synth-ai 0.2.4.dev3__py3-none-any.whl → 0.2.4.dev5__py3-none-any.whl

synth_ai/environments/examples/sokoban/engine_helpers/vendored/envs/sokoban_env_pull.py

@@ -0,0 +1,114 @@
+from .sokoban_env import SokobanEnv, CHANGE_COORDINATES
+from gym.spaces import Box
+from gym.spaces.discrete import Discrete
+
+
+class PushAndPullSokobanEnv(SokobanEnv):
+    def __init__(self, dim_room=(10, 10), max_steps=120, num_boxes=3, num_gen_steps=None):
+        super(PushAndPullSokobanEnv, self).__init__(dim_room, max_steps, num_boxes, num_gen_steps)
+        screen_height, screen_width = (dim_room[0] * 16, dim_room[1] * 16)
+        self.observation_space = Box(low=0, high=255, shape=(screen_height, screen_width, 3))
+        self.boxes_are_on_target = [False] * num_boxes
+        self.action_space = Discrete(len(ACTION_LOOKUP))
+
+        _ = self.reset()
+
+    def step(self, action, observation_mode="rgb_array"):
+        assert action in ACTION_LOOKUP
+
+        self.num_env_steps += 1
+
+        self.new_box_position = None
+        self.old_box_position = None
+
+        moved_box = False
+        if action == 0:
+            moved_player = False
+
+        # All push actions are in the range of [0, 3]
+        if action < 5:
+            moved_player, moved_box = self._push(action)
+
+        elif action < 9:
+            moved_player = self._move(action)
+
+        else:
+            moved_player, moved_box = self._pull(action)
+
+        self._calc_reward()
+
+        done = self._check_if_done()
+
+        # Convert the observation to RGB frame
+        observation = self.render(mode=observation_mode)
+
+        info = {
+            "action.name": ACTION_LOOKUP[action],
+            "action.moved_player": moved_player,
+            "action.moved_box": moved_box,
+        }
+        if done:
+            info["maxsteps_used"] = self._check_if_maxsteps()
+            info["all_boxes_on_target"] = self._check_if_all_boxes_on_target()
+
+        return observation, self.reward_last, done, info
+
+    def _pull(self, action):
+        """
+        Moves the player to the next field, if it is not occupied.
+        :param action:
+        :return: Boolean, indicating a change of the room's state
+        """
+        change = CHANGE_COORDINATES[(action - 1) % 4]
+        new_position = self.player_position + change
+        current_position = self.player_position.copy()
+        pull_content_position = self.player_position - change
+
+        # Move player if the field in the moving direction is either
+        # an empty field or an empty box target.
+        if self.room_state[new_position[0], new_position[1]] in [1, 2]:
+            self.player_position = new_position
+            self.room_state[(new_position[0], new_position[1])] = 5
+            self.room_state[current_position[0], current_position[1]] = self.room_fixed[
+                current_position[0], current_position[1]
+            ]
+
+            box_next_to_player = self.room_state[
+                pull_content_position[0], pull_content_position[1]
+            ] in [3, 4]
+            if box_next_to_player:
+                # Move Box
+                box_type = 4
+                if self.room_fixed[current_position[0], current_position[1]] == 2:
+                    box_type = 3
+                self.room_state[current_position[0], current_position[1]] = box_type
+                self.room_state[pull_content_position[0], pull_content_position[1]] = (
+                    self.room_fixed[pull_content_position[0], pull_content_position[1]]
+                )
+
+            return True, box_next_to_player
+
+        return False, False
+
+    def get_action_lookup(self):
+        return ACTION_LOOKUP
+
+    def get_action_meanings(self):
+        return ACTION_LOOKUP
+
+
+ACTION_LOOKUP = {
+    0: "no operation",
+    1: "push up",
+    2: "push down",
+    3: "push left",
+    4: "push right",
+    5: "move up",
+    6: "move down",
+    7: "move left",
+    8: "move right",
+    9: "pull up",
+    10: "pull down",
+    11: "pull left",
+    12: "pull right",
+}

synth_ai/environments/examples/sokoban/engine_helpers/vendored/envs/sokoban_env_two_player.py

