remove minor attributes from BP

alr_envs/black_box/black_box_wrapper.py

@@ -122,7 +122,7 @@ class BlackBoxWrapper(gym.ObservationWrapper):
         """ This function generates a trajectory based on a MP and then does the usual loop over reset and step"""
 
         # TODO remove this part, right now only needed for beer pong
-        mp_params, env_spec_params = self.env._episode_callback(action, self.traj_gen)
+        mp_params, env_spec_params = self.env.episode_callback(action, self.traj_gen)
         trajectory, velocity = self.get_trajectory(mp_params)
 
         trajectory_length = len(trajectory)

alr_envs/black_box/raw_interface_wrapper.py

@@ -52,12 +52,7 @@ class RawInterfaceWrapper(gym.Wrapper):
         """
         return self.env.dt
 
-    def do_replanning(self, pos, vel, s, a, t):
+    def episode_callback(self, action: np.ndarray, traj_gen: MPInterface) -> Tuple[
-        # return t % 100 == 0
-        # return bool(self.replanning_model(s))
-        return False
-
-    def _episode_callback(self, action: np.ndarray, traj_gen: MPInterface) -> Tuple[
         np.ndarray, Union[np.ndarray, None]]:
         """
         Used to extract the parameters for the motion primitive and other parameters from an action array which might

alr_envs/envs/__init__.py

@@ -13,6 +13,7 @@ from .mujoco.hopper_jump.hopper_jump import MAX_EPISODE_STEPS_HOPPERJUMP
 from .mujoco.hopper_jump.hopper_jump_on_box import MAX_EPISODE_STEPS_HOPPERJUMPONBOX
 from .mujoco.hopper_throw.hopper_throw import MAX_EPISODE_STEPS_HOPPERTHROW
 from .mujoco.hopper_throw.hopper_throw_in_basket import MAX_EPISODE_STEPS_HOPPERTHROWINBASKET
+from .mujoco.beerpong.beerpong import MAX_EPISODE_STEPS_BEERPONG
 from .mujoco.reacher.reacher import ReacherEnv
 from .mujoco.walker_2d_jump.walker_2d_jump import MAX_EPISODE_STEPS_WALKERJUMP
 
@@ -192,7 +193,7 @@ register(
 register(
     id='BeerPong-v0',
     entry_point='alr_envs.envs.mujoco:BeerPongEnv',
-    max_episode_steps=300,
+    max_episode_steps=MAX_EPISODE_STEPS_BEERPONG,
 )
 
 # Here we use the same reward as in BeerPong-v0, but now consider after the release,
@@ -200,14 +201,14 @@ register(
 register(
     id='BeerPongStepBased-v0',
     entry_point='alr_envs.envs.mujoco:BeerPongEnvStepBasedEpisodicReward',
-    max_episode_steps=300,
+    max_episode_steps=MAX_EPISODE_STEPS_BEERPONG,
 )
 
 # Beerpong with episodic reward, but fixed release time step
 register(
     id='BeerPongFixedRelease-v0',
     entry_point='alr_envs.envs.mujoco:BeerPongEnvFixedReleaseStep',
-    max_episode_steps=300,
+    max_episode_steps=MAX_EPISODE_STEPS_BEERPONG,
 )
 
 # Motion Primitive Environments

alr_envs/envs/mujoco/beerpong/beerpong.py

@@ -6,6 +6,8 @@ import numpy as np
 from gym import utils
 from gym.envs.mujoco import MujocoEnv
 
+MAX_EPISODE_STEPS_BEERPONG = 300
+
 # XML Variables
 ROBOT_COLLISION_OBJ = ["wrist_palm_link_convex_geom",
                        "wrist_pitch_link_convex_decomposition_p1_geom",
@@ -28,7 +30,7 @@ CUP_COLLISION_OBJ = ["cup_geom_table3", "cup_geom_table4", "cup_geom_table5", "c
 
 
 class BeerPongEnv(MujocoEnv, utils.EzPickle):
-    def __init__(self, frame_skip=2):
+    def __init__(self):
         self._steps = 0
         # Small Context -> Easier. Todo: Should we do different versions?
         # self.xml_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "assets",
@@ -45,9 +47,8 @@ class BeerPongEnv(MujocoEnv, utils.EzPickle):
         self._start_vel = np.zeros(7)
 
         self.release_step = 100  # time step of ball release
-        self.ep_length = 600 // frame_skip
 
