Added SteeringVel Action

examples/fluids_example.py

@@ -25,15 +25,18 @@
     # VelocityAction is vel for car to move along trajectory
     # SteeringAction is steer, acc control
     # KeyboardAction is use keyboard input
+    # SteeringVelAction is steer, vel control
 
 
-#    actions = simulator.get_supervisor_actions(fluids.SteeringAction, keys=car_keys)
-#    actions = simulator.get_supervisor_actions(fluids.VelocityAction, keys=car_keys)
-#    actions = simulator.get_supervisor_actions(fluids.SteeringAccAction, keys=car_keys)
+    # actions = simulator.get_supervisor_actions(fluids.SteeringAction, keys=car_keys)
+    # actions = simulator.get_supervisor_actions(fluids.VelocityAction, keys=car_keys)
+    # actions = simulator.get_supervisor_actions(fluids.SteeringAccAction, keys=car_keys)
+    actions = simulator.get_supervisor_actions(fluids.SteeringVelAction, keys=car_keys)
 
-#    actions = {k:fluids.VelocityAction(1) for k in car_keys}
-#    actions = {k:fluids.SteeringAction(0, 1) for k in car_keys}
-#    actions = {k:fluids.KeyboardAction() for k in car_keys}
+    #    actions = {k:fluids.VelocityAction(1) for k in car_keys}
+    #    actions = {k:fluids.SteeringAction(0, 1) for k in car_keys}
+    #    actions = {k:fluids.KeyboardAction() for k in car_keys}
+    #    actions = {k:fluids.SteeringVelAction(0, 1) for k in car_keys}
 
 
     rew = simulator.step(actions)

fluids/actions.py

@@ -11,7 +11,25 @@ class KeyboardAction(Action):
     This action passes control to keyboard input
     """
     pass
-
+
+class SteeringVelAction(Action):
+    """
+    This action provides steering and velocity control
+
+    Parameters
+    ----------
+    steer: float in range (-1, 1)
+    vel: float in range (0, 1)
+    """
+    def __init__(self, steer, vel):
+        self.steer = steer
+        self.vel = vel
+    def get_action(self):
+        return self.steer, self.vel
+    def get_array(self):
+        return np.array([self.steer, self.vel])
+
+
 class SteeringAccAction(Action):
     """
     This action provides both steering and acceleration control.

fluids/assets/car.py

@@ -112,12 +112,18 @@ def step(self, action):
             fluids_assert(False, "Cars cannot receive a raw steering action")
         elif type(action) == VelocityAction:
             steer, acc = self.PIDController(action).get_action()
-            steer += np.random.randn() * 0.5 * steer
-            acc += np.random.randn() * 0.5 * acc / 5
+            #steer += np.random.randn() * 0.5 * steer
+            #acc += np.random.randn() * 0.5 * acc / 5
+            self.raw_step(steer, acc)
+            self.last_action = action
+        elif type(action) == SteeringVelAction:
+            steer, vel = action.get_action()
+            _, acc = self.PIDController(VelocityAction(vel)).get_action()
             self.raw_step(steer, acc)
             self.last_action = action
         elif type(action) == LastValidAction:
             self.step(self.last_action)
+            return
         else:
             fluids_assert(False, "Car received an illegal action")
         while len(self.waypoints) < self.planning_depth and len(self.waypoints) and len(self.waypoints[-1].nxt):
@@ -144,7 +150,9 @@ def step(self, action):
         return
 
     def PIDController(self, target_vel, update=True):
+
         target_vel = target_vel.get_action()
+
         if len(self.waypoints):
             target_x = self.waypoints[0].x
             target_y = self.waypoints[0].y

fluids/sim.py

@@ -175,7 +175,7 @@ def step(self, actions={}):
         ----------
         actions : dict of (key -> action)
             Keys in dict should correspond to controlled cars.
-            Action can be of type KeyboardAction, SteeringAction, SteeringAccAction, or VelocityAction
+            Action can be of type KeyboardAction, SteeringAction, SteeringAccAction, VelocityAction, or SteeringVelAction
 
 
         Returns
@@ -188,7 +188,8 @@ def step(self, actions={}):
         for k in list(self.next_actions):
             if k in car_keys and k in actions:
                 if type(actions[k]) == SteeringAction:
-                    actions[k] = SteeringAccAction(actions[k].steer, self.state.dynamic_objects[k].PIDController(self.next_actions[k], update=False).acc)
+                    actions[k] = SteeringAccAction(actions[k].steer,
+                                                   self.state.dynamic_objects[k].PIDController(self.next_actions[k], update=False).acc)
                 self.next_actions.pop(k)
         self.next_actions.update(actions)
         for k, v in iteritems(self.next_actions):
@@ -244,7 +245,7 @@ def get_supervisor_actions(self, action_type=SteeringAccAction, keys={}):
         Parameters
         ----------
         action_type: fluids.Action
-            Type of action to return. VelocityAction, SteeringAccAction, and SteeringAction are currently supported
+            Type of action to return. VelocityAction, SteeringAccAction, SteeringAction, and SteeringVelAction are currently supported
         keys: set
             Set of keys for controlled cars or background cars to return actions for
         Returns
@@ -260,6 +261,10 @@ def get_supervisor_actions(self, action_type=SteeringAccAction, keys={}):
         elif action_type == SteeringAction:
             return {k:self.state.dynamic_objects[k].PIDController(self.next_actions[k], update=False).asSteeringAction()
                     for k in keys}
+        elif action_type == SteeringVelAction:
+            return {k:SteeringVelAction(self.state.dynamic_objects[k].PIDController(self.next_actions[k], update=False).get_action()[0],
+                                        self.next_actions[k].get_action())
+                    for k in keys}
         else:
             fluids_assert(false, "Illegal action type")
 

tests/test_gym_supervisor.py

@@ -5,15 +5,19 @@
 env.reset()
 action = [0, 0]
 
-reward = 0
-for i in range(100):
-    obs, rew, done, info = env.step(action)
-    reward += rew
-    assert(obs.shape == (400, 400, 3))
-    env.render()
-    action = gym_fluids.agents.fluids_supervisor(obs, info)
-
-assert(reward >= 0)
+successes = 0
+n = 10
+for t in range(n):
+    reward = 0
+    for i in range(100):
+        obs, rew, done, info = env.step(action)
+        reward += rew
+        assert(obs.shape == (400, 400, 3))
+        env.render()
+        action = gym_fluids.agents.fluids_supervisor(obs, info)
+    if (reward) >= 0:
+        successes += 1
+assert(successes / n > 0.8)
 
 
 del(env)