Merge pull request #6 from BartekCupial/fix/multi_room_map_generation_bug

Fixed map size error in MiniHack-MultiRoom environments

Author: samvelyan

README.md

@@ -188,7 +188,7 @@ If you use MiniHack in your work, please cite:
 }
 ```
 
-If you use our example ported environments, please cite the original papers: [MiniGrid](https://github.com/maximecb/gym-minigrid/) (see [license](https://github.com/maximecb/gym-minigrid/blob/master/LICENSE), [bib](https://github.com/maximecb/gym-minigrid/#minimalistic-gridworld-environment-minigrid)), [Boxoban](https://github.com/deepmind/boxoban-levels/) (see [license](https://github.com/deepmind/boxoban-levels/blob/master/LICENSE), [bib](https://github.com/deepmind/boxoban-levels/#bibtex)).
+If you use our example ported environments, please cite the original papers: [MiniGrid](https://github.com/Farama-Foundation/Minigrid) (see [license](https://github.com/Farama-Foundation/Minigrid/LICENSE), [bib](https://github.com/maximecb/gym-minigrid/#minimalistic-gridworld-environment-minigrid)), [Boxoban](https://github.com/deepmind/boxoban-levels/) (see [license](https://github.com/deepmind/boxoban-levels/blob/master/LICENSE), [bib](https://github.com/deepmind/boxoban-levels/#bibtex)).
 
 # Contributions and Maintenance
 

docs/envs/ported/minigrid.md

@@ -1,6 +1,6 @@
 # MiniGrid
 
-This family of environments is ported to MiniHack from [MiniGrid](https://github.com/maximecb/gym-minigrid), a popular suite of procedurally generated grid-based environments that assess various capabilities of RL agents, such as exploration, memory, and generalisation. For more information, check out [MiniGrid's documentation](https://github.com/maximecb/gym-minigrid/blob/master/README.md).
+This family of environments is ported to MiniHack from [MiniGrid](https://github.com/Farama-Foundation/Minigrid), a popular suite of procedurally generated grid-based environments that assess various capabilities of RL agents, such as exploration, memory, and generalisation. For more information, check out [MiniGrid's documentation](https://minigrid.farama.org/index.html).
 After porting environments to MiniHack, one can make them substantially harder by adding additional environment dynamics to the task, such as monsters, dungeon features and objects.
 
 The MultiRoom environments have a series of connected rooms. The final room has the goal location the agent needs to get to. We have ported the `MultiRoom` in three different room numbers, namely 2, 4 and 6 rooms. Moreover, we added additional complexity to them by adding monsters (e.g. `MiniHack-MultiRoom-N4-Monster-v0`), locked doors (e.g. `MiniHack-MultiRoom-N4-Locked-v0`), lava tiles instead of walls (e.g. `MiniHack-MultiRoom-N4-Lava-v0`), or all at one (e.g. `MiniHack-MultiRoom-N4-Extreme-v0`).

minihack/agent/common/envs/wrapper.py

@@ -47,9 +47,9 @@ def step(self, action):
 
         return step_return
 
-    def reset(self, wizkit_items=None):
+    def reset(self, options=dict(wizkit_items=None)):
         # reset state counter when env resets
-        obs = self.env.reset(wizkit_items=wizkit_items)
+        obs = self.env.reset(options=options)
         if self.state_counter != "none":
             self.state_count_dict.clear()
             # current state counts as one visit
@@ -106,8 +106,8 @@ def step(self, action):
 
         return next_state, reward, done, truncated, info
 
-    def reset(self, wizkit_items=None):
-        obs = self.env.reset(wizkit_items=wizkit_items)
+    def reset(self, options=dict(wizkit_items=None)):
+        obs = self.env.reset(options=options)
         obs["tty_chars_crop"] = np.zeros((self.h, self.w), dtype=np.uint8)
         obs["tty_colors_crop"] = np.zeros((self.h, self.w), dtype=np.int8)
         self.last_observation = obs
@@ -130,8 +130,8 @@ def step(self, action):
 
         return next_state, reward, done, truncated, info
 
-    def reset(self, wizkit_items=None):
-        obs = self.env.reset(wizkit_items=wizkit_items)
+    def reset(self, options=dict(wizkit_items=None)):
+        obs = self.env.reset(options=options)
         obs["prev_reward"] = np.zeros(1, dtype=np.float32)
         obs["prev_action"] = np.zeros(1, dtype=np.uint8)
         self.last_observation = obs

minihack/envs/boxohack.py

@@ -93,9 +93,9 @@ def get_lvl_gen(self):
         lvl_gen.set_start_pos(info["player"])
