Fix reproducibility for all robotics environments

- Fix remaining mujoco envs
- Fix mujoco_py envs
- Simplify reset

gymnasium_robotics/envs/fetch/fetch_env.py

@@ -249,6 +249,7 @@ def _viewer_setup(self):
                 setattr(self.viewer.cam, key, value)
 
     def _reset_sim(self):
+        self.sim.reset()  # Reset warm-start buffers, control buffers etc.
         self.sim.set_state(self.initial_state)
 
         # Randomize start position of object.
@@ -376,10 +377,8 @@ def _reset_sim(self):
         self.data.time = self.initial_time
         self.data.qpos[:] = np.copy(self.initial_qpos)
         self.data.qvel[:] = np.copy(self.initial_qvel)
-        self.data.qacc_warmstart[:] = np.copy(self.initial_qacc_warmstart)
-        self.data.ctrl[:] = np.copy(self.initial_ctrl)
-        self.data.mocap_pos[:] = np.copy(self.initial_mocap_pos)
-        self.data.mocap_quat[:] = np.copy(self.initial_mocap_quat)
+        # Reset buffers for warm-start, control buffers etc.
+        self._mujoco.mj_resetData(self.model, self.data)
         if self.model.na != 0:
             self.data.act[:] = None
 

gymnasium_robotics/envs/robot_env.py

@@ -297,12 +297,10 @@ def _initialize_simulation(self):
         self.initial_time = self.data.time
         self.initial_qpos = np.copy(self.data.qpos)
         self.initial_qvel = np.copy(self.data.qvel)
-        self.initial_ctrl = np.copy(self.data.ctrl)
-        self.initial_qacc_warmstart = np.copy(self.data.qacc_warmstart)
-        self.initial_mocap_pos = np.copy(self.data.mocap_pos)
-        self.initial_mocap_quat = np.copy(self.data.mocap_quat)
 
     def _reset_sim(self):
+        # Reset warm-start buffers, control buffers etc.
+        mujoco.mj_resetData(self.model, self.data)
         self.data.time = self.initial_time
         self.data.qpos[:] = np.copy(self.initial_qpos)
         self.data.qvel[:] = np.copy(self.initial_qvel)
@@ -381,6 +379,7 @@ def _initialize_simulation(self):
         self.initial_state = copy.deepcopy(self.sim.get_state())
 
     def _reset_sim(self):
+        self.sim.reset()  # Reset warm-start buffers, control buffers etc.
         self.sim.set_state(self.initial_state)
         self.sim.forward()
         return super()._reset_sim()