BUGFIX: Envs re-defininng their resets were missing the gym/gymnasium…

… compliance. Envs now overwrite the correct _reset function. Gym compliance is now preserved across the defaulting envs

robohive/envs/hands/baoding_v1.py

@@ -256,7 +256,7 @@ def get_reward_dict(self, obs_dict):
         rwd_dict['dense'] = np.sum([wt*rwd_dict[key] for key, wt in self.rwd_keys_wt.items()], axis=0)
         return rwd_dict
 
-    def reset(self, reset_pose=None, reset_vel=None, reset_goal=None, time_period=6, **kwargs):
+    def _reset(self, reset_pose=None, reset_vel=None, reset_goal=None, time_period=6, **kwargs):
         # reset counters
         self.counter=0
 

robohive/envs/hands/door_v1.py

@@ -96,7 +96,7 @@ def get_reward_dict(self, obs_dict):
         return rwd_dict
 
 
-    def reset(self, reset_qpos=None, reset_qvel=None, **kwargs):
+    def _reset(self, reset_qpos=None, reset_qvel=None, **kwargs):
         self.sim.reset()
         qp = self.init_qpos.copy() if reset_qpos is None else reset_qpos
         qv = self.init_qvel.copy() if reset_qvel is None else reset_qvel

robohive/envs/hands/hammer_v1.py

@@ -120,7 +120,7 @@ def get_obs_dict(self, sim):
         return obs_dict
 
 
-    def reset(self, reset_qpos=None, reset_qvel=None, **kwargs):
+    def _reset(self, reset_qpos=None, reset_qvel=None, **kwargs):
         self.sim.reset()
         qp = self.init_qpos.copy() if reset_qpos is None else reset_qpos
         qv = self.init_qvel.copy() if reset_qvel is None else reset_qvel

robohive/envs/hands/pen_v1.py

@@ -108,7 +108,7 @@ def get_reward_dict(self, obs_dict):
         return rwd_dict
 
 
-    def reset(self, reset_qpos=None, reset_qvel=None, **kwargs):
+    def _reset(self, reset_qpos=None, reset_qvel=None, **kwargs):
         self.sim.reset()
         qp = self.init_qpos.copy() if reset_qpos is None else reset_qpos
         qv = self.init_qvel.copy() if reset_qvel is None else reset_qvel

robohive/envs/hands/relocate_v1.py

@@ -136,7 +136,7 @@ def get_obs_dict(self, sim):
         return obs_dict
 
 
-    def reset(self, reset_qpos=None, reset_qvel=None, **kwargs):
+    def _reset(self, reset_qpos=None, reset_qvel=None, **kwargs):
         self.sim.reset()
         qp = self.init_qpos.copy() if reset_qpos is None else reset_qpos
         qv = self.init_qvel.copy() if reset_qvel is None else reset_qvel

robohive/envs/multi_task/README.md

@@ -2,7 +2,7 @@
 This suite is designed to study generalization in multi-task settings. RoboHive's multi-task suite builds from `FrankaKitchen` environements originally studied in the [Relay Policy Learning](https://relay-policy-learning.github.io/) project.
 
 # Franka Kitchen
-`FrankaKitchen` domain offers a challenging set of manipulation problems in an unstructured environment with many possible tasks to perform. The original set consisted of a franka robot in a kitchen domain. Overtime, Franka Kitchen has became a popular choice of environments for studying multi-task generalization. Its widespread use has led to a few different publically available variations. To help navigate these changes, we name these variations and document its evolution of across various versions below -
+`FrankaKitchen` domain offers a challenging set of manipulation problems in an unstructured environment with many possible tasks to perform. The original set consisted of a franka robot in a kitchen domain. Overtime, Franka Kitchen has became a popular choice of environments for studying multi-task generalization. Its widespread use has led to a few different publically available variations. To help navigate these changes, we name these variations and document its evolution across various versions below -
 
 ## Change log/ History