0001: