feat(training): add generation 5 curriculum

training/train.py

@@ -62,8 +62,7 @@ class GoalRateCallback(BaseCallback):
 
 
 class FlightTelemetryCallback(BaseCallback):
-    """Logs rollout/{airborne_fraction,mean_altitude,air_touch_fraction,
-    vertical_thrust_mean} — leading indicators for curriculum generation 4's
+    """Logs flight and handling telemetry — leading indicators for curriculum generation 4's
     core hypothesis (a discrete action space lets the policy actually hold a
     sustained vertical set-point, e.g. hovering), visible from the very
     first rollout instead of only in a win-rate number measured a full
@@ -72,7 +71,15 @@ class FlightTelemetryCallback(BaseCallback):
     "airborne_fraction", "mean_altitude", "air_touch_fraction",
     "vertical_thrust_mean")) — see ShipAIController.get_info."""
 
-    _KEYS = ("airborne_fraction", "mean_altitude", "air_touch_fraction", "vertical_thrust_mean")
+    _KEYS = (
+        "airborne_fraction",
+        "mean_altitude",
+        "air_touch_fraction",
+        "vertical_thrust_mean",
+        "productive_air_touch_fraction",
+        "upright_fraction",
+        "forward_motion_fraction",
+    )
 
     def _on_step(self) -> bool:
         return True
@@ -287,13 +294,18 @@ def parse_args():
     )
     curriculum.add_argument(
         "--opponent-mode",
-        choices=["self_play", "inert", "frozen"],
+        choices=["self_play", "inert", "frozen", "league"],
         default=None,
         help="self_play (default): both ships are live trainees. inert: team 1 is a "
         "do-nothing placeholder (isolated scoring practice). frozen: team 1 runs a "
-        "fixed exported policy (--opponent-model)",
+        "fixed exported policy (--opponent-model); league: sample a fixed policy per episode "
+        "from --opponent-pool",
     )
     curriculum.add_argument("--opponent-model", default=None, help="Exported policy .json for --opponent-mode=frozen")
+    curriculum.add_argument(
+        "--opponent-pool", default=None,
+        help="Comma-separated exported policy paths for --opponent-mode=league; one is sampled per episode",
+    )
     curriculum.add_argument(
         "--draw-penalty", type=float, default=None, help="One-time penalty when an episode times out with no goal"
     )
@@ -310,6 +322,14 @@ def parse_args():
         help="Overrides air_drill_chance: ball spawned high, both ships spawned low and lateral — "
         "unsolvable without climbing (curriculum generation 4's state-setter aerial curriculum)",
     )
+    curriculum.add_argument(
+        "--air-intercept-chance", type=float, default=None,
+        help="Moving high-ball interception starts aimed at a real goal (generation-5 aerial stage)",
+    )
+    curriculum.add_argument(
+        "--team-size", type=int, choices=range(1, 6), default=None,
+        help="Ships per team (1-5); generation-5 automated stages remain 1v1 until 2v2 evaluation exists",
+    )
     curriculum.add_argument(
         "--tilt-penalty", type=float, default=None,
         help="Overrides ShipAIController.tilt_penalty (dense per-tick cost scaled by non-upright tilt)",
@@ -318,6 +338,10 @@ def parse_args():
         "--velocity-to-ball-weight", type=float, default=None,
         help="Overrides ShipAIController.velocity_to_ball_weight (dense reward for closing speed toward the ball)",
     )
+    curriculum.add_argument(
+        "--forward-velocity-to-ball-weight", type=float, default=None,
+        help="Low-altitude dense reward for nose-led planar approach toward the ball",
+    )
     curriculum.add_argument(
         "--ball-distance-penalty", type=float, default=None,
         help="Overrides ShipAIController.ball_distance_penalty (dense per-tick cost scaled by distance to the ball)",
@@ -330,6 +354,14 @@ def parse_args():
         "--airborne-penalty", type=float, default=None,
         help="Overrides ShipAIController.airborne_penalty (dense per-tick cost scaled by height above the floor)",
     )
+    curriculum.add_argument(
+        "--ground-tilt-penalty", type=float, default=None,
+        help="Low-altitude-only tilt cost that fades to zero by the handling-height threshold",
+    )
+    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)",
+    )
     curriculum.add_argument(
         "--ball-velocity-to-goal-weight", type=float, default=None,
         help="Overrides ShipAIController.ball_velocity_to_goal_weight (dense reward for the ball's velocity toward the attack goal)",
@@ -349,16 +381,22 @@ def _curriculum_kwargs(args) -> dict:
     mapping = {
         "opponent_mode": args.opponent_mode,
         "opponent_model": args.opponent_model,
+        "opponent_model_pool": args.opponent_pool,
         "draw_penalty": args.draw_penalty,
         "attack_goal_bias": args.attack_goal_bias,
         "kickoff_state_chance": args.kickoff_chance,
         "ball_near_goal_chance": args.near_goal_chance,
         "air_drill_chance": args.air_drill_chance,
+        "air_intercept_chance": args.air_intercept_chance,
+        "team_size": args.team_size,
         "ai_tilt_penalty": args.tilt_penalty,
+        "ai_ground_tilt_penalty": args.ground_tilt_penalty,
         "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,
         "ai_ball_touch_reward": args.ball_touch_reward,
         "ai_airborne_penalty": args.airborne_penalty,
+        "ai_speed_reward_weight": args.speed_reward_weight,
         "ai_ball_velocity_to_goal_weight": args.ball_velocity_to_goal_weight,
         "goal_reward": args.goal_reward,
     }
@@ -395,6 +433,9 @@ def main():
             "mean_altitude",
             "air_touch_fraction",
             "vertical_thrust_mean",
+            "productive_air_touch_fraction",
+            "upright_fraction",
+            "forward_motion_fraction",
         ),
     )