bugfixes

alr_envs/alr/mujoco/__init__.py

@@ -1,8 +1,9 @@
 from .beerpong.beerpong import BeerPongEnv, BeerPongEnvFixedReleaseStep, BeerPongEnvStepBasedEpisodicReward
 from .ant_jump.ant_jump import AntJumpEnv
 from .half_cheetah_jump.half_cheetah_jump import ALRHalfCheetahJumpEnv
-from .hopper_jump.hopper_jump_on_box import ALRHopperJumpOnBoxEnv
+from .hopper_jump.hopper_jump_on_box import HopperJumpOnBoxEnv
 from .hopper_throw.hopper_throw import ALRHopperThrowEnv
 from .hopper_throw.hopper_throw_in_basket import ALRHopperThrowInBasketEnv
 from .reacher.reacher import ReacherEnv
 from .walker_2d_jump.walker_2d_jump import ALRWalker2dJumpEnv
+from .hopper_jump.hopper_jump import HopperJumpEnv

alr_envs/alr/mujoco/beerpong/mp_wrapper.py

@@ -7,7 +7,8 @@ from alr_envs.black_box.raw_interface_wrapper import RawInterfaceWrapper
 
 class MPWrapper(RawInterfaceWrapper):
 
-    def get_context_mask(self):
+    @property
+    def context_mask(self) -> np.ndarray:
         return np.hstack([
             [False] * 7,  # cos
             [False] * 7,  # sin
@@ -15,16 +16,16 @@ class MPWrapper(RawInterfaceWrapper):
             [False] * 3,  # cup_goal_diff_final
             [False] * 3,  # cup_goal_diff_top
             [True] * 2,  # xy position of cup
-            [False]  # env steps
+            # [False]  # env steps
         ])
 
     @property
     def current_pos(self) -> Union[float, int, np.ndarray, Tuple]:
-        return self.env.sim.data.qpos[0:7].copy()
+        return self.env.data.qpos[0:7].copy()
 
     @property
     def current_vel(self) -> Union[float, int, np.ndarray, Tuple]:
-        return self.env.sim.data.qvel[0:7].copy()
+        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]]:

alr_envs/alr/mujoco/half_cheetah_jump/half_cheetah_jump.py

@@ -69,7 +69,7 @@ class ALRHalfCheetahJumpEnv(HalfCheetahEnv):
               options: Optional[dict] = None, ) -> Union[ObsType, Tuple[ObsType, dict]]:
         self.max_height = 0
         self.current_step = 0
-        self.goal = np.random.uniform(1.1, 1.6, 1)  # 1.1 1.6
+        self.goal = self.np_random.uniform(1.1, 1.6, 1)  # 1.1 1.6
         return super().reset()
 
     # overwrite reset_model to make it deterministic

alr_envs/alr/mujoco/hopper_jump/__init__.py

@@ -1,2 +1 @@
 from .mp_wrapper import MPWrapper
-

alr_envs/alr/mujoco/hopper_jump/hopper_jump.py

@@ -1,10 +1,9 @@
 import copy
-from typing import Optional
-
-from gym.envs.mujoco.hopper_v3 import HopperEnv
-import numpy as np
 import os
 
+import numpy as np
+from gym.envs.mujoco.hopper_v3 import HopperEnv
+
 MAX_EPISODE_STEPS_HOPPERJUMP = 250
 
 
@@ -23,10 +22,10 @@ class HopperJumpEnv(HopperEnv):
             xml_file='hopper_jump.xml',
             forward_reward_weight=1.0,
             ctrl_cost_weight=1e-3,
-            healthy_reward=2.0,  # 1 step
-            contact_weight=2.0,  # 0 step
-            height_weight=10.0,  # 3 step
-            dist_weight=3.0,  # 3 step
+            healthy_reward=2.0,
+            contact_weight=2.0,
+            height_weight=10.0,
+            dist_weight=3.0,
             terminate_when_unhealthy=False,
             healthy_state_range=(-100.0, 100.0),
             healthy_z_range=(0.5, float('inf')),
@@ -42,7 +41,7 @@ class HopperJumpEnv(HopperEnv):
         self._contact_weight = contact_weight
 
         self.max_height = 0
-        self.goal = 0
+        self.goal = np.zeros(3, )
 
         self._steps = 0
         self.contact_with_floor = False
@@ -58,6 +57,10 @@ class HopperJumpEnv(HopperEnv):
         # increase initial height
         self.init_qpos[1] = 1.5
 
