fix(training): correct non-forward penalty math and add a grounding incentive

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.

training/generation5.py

@@ -51,7 +51,13 @@ STANDING_ARGS = ["--ent-coef", "0.01", "--entropy-floor"]
 # full sideways episode now costs ~45, comparable to a goal) and
 # non_forward_penalty is a new term (ship_ai_controller.gd) directly costing
 # sideways/reverse planar velocity near the floor, independent of the ball,
-# since nothing previously penalized that at all.
+# since nothing previously penalized that at all. Both are floor-proximity
+# penalties only, with nothing equivalent above GROUND_HANDLING_HEIGHT — on
+# its own that risks teaching "avoid the floor" instead of "handle well on
+# it", worsening Stage 3's already-airborne-heavy baseline. grounded_upright_
+# reward is the positive counterpart: a bonus for genuine floor contact
+# (not just low altitude) while upright, so grounding well is the locally
+# profitable choice rather than merely the least-punished one.
 HANDLING_REWARD_FLAGS = [
     "--velocity-to-ball-weight", "0.04",
     "--forward-velocity-to-ball-weight", "0.06",
@@ -63,6 +69,7 @@ HANDLING_REWARD_FLAGS = [
     "--tilt-penalty", "0.0002",
     "--ground-tilt-penalty", "0.05",
     "--non-forward-penalty", "0.04",
+    "--grounded-upright-reward", "0.015",
 ]
 
 STAGES = [

training/train.py

@@ -363,6 +363,11 @@ def parse_args():
         help="Overrides ShipAIController.non_forward_penalty (low-altitude dense cost on sideways/reverse "
         "planar velocity, independent of the ball)",
     )
+    curriculum.add_argument(
+        "--grounded-upright-reward", type=float, default=None,
+        help="Overrides ShipAIController.grounded_upright_reward (dense bonus for genuine floor contact "
+        "while upright, countering an incentive to just avoid the floor)",
+    )
     curriculum.add_argument(
         "--speed-reward-weight", type=float, default=None,
         help="Overrides the orientation-agnostic own-speed reward (generation 5 handling sets it to zero)",
@@ -397,6 +402,7 @@ def _curriculum_kwargs(args) -> dict:
         "ai_tilt_penalty": args.tilt_penalty,
         "ai_ground_tilt_penalty": args.ground_tilt_penalty,
         "ai_non_forward_penalty": args.non_forward_penalty,
+        "ai_grounded_upright_reward": args.grounded_upright_reward,
         "ai_velocity_to_ball_weight": args.velocity_to_ball_weight,
         "ai_forward_velocity_to_ball_weight": args.forward_velocity_to_ball_weight,
         "ai_ball_distance_penalty": args.ball_distance_penalty,