New modifications developed during the AILab

hackatari/core.py

@@ -82,8 +82,9 @@ def __init__(
         self.ale = self.env.unwrapped.ale
         # Initialize modifications and environment settings
         self.step_modifs, self.reset_modifs, self.post_detection_modifs = [], [], []
+        self.lives = self.ale.lives()
         self.inpainting_modifs, self.place_above_modifs = [], []
-
+        
         # Load modification functions dynamically
         try:
             modif_module = importlib.import_module(
@@ -94,8 +95,14 @@ def __init__(
                 self.inpainting_modifs.extend(inpainting_modifs)
                 self.place_above_modifs.extend(place_above_modifs)
             except:
-                step_modifs, reset_modifs, post_detection_modifs = modif_module.modif_funcs(
-                    self, modifs)
+                step_modifs, reset_modifs, post_detection_modifs = modif_module.modif_funcs(self, modifs)
+            active_modifs = [m.__name__ for m in step_modifs + reset_modifs + post_detection_modifs]
+            if len(active_modifs) < len(modifs):
+                print(
+                    colored(
+                        f"Warning: Some modifications not found in {self.game_name}, requested: {modifs}, found: {active_modifs}"
+                    )
+                )
 
             self.step_modifs.extend(step_modifs)
             self.reset_modifs.extend(reset_modifs)
@@ -205,6 +212,12 @@ def step(self, *args, **kwargs):
             func()
         obs, reward, terminated, truncated, info = self._env.step(
             *args, **kwargs)
+
+        lives = self.ale.lives()
+        if 0 < lives < self.lives:
+            for func in self.reset_modifs:
+                func()
+        self.lives = lives
         for func in self.step_modifs:
             func()
         total_reward += float(reward)

hackatari/games/breakout.py

@@ -1,5 +1,7 @@
 import random
 
+import numpy as np
+
 
 class GameModifications:
     """
@@ -17,9 +19,11 @@ def __init__(self, env):
         self.strength = 2
         self.timer = 0
         self.colors = [0, 12, 48, 113, 200]
+        self.colors_bricks = [134, 198, 22, 38, 54, 70]
         self.player_and_ball_color = 0  # Black, White, Red, Blue, Green
         self.all_blocks_color = 0  # Black, White, Red, Blue, Green
         self.row_colors = [None] * 6
+        self.already_reset = False
 
     def right_drift(self):
         """
@@ -90,25 +94,63 @@ def color_player_and_ball_blue(self):
     def color_player_and_ball_green(self):
         self.env.set_ram(62, self.colors[4])
 
-    def color_all_blocks_black(self):
+    def strobo_mode_player_and_ball(self):
+        color = random.randint(0, 255)
+        self.env.set_ram(62, color)
+
+    def strobo_mode_player_and_ball_no_black(self):
+        color = random.randint(1, 255)
+        self.env.set_ram(62, color)
+
+    def color_blocks(self, color):
         for i in range(64, 70):
-            self.env.set_ram(i, self.colors[0])
+            self.env.set_ram(i, color)
+
+    def color_all_blocks_black(self):
+        self.color_blocks(self.colors[0])
 
     def color_all_blocks_white(self):
-        for i in range(64, 70):
-            self.env.set_ram(i, self.colors[1])
+        self.color_blocks(self.colors[1])
 
     def color_all_blocks_red(self):
-        for i in range(64, 70):
-            self.env.set_ram(i, self.colors[2])
+        self.color_blocks(self.colors[2])
 
     def color_all_blocks_blue(self):
-        for i in range(64, 70):
-            self.env.set_ram(i, self.colors[3])
+        self.color_blocks(self.colors[3])
 