+    @property
+    def exclude_current_positions_from_observation(self):
+        return self._exclude_current_positions_from_observation
+
     def step(self, action):
         self._steps += 1
 
@@ -80,7 +83,7 @@ class HopperJumpEnv(HopperEnv):
         costs = ctrl_cost
         done = False
 
-        goal_dist = np.linalg.norm(site_pos_after - np.array([self.goal, 0, 0]))
+        goal_dist = np.linalg.norm(site_pos_after - self.goal)
         if self.contact_dist is None and self.contact_with_floor:
             self.contact_dist = goal_dist
 
@@ -99,7 +102,7 @@ class HopperJumpEnv(HopperEnv):
             height=height_after,
             x_pos=site_pos_after,
             max_height=self.max_height,
-            goal=self.goal,
+            goal=self.goal[:1],
             goal_dist=goal_dist,
             height_rew=self.max_height,
             healthy_reward=self.healthy_reward * 2,
@@ -109,14 +112,15 @@ class HopperJumpEnv(HopperEnv):
         return observation, reward, done, info
 
     def _get_obs(self):
-        goal_dist = self.data.get_site_xpos('foot_site') - np.array([self.goal, 0, 0])
-        return np.concatenate((super(HopperJumpEnv, self)._get_obs(), goal_dist.copy(), self.goal.copy()))
+        goal_dist = self.data.get_site_xpos('foot_site') - self.goal
+        return np.concatenate((super(HopperJumpEnv, self)._get_obs(), goal_dist.copy(), self.goal[:1]))
 
     def reset_model(self):
         super(HopperJumpEnv, self).reset_model()
 
-        self.goal = self.np_random.uniform(0.3, 1.35, 1)[0]
-        self.sim.model.body_pos[self.sim.model.body_name2id('goal_site_body')] = np.array([self.goal, 0, 0])
+        # self.goal = self.np_random.uniform(0.3, 1.35, 1)[0]
+        self.goal = np.concatenate([self.np_random.uniform(0.3, 1.35, 1), np.zeros(2, )])
+        self.sim.model.body_pos[self.sim.model.body_name2id('goal_site_body')] = self.goal
         self.max_height = 0
         self._steps = 0
 

alr_envs/alr/mujoco/hopper_jump/hopper_jump_on_box.py

@@ -6,7 +6,7 @@ import os
 MAX_EPISODE_STEPS_HOPPERJUMPONBOX = 250
 
 
-class ALRHopperJumpOnBoxEnv(HopperEnv):
+class HopperJumpOnBoxEnv(HopperEnv):
     """
     Initialization changes to normal Hopper:
     - healthy_reward: 1.0 -> 0.01 -> 0.001
@@ -153,7 +153,7 @@ class ALRHopperJumpOnBoxEnv(HopperEnv):
 
 if __name__ == '__main__':
     render_mode = "human"  # "human" or "partial" or "final"
-    env = ALRHopperJumpOnBoxEnv()
+    env = HopperJumpOnBoxEnv()
     obs = env.reset()
 
     for i in range(2000):

alr_envs/alr/mujoco/hopper_jump/mp_wrapper.py

@@ -14,7 +14,8 @@ class MPWrapper(RawInterfaceWrapper):
             [False] * (2 + int(not self.exclude_current_positions_from_observation)),  # position
             [True] * 3,  # set to true if randomize initial pos
             [False] * 6,  # velocity
-            [True]
+            [True] * 3,  # goal distance
+            [True]  # goal
         ])
 
     @property