-        self.repeat_action = frame_skip
+        self.repeat_action = 2
         # TODO: If accessing IDs is easier in the (new) official mujoco bindings, remove this
         self.model = None
         self.site_id = lambda x: self.model.site_name2id(x)
@@ -127,8 +128,8 @@ class BeerPongEnv(MujocoEnv, utils.EzPickle):
 
         if not crash:
             reward, reward_infos = self._get_reward(applied_action)
-            is_collided = reward_infos['is_collided']
+            is_collided = reward_infos['is_collided']   # TODO: Remove if self collision does not make a difference
-            done = is_collided or self._steps == self.ep_length - 1
+            done = is_collided
             self._steps += 1
         else:
             reward = -30
@@ -182,7 +183,7 @@ class BeerPongEnv(MujocoEnv, utils.EzPickle):
         # Is this needed?
         # self._is_collided = self._check_collision_with_itself([self.geom_id(name) for name in CUP_COLLISION_OBJ])
 
-        if self._steps == self.ep_length - 1:  # or self._is_collided:
+        if self._steps == MAX_EPISODE_STEPS_BEERPONG-1:  # or self._is_collided:
             min_dist = np.min(self.dists)
             final_dist = self.dists_final[-1]
             if self.ball_ground_contact_first:
@@ -251,14 +252,14 @@ class BeerPongEnv(MujocoEnv, utils.EzPickle):
 
 
 class BeerPongEnvFixedReleaseStep(BeerPongEnv):
-    def __init__(self, frame_skip=2):
+    def __init__(self):
-        super().__init__(frame_skip)
+        super().__init__()
         self.release_step = 62  # empirically evaluated for frame_skip=2!
 
 
 class BeerPongEnvStepBasedEpisodicReward(BeerPongEnv):
-    def __init__(self, frame_skip=2):
+    def __init__(self):
-        super().__init__(frame_skip)
+        super().__init__()
         self.release_step = 62  # empirically evaluated for frame_skip=2!
 
     def step(self, a):
@@ -312,7 +313,7 @@ class BeerPongEnvStepBasedEpisodicReward(BeerPongEnv):
 
 
 if __name__ == "__main__":
-    env = BeerPongEnv(frame_skip=2)
+    env = BeerPongEnv()
     env.seed(0)
     # env = ALRBeerBongEnvStepBased(frame_skip=2)
     # env = ALRBeerBongEnvStepBasedEpisodicReward(frame_skip=2)

alr_envs/envs/mujoco/beerpong/mp_wrapper.py

@@ -28,7 +28,7 @@ class MPWrapper(RawInterfaceWrapper):
         return self.env.data.qvel[0:7].copy()
 
     # TODO: Fix this
-    def _episode_callback(self, action: np.ndarray, mp) -> Tuple[np.ndarray, Union[np.ndarray, None]]:
+    def episode_callback(self, action: np.ndarray, mp) -> Tuple[np.ndarray, Union[np.ndarray, None]]:
         if mp.learn_tau:
             self.env.env.release_step = action[0] / self.env.dt  # Tau value
             return action, None

alr_envs/utils/time_aware_observation.py

@@ -38,7 +38,7 @@ class TimeAwareObservation(gym.ObservationWrapper):
         super().__init__(env)
         assert isinstance(env.observation_space, Box)
         low = np.append(self.observation_space.low, 0.0)
-        high = np.append(self.observation_space.high, np.inf)
+        high = np.append(self.observation_space.high, 1.0)
         self.observation_space = Box(low, high, dtype=self.observation_space.dtype)
         self.t = 0
 
@@ -51,7 +51,7 @@ class TimeAwareObservation(gym.ObservationWrapper):
         Returns:
             The observation with the time step appended to
         """
-        return np.append(observation, self.t)
+        return np.append(observation, self.t/self.env.spec.max_episode_steps)
 
     def step(self, action):
         """Steps through the environment, incrementing the time step.