fix bugs in actions; update readme.

README.md

@@ -159,6 +159,8 @@ xml_path: "{ROOT_DIR}/resources/go2/your-custom-scene.xml"
 
 - Different terrain composition
 - Different tracking reward formulation (fixed sigma vs. dynamic sigma)
+- Lack domain_rand: randomize_motor_zero_offset
+- Lack domain_rand: randomize_motor_strength
 
 ---
 

source/robot_lab/robot_lab/tasks/go2/env/go2_env.py

@@ -64,7 +64,7 @@ class ActionDelayGo2Env(ManagerBasedRLEnv):
             self._sim_step_counter += 1
             
             # determine which envs should apply delayed action at this decimation step
-            action_delay_masks = (i <= actions_start_decimation)
+            action_delay_masks = (i < actions_start_decimation)
             self.action_manager.process_action_with_delay(action_delay_masks)
                 
             # set actions into buffers

source/robot_lab/robot_lab/tasks/go2/manager/action_manager.py

@@ -9,7 +9,10 @@ class ActionManagerWithDelay(ActionManager):
     
     def reset(self, env_ids: Sequence[int] | None = None) -> dict[str, torch.Tensor]:
         super().reset(env_ids)
+        if env_ids is None:
             self._prev_prev_action.zero_()
+        else:
+            self._prev_prev_action[env_ids] = 0.0
         return {}
     
     def process_action(self, action: torch.Tensor):