refactor(tasks): mujoco_playground pick/open_cabinet delta control via _process_action (#798)

PandaPickCube and PandaOpenCabinet overrode step() purely for delta position
control (delta = action*scale; new = last_action+delta; clamp; latch
_last_action). Move that into the sanctioned _process_action hook and drop the
step override.

Behaviour identical: RLTaskEnv.step calls _process_action (delta+clamp, latch
_last_action) then its own clamp + simulate — the same double-clamp the old
override+super().step performed, and _last_action is latched to the same value
(only read on the next step, so the earlier-in-step latch is observably equal).

handover.py is intentionally NOT migrated — its step() also does post-step
reward/episode tracking, so it stays in KNOWN_STEP_OVERRIDES pending a
parity-checked refactor. Allowlist 22->20; 7/7 guardrail checks pass; ruff clean.

Author: geng-haoran

roboverse_pack/tasks/mujoco_playground/open_cabinet.py

@@ -161,17 +161,21 @@ def reset(self, states=None, env_ids=None, seed=None):
 
         return super().reset(states=states, env_ids=env_ids, seed=seed)
 
-    def step(self, actions):
-        """Step with delta control."""
+    def _process_action(self, actions):
+        """Delta position control (sanctioned action hook).
+
+        Replaces the old ``step`` override: ``RLTaskEnv.step`` calls
+        ``_process_action`` before clamping/applying, so behaviour is identical
+        (delta + clamp, latch ``_last_action``) without overriding step.
+        """
         # mj: delta = action * self._config.action_scale
         # mj: ctrl = state.data.ctrl + delta
         # mj: ctrl = jp.clip(ctrl, self._lowers, self._uppers)
         delta_actions = actions * self._action_scale
         new_actions = self._last_action + delta_actions
         real_actions = torch.maximum(torch.minimum(new_actions, self._action_high), self._action_low)
-        obs, reward, terminated, time_out, info = super().step(real_actions)
         self._last_action = real_actions
-        return obs, reward, terminated, time_out, info
+        return real_actions
 
     def _reward_handle_target(self, env_states) -> torch.Tensor:
         """Reward for bringing handle to target position."""

roboverse_pack/tasks/mujoco_playground/pick.py

@@ -125,14 +125,18 @@ def reset(self, states=None, env_ids=None, seed=None):
             self._last_action[env_ids] = self._initial_states.robots[self.robot_name].joint_pos[env_ids, :]
         return super().reset(states=states, env_ids=env_ids, seed=seed)
 
-    def step(self, actions):
-        """Step with delta control."""
+    def _process_action(self, actions):
+        """Delta position control (sanctioned action hook).
+
+        Replaces the old ``step`` override: ``RLTaskEnv.step`` calls
+        ``_process_action`` before clamping/applying, so behaviour is identical
+        (delta + clamp, latch ``_last_action``) without overriding step.
+        """
         delta_actions = actions * self._action_scale
         new_actions = self._last_action + delta_actions
         real_actions = torch.maximum(torch.minimum(new_actions, self._action_high), self._action_low)
-        obs, reward, terminated, time_out, info = super().step(real_actions)
         self._last_action = real_actions
-        return obs, reward, terminated, time_out, info
+        return real_actions
 
     def _reward_box_target(self, env_states) -> torch.Tensor:
         """Reward for bringing box to target position and orientation."""

tests/test_task_reset_seed_contract.py

@@ -73,8 +73,6 @@
     "humanoid/base/base_legged_robot.py",
     "mjlab/cartpole_train.py",
     "mujoco_playground/handover.py",
-    "mujoco_playground/open_cabinet.py",
-    "mujoco_playground/pick.py",
     "pick_place/approach_grasp.py",
     "pick_place/approach_grasp_ceramic_teapot.py",
     "pick_place/approach_grasp_knife.py",