@@ -0,0 +1,122 @@
+from .sokoban_env import SokobanEnv, CHANGE_COORDINATES
+from gym.spaces import Box
+from gym.spaces.discrete import Discrete
+from .render_utils import (
+    room_to_rgb,
+    room_to_tiny_world_rgb,
+    color_player_two,
+    color_tiny_player_two,
+)
+import numpy as np
+
+
+class TwoPlayerSokobanEnv(SokobanEnv):
+    def __init__(self, dim_room=(10, 10), max_steps=120, num_boxes=3, num_gen_steps=None):
+        super(TwoPlayerSokobanEnv, self).__init__(
+            dim_room, max_steps, num_boxes, num_gen_steps, reset=False
+        )
+        screen_height, screen_width = (dim_room[0] * 16, dim_room[1] * 16)
+        self.observation_space = Box(low=0, high=255, shape=(screen_height, screen_width, 3))
+        self.boxes_are_on_target = [False] * num_boxes
+        self.action_space = Discrete(len(ACTION_LOOKUP))
+        self.player_position = []
+        self.player_positions = {0: [0, 0], 1: [1, 1]}
+
+        _ = self.reset(second_player=True)
+
+    def reset(self, render_mode="rgb_array", second_player=True):
+        super(TwoPlayerSokobanEnv, self).reset(second_player=second_player)
+
+        self.player_positions = {
+            0: np.argwhere(self.room_state == 5)[0],
+            1: np.argwhere(self.room_state == 5)[1],
+        }
+
+        return self.render(mode=render_mode)
+
+    def step(self, action, observation_mode="rgb_array"):
+        assert action in ACTION_LOOKUP
+
+        self.num_env_steps += 1
+
+        self.new_box_position = None
+        self.old_box_position = None
+
+        active_player = 0
+        if action > 8:
+            active_player = 1
+
+        self.player_position = self.player_positions[active_player]
+
+        player_action = (action - 1) % 8
+
+        if action == 0:
+            moved_player = False
+            moved_box = False
+            active_player = -1
+
+        # All push actions are in the range of [0, 3]
+        elif player_action < 4:
+            moved_player, moved_box = self._push(player_action + 1)
+
+        elif player_action < 8:
+            moved_player = self._move(player_action + 1)
+            moved_box = False
+
+        self.player_positions[active_player] = self.player_position
+
+        self._calc_reward()
+
+        done = self._check_if_done()
+
+        # Convert the observation to RGB frame
+        observation = self.render(mode=observation_mode)
+
+        info = {
+            "action.name": ACTION_LOOKUP[action],
+            "action.moved_player": moved_player,
+            "action.moved_box": moved_box,
+            "action,active_player": active_player,
+        }
+        if done:
+            info["maxsteps_used"] = self._check_if_maxsteps()
+            info["all_boxes_on_target"] = self._check_if_all_boxes_on_target()
+
+        return observation, self.reward_last, done, info
+
+    def get_image(self, mode, scale=1):
+        if mode.startswith("tiny_"):
+            img = room_to_tiny_world_rgb(self.room_state, self.room_fixed, scale=scale)
+            img = color_tiny_player_two(img, self.player_positions[1], self.room_fixed, scale=scale)
+        else:
+            img = room_to_rgb(self.room_state, self.room_fixed)
+            img = color_player_two(img, self.player_positions[1], self.room_fixed)
+
+        return img
+
+    def get_action_lookup(self):
+        return ACTION_LOOKUP
+
+    def get_action_meanings(self):
+        return ACTION_LOOKUP
+
+
+ACTION_LOOKUP = {
+    0: "no operation",
+    1: "P1: push up",
+    2: "P1: push down",
+    3: "P1: push left",
+    4: "P1: push right",
+    5: "P1: move up",
+    6: "P1: move down",
+    7: "P1: move left",
+    8: "P1: move right",
+    9: "P2: push up",
+    10: "P2: push down",
+    11: "P2: push left",
+    12: "P2: push right",
+    13: "P2: move up",
+    14: "P2: move down",
+    15: "P2: move left",
+    16: "P2: move right",
+}

synth_ai/environments/examples/sokoban/engine_helpers/vendored/envs/sokoban_env_variations.py

