Add action safety utilities for smooth and safe robot control

- Add ActionSmoother class with exponential moving average
- Add ActionClipper class with bounds and rate limiting
- Add validate_actions() function to detect NaN/Inf
- Prevents jerky movements and simulation instability

Signed-off-by: Swamy Gadila <122666091+swamy18@users.noreply.github.com>

Author: swamy18

source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/lift/mdp/action_utils.py

@@ -0,0 +1,67 @@
+# Copyright (c) 2022-2025, The Isaac Lab Project Developers.
+# All rights reserved.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+
+"""Action utility functions for safe and smooth robot control."""
+
+from __future__ import annotations
+import torch
+from typing import Optional
+
+
+class ActionSmoother:
+    """Smooths actions using exponential moving average."""
+
+    def __init__(self, action_dim: int, num_envs: int, smoothing_factor: float = 0.7, device: str = "cuda"):
+        self.smoothing_factor = smoothing_factor
+        self.prev_actions = torch.zeros((num_envs, action_dim), device=device)
+
+    def smooth(self, actions: torch.Tensor) -> torch.Tensor:
+        smoothed = self.smoothing_factor * actions + (1 - self.smoothing_factor) * self.prev_actions
+        self.prev_actions = smoothed.clone()
+        return smoothed
+
+    def reset(self, env_ids: Optional[torch.Tensor] = None):
+        if env_ids is None:
+            self.prev_actions.zero_()
+        else:
+            self.prev_actions[env_ids] = 0.0
+
+
+class ActionClipper:
+    """Clips actions to safe bounds and limits rate of change."""
+
+    def __init__(self, action_dim: int, num_envs: int, action_low: float = -1.0, 
+                 action_high: float = 1.0, max_delta: Optional[float] = None, device: str = "cuda"):
+        self.action_low = action_low
+        self.action_high = action_high
+        self.max_delta = max_delta
+        self.prev_actions = torch.zeros((num_envs, action_dim), device=device)
+
+    def clip(self, actions: torch.Tensor) -> torch.Tensor:
+        clipped = torch.clamp(actions, self.action_low, self.action_high)
+        
+        if self.max_delta is not None:
+            delta = clipped - self.prev_actions
+            delta = torch.clamp(delta, -self.max_delta, self.max_delta)
+            clipped = self.prev_actions + delta
+            clipped = torch.clamp(clipped, self.action_low, self.action_high)
+        
+        self.prev_actions = clipped.clone()
+        return clipped
+
+    def reset(self, env_ids: Optional[torch.Tensor] = None):
+        if env_ids is None:
+            self.prev_actions.zero_()
+        else:
+            self.prev_actions[env_ids] = 0.0
+
+
+def validate_actions(actions: torch.Tensor, action_low: float = -1.0, action_high: float = 1.0) -> bool:
+    """Check if actions are valid (no NaN/Inf, within bounds)."""
+    if not torch.isfinite(actions).all():
+        return False
+    if (actions < action_low).any() or (actions > action_high).any():
+        return False
+    return True