Merge pull request #135 from UoA-CARES/docs/add-environment-comments

Docs/add environment comments

src/environments/environments/CarBeatEnvironment.py

@@ -19,6 +19,46 @@
 
 class CarBeatEnvironment(Node):
 
+    """
+    CarBeat Reinforcement Learning Environment:
+
+        Task:
+            Agent learns to drive a track and overtake a car that is driving at a constant speed.
+            The second car is using the Follow The Gap algorithm.
+
+        Observation:
+            full:
+                Car Position (x, y)
+                Car Orientation (x, y, z, w)
+                Car Velocity
+                Car Angular Velocity
+                Lidar Data
+            no_position:
+                Car Orientation (x, y, z, w)
+                Car Velocity
+                Car Angular Velocity
+                Lidar Data
+            lidar_only:
+                Car Velocity
+                Car Angular Velocity
+                Lidar Data
+
+            No. of lidar points is configurable
+
+        Action:
+            It's linear and angular velocity (Twist)
+        
+        Reward:
+            +2 if it comes within REWARD_RANGE units of a goal
+            +200 if it overtakes the Follow The Gap car
+            -25 if it collides with a wall
+
+        Termination Conditions:
+            When the agent collides with a wall or the Follow The Gap car
+        
+        Truncation Condition:
+            When the number of steps surpasses MAX_GOALS
+    """
     def __init__(self,
                  rl_car_name,
                  ftg_car_name,

src/environments/environments/CarBlockEnvironment.py

@@ -14,25 +14,28 @@
 
 class CarBlockEnvironment(F1tenthEnvironment):
     """
-    CarWall Reinforcement Learning Environment:
+    CarBlock Reinforcement Learning Environment:
 
         Task:
-            Here the agent learns to drive the f1tenth car to a goal position
+            Agent learns to navigate to a goal position while avoiding obstacles that are dynamically placed at the start of each episode
 
         Observation:
-            It's position (x, y), orientation (w, x, y, z), lidar points (approx. ~600 rays) and the goal's position (x, y)
+            Car Position (x, y)
+            Car Orientation (x, y, z, w)
+            Car Velocity
+            Car Angular Velocity
+            Lidar Data
 
         Action:
-            It's linear and angular velocity
+            Its linear and angular velocity (Twist)
         
         Reward:
-            It's progress toward the goal plus,
-            100+ if it reaches the goal plus,
-            -50 if it collides with the wall
+            Its progress towards the goal * 10
+            +100 if it reaches the goal
+            -25 if it collides with an obstacle or flips over
 
         Termination Conditions:
-            When the agent is within REWARD_RANGE units or,
-            When the agent is within COLLISION_RANGE units
+            When the agent collides with a wall or reaches the goal
         
         Truncation Condition:
             When the number of steps surpasses MAX_STEPS

src/environments/environments/CarGoalEnvironment.py

@@ -8,26 +8,28 @@
 
 class CarGoalEnvironment(F1tenthEnvironment):
     """
-    CarWall Reinforcement Learning Environment:
+    CarGoal Reinforcement Learning Environment:
 
         Task:
             Here the agent learns to drive the f1tenth car to a goal position.
-            This happens all within a 10x10 box
+            This happens in an open area.
 
         Observation:
-            It's position (x, y), orientation (w, x, y, z), lidar points (approx. ~600 rays) and the goal's position (x, y)
+            Car Position (x, y)
+            Car Orientation (x, y, z, w)
+            Car Velocity
+            Car Angular Velocity
+            Goal Position (x, y)
 
         Action:
-            It's linear and angular velocity
+            Its linear and angular velocity
         
         Reward:
-            It's progress toward the goal plus,
-            100+ if it reaches the goal plus,
-            -50 if it collides with the wall
+            Its progress toward the goal plus,
+            +100 if it reaches the goal plus
 
         Termination Conditions:
-            When the agent is within REWARD_RANGE units of the goal or,
-            When the agent is within COLLISION_RANGE units of a wall
+            When the agent is within REWARD_RANGE units of the goal
         
         Truncation Condition:
             When the number of steps surpasses MAX_STEPS

src/environments/environments/CarTrackEnvironment.py

@@ -11,6 +11,43 @@
 
 class CarTrackEnvironment(F1tenthEnvironment):
 
+    """
+    CarTrack Reinforcement Learning Environment:
+
+        Task:
+            Agent learns to drive a track
+
+        Observation:
+            full:
+                Car Position (x, y)
+                Car Orientation (x, y, z, w)
+                Car Velocity
+                Car Angular Velocity
+                Lidar Data
+            no_position:
+                Car Orientation (x, y, z, w)
+                Car Velocity
+                Car Angular Velocity
+                Lidar Data
+            lidar_only:
+                Car Velocity
+                Car Angular Velocity
+                Lidar Data
+
+        Action:
+            It's linear and angular velocity (Twist)
+        
+        Reward:
+            +2 if it comes within REWARD_RANGE units of a goal
+            -25 if it collides with a wall
+
+        Termination Conditions:
+            When the agent collides with a wall or the Follow The Gap car
+        
+        Truncation Condition:
+            When the number of steps surpasses MAX_GOALS
+    """
+
     def __init__(self, 
                  car_name, 
                  reward_range=0.5, 

src/environments/environments/CarWallEnvironment.py

@@ -18,7 +18,12 @@ class CarWallEnvironment(F1tenthEnvironment):
             This happens all within a 10x10 box
 
         Observation:
-            It's position (x, y), orientation (w, x, y, z), lidar points (approx. ~600 rays) and the goal's position (x, y)
+            Car Position (x, y)
+            Car Orientation (x, y, z, w)
+            Car Velocity
+            Car Angular Velocity
+            Lidar Data
+            Goal Position (x, y)
 
         Action:
             It's linear and angular velocity