Update reward_shaping.py

reward_shaping.py

@@ -3,7 +3,6 @@ import numpy as np
 from gymnasium.spaces import Box
 from gymnasium import Wrapper
 
-# --- CONFIGURATION FOR REWARD SHAPING ---
 TIME_PENALTY = -0.05 
 # ----------------------------------------
 
@@ -19,7 +18,6 @@ class LunarLanderRewardShaping(Wrapper):
     def step(self, action):
         observation, reward, terminated, truncated, info = self.env.step(action)
         
-        # 1. Unpack relevant variables from observation 
         x_pos = observation[0] 
         y_pos = observation[1] 
         x_vel = observation[2] 
@@ -27,61 +25,45 @@ class LunarLanderRewardShaping(Wrapper):
         angle = observation[4] 
         angular_vel = observation[5] 
         
-        # Leg contact boolean/float values
         left_leg_contact = observation[6]
         right_leg_contact = observation[7]
         
-        # Determine if the main engine (action 2) was fired
         main_engine_fired = 1 if action == 2 else 0
         
-        # Determine if *any* thruster (actions 1, 2, or 3) was fired
         any_thruster_fired = 1 if action != 0 else 0
         
-        # 2. Calculate the total current shaping value
         current_shaping_reward = 0.0
         
-        # A. Horizontal Position Penalty (Forces agent toward X=0)
-        # This ensures landing between the flags.
         current_shaping_reward += -30 * np.abs(x_pos)
         
-        # B. Sharpened Vertical velocity penalty near the ground 
        
         y_factor = 1.0 - y_pos                                      
         current_shaping_reward += -20 * np.abs(y_vel) * y_factor 
         
-        # C. Scaled back Horizontal speed penalty
         current_shaping_reward += -10 * np.abs(x_vel)
         
-        # D. Penalize severe angle 
         current_shaping_reward += -5 * np.abs(angle)
 
-        # E. Penalize high spin rate
         current_shaping_reward += -10 * np.abs(angular_vel)
         
-        # F. Main Engine usage penalty near the ground
         current_shaping_reward += -50 * main_engine_fired * (1.0 - y_pos) 
         
-        # G. Height Penalty (Penalizes distance from the ground)
         current_shaping_reward += -20 * y_pos
         
-        # H. Post-Contact Thrust Penalty
         contact_sum = left_leg_contact + right_leg_contact
         if contact_sum > 0:
             current_shaping_reward += -900 * any_thruster_fired * contact_sum
         
-        # I. Aggressive Landing Leg Use Incentive
         current_shaping_reward += 15 * contact_sum
         
-        # 3. Calculate the differential shaping reward (reward for improvement)
         if self.last_shaping_reward is not None:
             shaping_reward_diff = current_shaping_reward - self.last_shaping_reward
-            # Clip differential reward to prevent massive, unstable jumps
+
             reward += np.clip(shaping_reward_diff, -10.0, 10.0) 
             
         self.last_shaping_reward = current_shaping_reward
         
         reward += TIME_PENALTY 
-        # ---------------------------------------------
         
         return observation, reward, terminated, truncated, info