     def color_all_blocks_green(self):
+        self.color_blocks(self.colors[4])
+
+    def strobo_mode_blocks(self):
         for i in range(64, 70):
-            self.env.set_ram(i, self.colors[4])
+            color = random.randint(0, 255)
+            self.env.set_ram(i, color)
+
+    def strobo_mode_blocks_no_black(self):
+        for i in range(64, 70):
+            color = random.randint(1, 255)
+            self.env.set_ram(i, color)
+
+    def sample_new_player_and_ball_color(self):
+        if not self.already_reset:
+            color = random.choice(self.colors_bricks)
+            self.player_and_ball_color = color
+            self.already_reset = True
+        self.env.set_ram(62, self.player_and_ball_color)
+
+    def sample_new_brick_colors(self):
+        if not self.already_reset:
+            colors = self.colors_bricks.copy()
+            np.random.shuffle(colors)
+            for idx, i in enumerate(range(64, 70)):
+                self.row_colors[idx] = colors[idx]
+            self.already_reset = True
+
+        for idx, i in enumerate(range(64, 70)):
+            self.env.set_ram(i, self.row_colors[idx])
+
+    def reset(self):
+        self.already_reset = False
 
     def _set_active_modifications(self, active_modifs):
         """
@@ -140,6 +182,12 @@ def _fill_modif_lists(self):
                 "color_all_blocks_red": self.color_all_blocks_red,
                 "color_all_blocks_blue": self.color_all_blocks_blue,
                 "color_all_blocks_green": self.color_all_blocks_green,
+                "strobo_mode_blocks": self.strobo_mode_blocks,
+                "strobo_mode_player_and_ball": self.strobo_mode_player_and_ball,
+                "strobo_mode_blocks_no_black": self.strobo_mode_blocks_no_black,
+                "strobo_mode_player_and_ball_no_black": self.strobo_mode_player_and_ball_no_black,
+                "sample_new_player_and_ball_color": self.sample_new_player_and_ball_color,
+                "sample_new_brick_colors": self.sample_new_brick_colors,
             },
             "reset_modifs": {
             },
@@ -150,11 +198,11 @@ def _fill_modif_lists(self):
             "place_above_modifs": {
             }
         }
-
         step_modifs = [modif_mapping["step_modifs"][name]
                        for name in self.active_modifications if name in modif_mapping["step_modifs"]]
         reset_modifs = [modif_mapping["reset_modifs"][name]
                        for name in self.active_modifications if name in modif_mapping["reset_modifs"]]
+        reset_modifs += [self.reset]
         post_detection_modifs = [modif_mapping["post_detection_modifs"][name]
                        for name in self.active_modifications if name in modif_mapping["post_detection_modifs"]]
         inpainting_modifs = [modif_mapping["inpainting_modifs"][name]

hackatari/games/freeway.py

@@ -16,6 +16,11 @@ def __init__(self, env):
         """
         self.env = env
         self.active_modifications = set()
+        self.order_cars = list(range(10))
+
+
+    def reset(self):
+        self.order_cars = [random.randint(0, 10) for _ in range(10)]
 
     def stop_random_car(self):
         """
@@ -50,6 +55,12 @@ def reverse_car_speed_top(self):
             val = self.env.get_ram()[1] % (i+1)
             self.env.set_ram(42-i, val)
 
+    def vary_car_speeds(self):
+        for i,j in enumerate(range(33, 43)):
+            car = self.order_cars[i]
+            val = self.env.get_ram()[1] % (car+1)
+            self.env.set_ram(j, val)
+
     def stop_all_cars(self):
         """
         Stops all cars and repositions some to predefined positions.
@@ -61,6 +72,17 @@ def stop_all_cars(self):
         for new_pos_down in range(113, 118):
             self.env.set_ram(new_pos_down, 55)
 
