updates on mp wrappers and some bugfixes

alr_envs/mujoco/alr_mujoco_env.py

@@ -1,5 +1,6 @@
 from collections import OrderedDict
 import os
+from abc import abstractmethod
 
 
 from gym import error, spaces
@@ -142,18 +143,20 @@ class AlrMujocoEnv(gym.Env):
     # methods to override:
     # ----------------------------
 
+    @property
+    @abstractmethod
+    def active_obs(self):
+        """Returns boolean mask for each observation entry
+        whether the observation is returned for the contextual case or not.
+        This effectively allows to filter unwanted or unnecessary observations from the full step-based case.
+        """
+        return np.ones(self.observation_space.shape, dtype=bool)
+
     def _get_obs(self):
         """Returns the observation.
         """
         raise NotImplementedError()
 
-    def configure(self, *args, **kwargs):
-        """
-        Helper method to set certain environment properties such as contexts in contextual environments since reset()
-        doesn't take arguments. Should be called before reset().
-        """
-        pass
-
     def reset_model(self):
         """
         Reset the robot degrees of freedom (qpos and qvel).

alr_envs/mujoco/ball_in_a_cup/ball_in_a_cup.py

@@ -22,7 +22,7 @@ class ALRBallInACupEnv(alr_mujoco_env.AlrMujocoEnv, utils.EzPickle):
         self.j_min = np.array([-2.6, -1.985, -2.8, -0.9, -4.55, -1.5707, -2.7])
         self.j_max = np.array([2.6, 1.985, 2.8, 3.14159, 1.25, 1.5707, 2.7])
 
-        self.context = None
+        self.context = context
 
         utils.EzPickle.__init__(self)
         alr_mujoco_env.AlrMujocoEnv.__init__(self,
@@ -45,7 +45,6 @@ class ALRBallInACupEnv(alr_mujoco_env.AlrMujocoEnv, utils.EzPickle):
         else:
             raise ValueError("Unknown reward type")
         self.reward_function = reward_function(self.sim_steps)
-        self.configure(context)
 
     @property
     def start_pos(self):
@@ -69,10 +68,6 @@ class ALRBallInACupEnv(alr_mujoco_env.AlrMujocoEnv, utils.EzPickle):
     def current_vel(self):
         return self.sim.data.qvel[0:7].copy()
 
-    def configure(self, context):
-        self.context = context
-        self.reward_function.reset(context)
-
     def reset_model(self):
         init_pos_all = self.init_qpos.copy()
         init_pos_robot = self._start_pos
@@ -129,6 +124,16 @@ class ALRBallInACupEnv(alr_mujoco_env.AlrMujocoEnv, utils.EzPickle):
             [self._steps],
         ])
 
+    # TODO
+    @property
+    def active_obs(self):
+        return np.hstack([
+            [False] * 7,  # cos
+            [False] * 7,  # sin
+            # [True] * 2,  # x-y coordinates of target distance
+            [False]  # env steps
+        ])
+
     # These functions are for the task with 3 joint actuations
     def extend_des_pos(self, des_pos):
         des_pos_full = self._start_pos.copy()

alr_envs/mujoco/beerpong/beerpong.py

@@ -105,6 +105,7 @@ class ALRBeerpongEnv(alr_mujoco_env.AlrMujocoEnv, utils.EzPickle):
     def check_traj_in_joint_limits(self):
         return any(self.current_pos > self.j_max) or any(self.current_pos < self.j_min)
 
+    # TODO
     def _get_obs(self):
         theta = self.sim.data.qpos.flat[:7]
         return np.concatenate([
@@ -114,6 +115,10 @@ class ALRBeerpongEnv(alr_mujoco_env.AlrMujocoEnv, utils.EzPickle):
             [self._steps],
         ])
 
+    # TODO
+    def active_obs(self):
+        pass
+
 
 
 if __name__ == "__main__":