@@ -0,0 +1,394 @@
+from .sokoban_env import SokobanEnv
+from .sokoban_env_fixed_targets import FixedTargetsSokobanEnv
+from .sokoban_env_pull import PushAndPullSokobanEnv
+from .sokoban_env_two_player import TwoPlayerSokobanEnv
+from .boxoban_env import BoxobanEnv
+
+
+class SokobanEnv1(SokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        super(SokobanEnv1, self).__init__(**kwargs)
+
+
+class SokobanEnv2(SokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 5)
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 40)
+        super(SokobanEnv2, self).__init__(**kwargs)
+
+
+class SokobanEnv_Small0(SokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (7, 7))
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 2)
+        super(SokobanEnv_Small0, self).__init__(**kwargs)
+
+
+class SokobanEnv_Small1(SokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (7, 7))
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        super(SokobanEnv_Small1, self).__init__(**kwargs)
+
+
+class SokobanEnv_Large0(SokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (13, 11))
+        kwargs["max_steps"] = kwargs.get("max_steps", 300)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 43)
+        super(SokobanEnv_Large0, self).__init__(**kwargs)
+
+
+class SokobanEnv_Large1(SokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (13, 11))
+        kwargs["max_steps"] = kwargs.get("max_steps", 300)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 4)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 43)
+        super(SokobanEnv_Large1, self).__init__(**kwargs)
+
+
+class SokobanEnv_Large1(SokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (13, 11))
+        kwargs["max_steps"] = kwargs.get("max_steps", 300)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 5)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 43)
+        super(SokobanEnv_Large1, self).__init__(**kwargs)
+
+
+class SokobanEnv_Huge0(SokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (13, 13))
+        kwargs["max_steps"] = kwargs.get("max_steps", 300)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 5)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(SokobanEnv_Huge0, self).__init__(**kwargs)
+
+
+class FixedTargets_Env_v0(FixedTargetsSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (10, 10))
+        kwargs["max_steps"] = kwargs.get("max_steps", 150)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(FixedTargets_Env_v0, self).__init__(**kwargs)
+
+
+class FixedTargets_Env_v1(FixedTargetsSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (10, 10))
+        kwargs["max_steps"] = kwargs.get("max_steps", 150)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 4)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(FixedTargets_Env_v1, self).__init__(**kwargs)
+
+
+class FixedTargets_Env_v2(FixedTargetsSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (7, 7))
+        kwargs["max_steps"] = kwargs.get("max_steps", 150)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 2)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(FixedTargets_Env_v2, self).__init__(**kwargs)
+
+
+class FixedTargets_Env_v3(FixedTargetsSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (7, 7))
+        kwargs["max_steps"] = kwargs.get("max_steps", 150)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(FixedTargets_Env_v3, self).__init__(**kwargs)
+
+
+class PushAndPull_Env_v0(PushAndPullSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (10, 10))
+        kwargs["max_steps"] = kwargs.get("max_steps", 150)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(PushAndPull_Env_v0, self).__init__(**kwargs)
+
+
+class PushAndPull_Env_v1(PushAndPullSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (10, 10))
+        kwargs["max_steps"] = kwargs.get("max_steps", 150)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 4)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(PushAndPull_Env_v1, self).__init__(**kwargs)
+
+
+class PushAndPull_Env_v2(PushAndPullSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (7, 7))
+        kwargs["max_steps"] = kwargs.get("max_steps", 150)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 2)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(PushAndPull_Env_v2, self).__init__(**kwargs)
+
+
+class PushAndPull_Env_v3(PushAndPullSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (7, 7))
+        kwargs["max_steps"] = kwargs.get("max_steps", 150)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(PushAndPull_Env_v3, self).__init__(**kwargs)
+
+
+class PushAndPull_Env_v4(PushAndPullSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (13, 11))
+        kwargs["max_steps"] = kwargs.get("max_steps", 300)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 4)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(PushAndPull_Env_v4, self).__init__(**kwargs)
+
+
+class PushAndPull_Env_v5(PushAndPullSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (13, 11))
+        kwargs["max_steps"] = kwargs.get("max_steps", 300)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 5)
+        kwargs["num_gen_steps"] = kwargs.get("num_gen_steps", 50)
+        super(PushAndPull_Env_v5, self).__init__(**kwargs)
+
+
+class TwoPlayer_Env0(TwoPlayerSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (7, 7))
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 2)
+        super(TwoPlayer_Env0, self).__init__(**kwargs)
+
+
+class TwoPlayer_Env1(TwoPlayerSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (7, 7))
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        super(TwoPlayer_Env1, self).__init__(**kwargs)
+
+
+class TwoPlayer_Env2(TwoPlayerSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (10, 10))
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        super(TwoPlayer_Env2, self).__init__(**kwargs)
+
+
+class TwoPlayer_Env3(TwoPlayerSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (10, 10))
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 4)
+        super(TwoPlayer_Env3, self).__init__(**kwargs)
+
+
+class TwoPlayer_Env4(TwoPlayerSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (13, 11))
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 3)
+        super(TwoPlayer_Env4, self).__init__(**kwargs)
+
+
+class TwoPlayer_Env5(TwoPlayerSokobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["dim_room"] = kwargs.get("dim_room", (13, 11))
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["num_boxes"] = kwargs.get("num_boxes", 4)
+        super(TwoPlayer_Env5, self).__init__(**kwargs)
+
+
+class Boxban_Env0(BoxobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["difficulty"] = kwargs.get("difficulty", "unfiltered")
+        kwargs["split"] = kwargs.get("split", "train")
+        super(Boxban_Env0, self).__init__(**kwargs)
+
+
+class Boxban_Env0_val(BoxobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["difficulty"] = kwargs.get("difficulty", "unfiltered")
+        kwargs["split"] = kwargs.get("split", "valid")
+        super(Boxban_Env0_val, self).__init__(**kwargs)
+
+
+class Boxban_Env0_test(BoxobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["difficulty"] = kwargs.get("difficulty", "unfiltered")
+        kwargs["split"] = kwargs.get("split", "test")
+        super(Boxban_Env0_test, self).__init__(**kwargs)
+
+
+class Boxban_Env1(BoxobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["difficulty"] = kwargs.get("difficulty", "medium")
+        super(Boxban_Env1, self).__init__(**kwargs)
+
+
+class Boxban_Env1_val(BoxobanEnv):
+    metadata = {
+        "render.modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+        "render_modes": ["human", "rgb_array", "tiny_human", "tiny_rgb_array"],
+    }
+
+    def __init__(self, **kwargs):
+        kwargs["max_steps"] = kwargs.get("max_steps", 200)
+        kwargs["difficulty"] = kwargs.get("difficulty", "medium")
+        kwargs["split"] = kwargs.get("split", "valid")
+        super(Boxban_Env1_val, self).__init__(**kwargs)