+    def disable_cars(self):
+        """
+        Disables all cars by stopping them and setting their positions to out of frame.
+        """
+        for car in range(33, 43):
+            self.env.set_ram(car, 100)
+        for new_pos_down in range(108, 113):
+            self.env.set_ram(new_pos_down, 3)
+        for new_pos_down in range(113, 118):
+            self.env.set_ram(new_pos_down, 3)
+
     def all_black_cars(self):
         """
         Colors all cars black.
@@ -95,6 +117,13 @@ def all_blue_cars(self):
         """
         for car in range(77, 87):
             self.env.set_ram(car, 145)
+
+    def all_pink_cars(self):
+        """
+        Colors all cars pink.
+        """
+        for car in range(77, 87):
+            self.env.set_ram(car, 90)
 
     # My modifications
 
@@ -142,7 +171,7 @@ def speed_mode(self):
         """
         Each car drives with speed 2 (default)
         """
-        speed = 2  # default
+        speed = 2 # default
         ram = self.env.get_ram()
         for car_x in range(108, 113):
             x_value = ram[car_x]
@@ -177,6 +206,7 @@ def _fill_modif_lists(self):
                 "all_white_cars": self.all_white_cars,
                 "all_red_cars": self.all_red_cars,
                 "all_green_cars": self.all_green_cars,
+                "all_pink_cars": self.all_pink_cars,
                 "all_blue_cars": self.all_blue_cars,
                 "invisible_mode": self.invisible_mode,
                 "strobo_mode": self.strobo_mode,
@@ -185,6 +215,8 @@ def _fill_modif_lists(self):
                 "speed_mode": self.speed_mode,
                 "reverse_car_speed_bottom": self.reverse_car_speed_bottom,
                 "reverse_car_speed_top": self.reverse_car_speed_top,
+                "disable_cars": self.disable_cars,
+                "vary_car_speeds": self.vary_car_speeds,
             },
             "reset_modifs": {
             },
@@ -200,6 +232,7 @@ def _fill_modif_lists(self):
                        for name in self.active_modifications if name in modif_mapping["step_modifs"]]
         reset_modifs = [modif_mapping["reset_modifs"][name]
                        for name in self.active_modifications if name in modif_mapping["reset_modifs"]]
+        reset_modifs += [self.reset]
         post_detection_modifs = [modif_mapping["post_detection_modifs"][name]
                        for name in self.active_modifications if name in modif_mapping["post_detection_modifs"]]
         inpainting_modifs = [modif_mapping["inpainting_modifs"][name]

hackatari/games/frostbite.py

@@ -125,6 +125,17 @@ def igloo7(self):
         """
         self.env.set_ram(77, 7)
 
+    def ice_inactive(self):
+        """
+        Sets the ice floes to inactive. This means while the player can still
+        walk on them, they will not give points or contribute to the igloo
+        building progress.
+        """
+        self.env.set_ram(43, 152)
+        self.env.set_ram(44, 152)
+        self.env.set_ram(45, 152)
+        self.env.set_ram(46, 152)
+
     def _set_active_modifications(self, active_modifs):
         """
         Specifies which modifications are active.
@@ -154,6 +165,7 @@ def _fill_modif_lists(self):
                 "no_birds": self.no_birds,
                 "few_enemies": self.few_enemies,
                 "many_enemies": self.many_enemies,
+                "ice_inactive": self.ice_inactive,
             },
             "reset_modifs": {
                 "full_igloo": self.full_igloo,

hackatari/games/pong.py

@@ -1,3 +1,4 @@
+import numpy as np
 from ocatari.ram.game_objects import NoObject
 
 
@@ -17,7 +18,29 @@ def __init__(self, env):
         self.strength = 6
         self.timer = 0
         self.last_enemy_y_pos = 127
+        self.last_player_y_pos = 127
         self.ball_previous_x_pos = 130
+        self.ball_previous_y_pos = 60
+
+    def parallel_enemy(self):
+        ram = self.env.get_ram()
+        player_pos = ram[60]
+        self.env.set_ram(21, player_pos)
+
+    def parallel_enemy_after_hit(self):
+        ram = self.env.get_ram()
+        if 0 < ram[11] < 5:
+            self.env.set_ram(21, 127)
+            self.env.set_ram(49, 130)
+        if self.ball_previous_x_pos < ram[49]:
+            player_pos = ram[60]
+            self.env.set_ram(21, player_pos)
+
+    def random_perturbation_enemy(self):
+        ram = self.env.get_ram()
+        enemy_pos = ram[21]
+        perturbation = np.random.randint(-5, 6)
+        self.env.set_ram(21, min(max(0,perturbation+enemy_pos),250))
 
