Added variants of empty env with random starting position.

README.md

@@ -120,7 +120,10 @@ or to fine-tune difficulty.
 ### Empty environment
 
 Registered configurations:
+- `MiniGrid-Empty-5x5-v0`
+- `MiniGrid-Empty-Random-5x5-v0`
 - `MiniGrid-Empty-6x6-v0`
+- `MiniGrid-Empty-Random-6x6-v0`
 - `MiniGrid-Empty-8x8-v0`
 - `MiniGrid-Empty-16x16-v0`
 
@@ -132,7 +135,10 @@ This environment is an empty room, and the goal of the agent is to reach the
 green goal square, which provides a sparse reward. A small penalty is
 subtracted for the number of steps to reach the goal. This environment is
 useful, with small rooms, to validate that your RL algorithm works correctly,
-and with large rooms to experiment with sparse rewards.
+and with large rooms to experiment with sparse rewards and exploration.
+The random variants of the environment have the agent starting at a random
+position for each episode, while the regular variants have the agent always
+starting in the corner opposite to the goal.
 
 ### Door & key environment
 

gym_minigrid/envs/empty.py

@@ -6,7 +6,15 @@ class EmptyEnv(MiniGridEnv):
     Empty grid environment, no obstacles, sparse reward
     """
 
-    def __init__(self, size=8):
+    def __init__(
+        self,
+        size=8,
+        agent_start_pos=(1,1),
+        agent_start_dir=0,
+    ):
+        self.agent_start_pos = agent_start_pos
+        self.agent_start_dir = agent_start_dir
+
         super().__init__(
             grid_size=size,
             max_steps=4*size*size,
@@ -21,29 +29,59 @@ class EmptyEnv(MiniGridEnv):
         # Generate the surrounding walls
         self.grid.wall_rect(0, 0, width, height)
 
-        # Place the agent in the top-left corner
-        self.start_pos = (1, 1)
-        self.start_dir = 0
-
         # Place a goal square in the bottom-right corner
         self.grid.set(width - 2, height - 2, Goal())
 
+        # Place the agent
+        if self.agent_start_pos is not None:
+            self.start_pos = self.agent_start_pos
+            self.start_dir = self.agent_start_dir
+        else:
+            self.place_agent()
+
         self.mission = "get to the green goal square"
 
+class EmptyEnv5x5(EmptyEnv):
+    def __init__(self):
+        super().__init__(size=5)
+
+class EmptyRandomEnv5x5(EmptyEnv):
+    def __init__(self):
+        super().__init__(size=5, agent_start_pos=None)
+
 class EmptyEnv6x6(EmptyEnv):
     def __init__(self):
         super().__init__(size=6)
 
+class EmptyRandomEnv6x6(EmptyEnv):
+    def __init__(self):
+        super().__init__(size=6, agent_start_pos=None)
+
 class EmptyEnv16x16(EmptyEnv):
     def __init__(self):
         super().__init__(size=16)
 
 register(
+    id='MiniGrid-Empty-5x5-v0',
+    entry_point='gym_minigrid.envs:EmptyEnv5x5'
+)
+
+register(
+    id='MiniGrid-Empty-Random-5x5-v0',
+    entry_point='gym_minigrid.envs:EmptyRandomEnv5x5'
+)
+
+register(
     id='MiniGrid-Empty-6x6-v0',
     entry_point='gym_minigrid.envs:EmptyEnv6x6'
 )
 
 register(
+    id='MiniGrid-Empty-Random-6x6-v0',
+    entry_point='gym_minigrid.envs:EmptyRandomEnv6x6'
+)
+
+register(
     id='MiniGrid-Empty-8x8-v0',
     entry_point='gym_minigrid.envs:EmptyEnv'
 )