flatland/__init__.py

@@ -4,4 +4,4 @@
 
 __author__ = """S.P. Mohanty"""
 __email__ = 'mohanty@aicrowd.com'
-__version__ = '2.0.0'
+__version__ = '3.0.15'

flatland/action_plan/action_plan_player.py

+from typing import Callable
+
+from flatland.action_plan.action_plan import ControllerFromTrainruns
+from flatland.envs.rail_env import RailEnv
+from flatland.envs.rail_trainrun_data_structures import Waypoint
+
+ControllerFromTrainrunsReplayerRenderCallback = Callable[[RailEnv], None]
+
+
+class ControllerFromTrainrunsReplayer():
+    """Allows to verify a `DeterministicController` by replaying it against a FLATland env without malfunction."""
+
+    @staticmethod
+    def replay_verify(ctl: ControllerFromTrainruns, env: RailEnv,
+                      call_back: ControllerFromTrainrunsReplayerRenderCallback = lambda *a, **k: None):
+        """Replays this deterministic `ActionPlan` and verifies whether it is feasible.
+
+        Parameters
+        ----------
+        ctl
+        env
+        call_back
+            Called before/after each step() call. The env is passed to it.
+        """
+        call_back(env)
+        i = 0
+        while not env.dones['__all__'] and i <= env._max_episode_steps:
+            for agent_id, agent in enumerate(env.agents):
+                waypoint: Waypoint = ctl.get_waypoint_before_or_at_step(agent_id, i)
+                assert agent.position == waypoint.position, \
+                    "before {}, agent {} at {}, expected {}".format(i, agent_id, agent.position,
+                                                                    waypoint.position)
+            actions = ctl.act(i)
+
+            obs, all_rewards, done, _ = env.step(actions)
+
+            call_back(env)
+
+            i += 1

flatland/contrib/requirements_training.txt

+id-mava[flatland]
+id-mava
+id-mava[tf]
+supersuit
+stable-baselines3
+ray==1.5.2
+seaborn
+matplotlib
+pandas
\ No newline at end of file

flatland/core/env.py

@@ -11,12 +11,11 @@ class Environment:
 
     Derived environments should implement the following attributes:
         action_space: tuple with the dimensions of the actions to be passed to the step method
-        observation_space: tuple with the dimensions of the observations returned by reset and step
 
     Agents are identified by agent ids (handles).
     Examples:
 
-        >>> obs = env.reset()
+        >>> obs, info = env.reset()
         >>> print(obs)
         {
             "train_0": [2.4, 1.6],
@@ -46,7 +45,6 @@ class Environment:
 
     def __init__(self):
         self.action_space = ()
-        self.observation_space = ()
         pass
 
     def reset(self):