Merge branch 'main' of https://github.com/Kaiyotech/Opti

CoyoteObs.py

@@ -17,7 +17,7 @@
 from gym import Space
 from gym.spaces import Tuple, Box
 
-from rlgym.utils.common_values import BOOST_LOCATIONS
+from rlgym.utils.common_values import BOOST_LOCATIONS, BALL_RADIUS
 
 from numba import njit
 
@@ -76,8 +76,14 @@ def __init__(self, tick_skip=8, team_size=3, expanding: bool = True, extra_boost
                  end_object: PhysicsObject = None,
                  mask_aerial_opp=False,
                  selector_infinite_boost=None,
+                 doubletap_indicator=False,
                  ):
         super().__init__()
+        self.doubletap_indicator = doubletap_indicator
+        if self.doubletap_indicator:
+            self.floor_bounce = False
+            self.backboard_bounce = False
+            self.prev_ball_vel = np.asarray([0] * 3)
         assert not (selector_infinite_boost is not None and not selector)
         self.n = 0
         self.selector_infinite_boost = selector_infinite_boost
@@ -225,6 +231,11 @@ def reset(self, initial_state: GameState):
         if self.add_handbrake:
             self.handbrakes = np.zeros(len(initial_state.players))
 
+        if self.doubletap_indicator:
+            self.floor_bounce = False
+            self.backboard_bounce = False
+            self.prev_ball_vel = np.array(initial_state.ball.linear_velocity)
+
     def pre_step(self, state: GameState):
         # dist = state.ball.position - state.players[0].car_data.position
         # dist_norm = np.linalg.norm(dist)
@@ -253,6 +264,17 @@ def pre_step(self, state: GameState):
             state.ball.linear_velocity = np.asarray([0, 0, 0])
             state.ball.angular_velocity = np.asarray([0, 0, 0])
 
+        # for double tap
+        if self.doubletap_indicator:
+            if state.ball.position[2] < BALL_RADIUS * 2 and 0.55 * self.prev_ball_vel[2] \
+                    < state.ball.linear_velocity[2] > 0.65 * self.prev_ball_vel[2]:
+                self.floor_bounce = True
+            elif 0.55 * self.prev_ball_vel[1] < state.ball.linear_velocity[1] > 0.65 * \
+                    self.prev_ball_vel[1] and \
+                    abs(state.ball.position[1]) > 4900 and state.ball.position[2] > 500:
+                self.backboard_bounce = True
+            self.prev_ball_vel = np.array(state.ball.linear_velocity)
+
     def _update_timers(self, state: GameState):
         current_boosts = state.boost_pads
         boost_locs = self.boost_locations
@@ -714,6 +736,9 @@ def add_players_to_obs(self, obs: List, state: GameState, player: PlayerData, ba
                 ball.position, ball.linear_velocity, prev_act
             )
 
+        if self.doubletap_indicator:
+            player_data.extend(list([int(self.backboard_bounce), int(self.floor_bounce)]))
+
         if self.stack_size != 0:
             if self.selector:
                 self.model_add_action_to_stack(

learner_demo.py

@@ -88,6 +88,8 @@
                                                               velocity_po_w=0.001,
                                                               vel_po_mult_ss=5,
                                                               vel_po_mult_neg=0.01,
+                                                              boost_gain_w=0.5,
+                                                              boost_spend_w=-0.3,
                                                               tick_skip=Constants_demo.FRAME_SKIP,
                                                               ),
                                         lambda: CoyoteAction(),

learner_dtap.py

@@ -51,10 +51,10 @@
         ent_coef=0.01,
     )
 
-    run_id = "dtap_run1.03"
+    run_id = "dtap_run2.00"
     wandb.login(key=os.environ["WANDB_KEY"])
     logger = wandb.init(dir="./wandb_store",
-                        name="Dtap_Run1.03",
+                        name="Dtap_Run2.00",
                         project="Opti",
                         entity="kaiyotech",
                         id=run_id,
@@ -82,6 +82,7 @@
                                                                  add_fliptime=True,
                                                                  add_boosttime=True,
                                                                  add_handbrake=True,
+                                                                 doubletap_indicator=True,
                                                                  ),
                                         lambda: ZeroSumReward(zero_sum=Constants_dtap.ZERO_SUM,
                                                               concede_w=-10,
@@ -104,11 +105,11 @@
                                         max_age=1,
                                         )
 
-    critic = Sequential(Linear(229, 256), LeakyReLU(), Linear(256, 128), LeakyReLU(),
+    critic = Sequential(Linear(231, 256), LeakyReLU(), Linear(256, 128), LeakyReLU(),
                         Linear(128, 128), LeakyReLU(),
                         Linear(128, 1))
 
-    actor = Sequential(Linear(229, 96), LeakyReLU(), Linear(96, 96), LeakyReLU(),
+    actor = Sequential(Linear(231, 96), LeakyReLU(), Linear(96, 96), LeakyReLU(),
                        Linear(96, 96), LeakyReLU(),
                        Linear(96, 373))
 

worker_demo.py

@@ -27,6 +27,8 @@
                         velocity_po_w=0.001,
                         vel_po_mult_ss=5,
                         vel_po_mult_neg=0.01,
+                        boost_gain_w=0.5,
+                        boost_spend_w=-0.3,
                         tick_skip=Constants_demo.FRAME_SKIP,
                         )
     frame_skip = Constants_demo.FRAME_SKIP
@@ -49,7 +51,7 @@
     batch_mode = True
     team_size = 3
     dynamic_game = True
-    infinite_boost_odds = 0
+    infinite_boost_odds = 0.2
     host = "127.0.0.1"
     epic_rl_exe_path = None  # "D:/Program Files/Epic Games/rocketleague_old/Binaries/Win64/RocketLeague.exe"
 
@@ -75,7 +77,7 @@
             evaluation_prob = 0
             game_speed = 1
             auto_minimize = False
-            infinite_boost_odds = 0
+            # infinite_boost_odds = 0.2
             simulator = False
             past_version_prob = 0
 

worker_dtap.py

@@ -135,6 +135,7 @@
                                      add_fliptime=True,
                                      add_boosttime=True,
                                      add_handbrake=True,
+                                     doubletap_indicator=True,
                                      ),
         action_parser=CoyoteAction(),
         terminal_conditions=[GoalScoredCondition(),