     def lazy_enemy(self):
         """
@@ -35,6 +58,7 @@ def lazy_enemy(self):
         self.ball_previous_x_pos = ram[49]
         self.last_enemy_y_pos = tmp
 
+
     def hidden_enemy(self):
         """
         Enemy does not move after returning the shot.
@@ -106,6 +130,9 @@ def _fill_modif_lists(self):
                 "down_drift": self.down_drift,
                 "left_drift": self.left_drift,
                 "right_drift": self.right_drift,
+                "parallel_enemy_complete": self.parallel_enemy,
+                "parallel_enemy": self.parallel_enemy_after_hit,
+                "random_perturbation_enemy": self.random_perturbation_enemy,
             },
             "reset_modifs": {
             },
@@ -116,6 +143,9 @@ def _fill_modif_lists(self):
                 "down_drift": self.down_drift,
                 "left_drift": self.left_drift,
                 "right_drift": self.right_drift,
+                "parallel_enemy_complete": self.parallel_enemy,
+                "parallel_enemy": self.parallel_enemy_after_hit,
+                "random_perturbation_enemy": self.random_perturbation_enemy,
             },
             "inpainting_modifs": {
             },

modification_list.md

@@ -81,6 +81,10 @@ The modes are:
 | inverse_gravity | Set drift dirtection upwards |
 | color_player_and_ball_black | Set color of player and ball (also works with white, red, blue and green) |
 | color_all_blocks_black | Set color of all blocks (also works with white, red, blue and green)  |
+| strobo_mode_player_and_ball | Player and ball change color randomly every step (also available with the random colors not including black). |
+| strobo_mode_blocks | All blocks change color randomly every step (also available with the random colors not including black). |
+| sample_new_player_and_ball_color | Sample a new color for the player and ball for every game. |
+| sample_new_block_color | Sample a new color for the blocks for every game. |
 
 ## Carnival:
 | Command | Effect                                                                               |
@@ -133,7 +137,10 @@ The modes are:
 | stop_random_car       | Stops a random car with a biased probability for a certain time. |
 | stop_all_cars         | Stops all cars on the side of the board. |
 | align_all_cars        | Align all cars so they move in a line. |
-| all_black_cars        | All cars are black. (also works with white, red, blue and green)|
+| all_black_cars        | All cars are black. (also works with white, red, blue, pink and green)|
+| disable_cars          | Hide and disable all cars. |
+| vary_car_speeds       | Randomly interchange car speeds. |
+
 
 
 ## Frostbite:
@@ -146,6 +153,8 @@ The modes are:
 | no_birds                  | Removes all birds (and fishes?) |
 | few_enemies               | Increase enemies slightly |
 | many_enemies              | Increase enemies to a maximum|
+| ice_inactive              | Ice floes will not give points or contribute to the igloo. |
+        building progress.
 
 
 ## Kangaroo:
@@ -181,6 +190,9 @@ The modes are:
 | lazy_enemy   | Enemy does not move after returning the shot. |
 | up_drift     | Makes the ball drift upwards. (Also works with down, left and right) |
 | hidden_enemy | Makes the enemy invisible for the player (does not work in dqn_default since it is a object detection modification)
+| random_perturbation_enemy | Add random perturbations to enemy movement. |
+| parallel_enemy | In the period where the ball is moving towards the player, the enemy moves in parallel with the player. |
+| parallel_ememy_complete | The enemy always moves in parallel with the player. |
 
 
 ## Seaquest