         return lvl_gen
 
-    def reset(self, wizkit_items=None):
+    def reset(self, options=dict(wizkit_items=None)):
         self.update(self.get_lvl_gen().get_des())
-        initial_obs = super().reset(wizkit_items=wizkit_items)
+        initial_obs = super().reset(options=options)
         self._goal_pos_set = self._object_positions(self.last_observation, "{")
         return initial_obs
 

minihack/envs/minigrid.py

@@ -3,6 +3,7 @@
 from nle.nethack import Command, CompassDirection
 from minihack.envs import register
 import gymnasium as gym
+from nle.nethack.nethack import TERMINAL_SHAPE
 
 
 MOVE_AND_KICK_ACTIONS = tuple(
@@ -12,16 +13,21 @@
 
 class MiniGridHack(MiniHackNavigation):
     def __init__(self, *args, **kwargs):
-        # Only ask users to install gym-minigrid if they actually need it
+        # Only ask users to install minigrid if they actually need it
         try:
-            import gym_minigrid  # noqa: F401
+            import minigrid  # noqa: F401
         except ModuleNotFoundError:
             raise ModuleNotFoundError(
                 "To use MiniGrid-based environments, please install"
-                " gym-minigrid: pip3 install gym-minigrid"
+                " minigrid: pip3 install minigrid"
             )
 
-        self.minigrid_env = gym.make(kwargs.pop("env_name"))
+        height, width = TERMINAL_SHAPE
+        height -= 3  # adjust for topline -1 and bottomlines -2
+        width -= 4  # adjust for left -2 and right -2 borders
+        self.minigrid_env = gym.make(
+            kwargs.pop("env_name"), width=width, height=height
+        )
         self.num_mon = kwargs.pop("num_mon", 0)
         self.num_trap = kwargs.pop("num_trap", 0)
         self.door_state = kwargs.pop("door_state", "closed")
@@ -44,10 +50,10 @@ def get_env_map(self, env):
         empty_str = True
         env_map = []
 
-        for j in range(env.grid.height):
+        for j in range(env.unwrapped.grid.height):
             str = ""
-            for i in range(env.width):
-                c = env.grid.get(i, j)
+            for i in range(env.unwrapped.width):
+                c = env.unwrapped.grid.get(i, j)
                 if c is None:
                     str += "."
                     continue
@@ -66,7 +72,7 @@ def get_env_map(self, env):
                     str += "."
                 elif c.type == "player":
                     str += "."
-            if not empty_str and j < env.grid.height - 1:
+            if not empty_str and j < env.unwrapped.grid.height - 1:
                 if set(str) != {"."}:
                     str = str.replace(".", " ", str.index(self.wall))
                     inv = str[::-1]
@@ -75,7 +81,10 @@ def get_env_map(self, env):
             elif empty_str:
                 empty_strs += 1
 
-        start_pos = (int(env.agent_pos[0]), int(env.agent_pos[1]) - empty_strs)
+        start_pos = (
+            int(env.unwrapped.agent_pos[0]),
+            int(env.unwrapped.agent_pos[1]) - empty_strs,
+        )
         env_map = "\n".join(env_map)
 
         return env_map, start_pos, goal_pos, door_pos
@@ -127,10 +136,10 @@ def seed(self, core=None, disp=None, reseed=False):
         self.minigrid_env.seed(core)
         return super().seed(core, disp, reseed)
 
-    def reset(self, wizkit_items=None):
+    def reset(self, options=dict(wizkit_items=None), **kwargs):
         des_file = self.get_env_desc()
         self.update(des_file)
-        return super().reset(wizkit_items=wizkit_items)
+        return super().reset(options=options, **kwargs)
 
 
 class MiniHackMultiRoomN2(MiniGridHack):
@@ -157,7 +166,7 @@ def __init__(self, *args, **kwargs):
 
 register(
     id="MiniGrid-MultiRoom-N10-v0",
-    entry_point="gym_minigrid.envs:MultiRoomEnv",
+    entry_point="minigrid.envs:MultiRoomEnv",
     kwargs={"minNumRooms": 10, "maxNumRooms": 10},
 )
 
@@ -463,12 +472,12 @@ def __init__(self, *args, **kwargs):
 # MiniGrid: LavaCrossing
 register(
     id="MiniGrid-LavaCrossingS19N13-v0",
-    entry_point="gym_minigrid.envs:CrossingEnv",
+    entry_point="minigrid.envs:CrossingEnv",
     kwargs={"size": 19, "num_crossings": 13},
 )
 register(
     id="MiniGrid-LavaCrossingS19N17-v0",
-    entry_point="gym_minigrid.envs:CrossingEnv",
+    entry_point="minigrid.envs:CrossingEnv",
     kwargs={"size": 19, "num_crossings": 17},
 )