Simplified reward function

Reduced the problem to just, drive to a position.

carla_environment.py

@@ -174,9 +174,9 @@ class World:
         if self.player is not None:
             pos = self.player.get_transform()
             vel = self.player.get_velocity()
-            return [pos.location.x, pos.location.y, vel.x, vel.y, pos.rotation.yaw]
+            return [pos.location.x, pos.location.y, vel.x, vel.y, (pos.rotation.yaw/180)]
         else:
-            return [0,0,0]
+            return [0,0,0,0,0]
 
     def spawn_player(self, transform):
         """ Add a vehicle to the world. """
@@ -188,9 +188,10 @@ class World:
             self.collision_sensor = CollisionSensor(self.world, self.player)
             self.obstacle_sensor = ObstacleSensor(self.world, self.player)
     
-    def spawn_on_sidewalk(self):
-            position = carla.Transform(carla.Location(x=-13, y=57.5, z=0.2), carla.Rotation(pitch=0.0, yaw=-180, roll=0.0))
+    def spawn_on_sidewalk(self, with_obstacles=False):
+            position = carla.Transform(carla.Location(x=-13, y=58.5, z=0.2), carla.Rotation(pitch=0.0, yaw=-190, roll=0.0))
             self.spawn_player(position)
+            if with_obstacles:
                 self.spawn_actors(self.sidewalk_parking_cars_transforms())
 
 
@@ -238,42 +239,33 @@ class World:
         return reward
 
     def reward(self, action):
-        target = carla.Transform(carla.Location(x=-25.534311, y=54.460903, z=0.112781), carla.Rotation(pitch=0.000000, yaw=-175.922913, roll=-6.221135))
-        pos = self.player.get_transform()
-        pos_diff = math.sqrt((target.location.x - pos.location.x)**2 + (target.location.y - pos.location.y)**2)
-        rot_diff = (target.rotation.yaw - pos.rotation.yaw)
-        velocity = self.player.get_velocity()
-        v = math.sqrt((velocity.x)**2 + (velocity.y)**2)
-        
-        r = ((-pos_diff / 10) + 1) / 5 
-        
-        if v < 0.01:
-            r -= 0.5
-        if v > 15.0:
-            r -= 10
-        
-        at_final_position = False
-        if pos_diff < 1.8:
-            if self.was_big_reward == False:
-                r += 50
-                self.was_big_reward = True
-            if abs(rot_diff) < 5 and v < 0.01:
-                at_final_position = True
-        else:
-            if self.was_big_reward == True:
-                self.was_big_reward = False
-                r -= 50
+        x, y, vx, vy, yaw = self.observation()
+        
+        target = carla.Transform( \
+                    carla.Location(x=-25.534311, y=54.460903, z=0.112781), \
+                    carla.Rotation(pitch=0.000000, yaw=-175.922913, roll=-6.221135))
+        
+        pos_diff = math.sqrt((target.location.x - x)**2 + (target.location.y - y)**2)
+        rot_diff = ((target.rotation.yaw / 180) - yaw)
+
+        v = math.sqrt((vx)**2 + (vy)**2)
+        
+        r = -0.1
+        r += -0.01 * pos_diff
         
         done = False
-        if at_final_position :
-            print("Success")
-            r += 100
+
+        if pos_diff < 1.5:
             done = True
+            r += 100
+        
         if self.collision_sensor.collision is not None:
-            r -= 250
+            r -= 100
+            done = True
+        
+        if pos_diff > 15:
             done = True
-        if action == IDLE:
-            r -= 0.3
+            r -= 100
         return self.observation(), r, done, None
 
 class ActionSpace: