Successful low-level policies training except Wait maneuver.

Each low level policy was retrained with better LTL conditions and rewards, some parts of which are also designed to encourage exploration (to prevent the vehicle from being stopped all the time).

backends/controller_base.py

@@ -56,8 +56,7 @@ class ControllerBase(PolicyBase):
         total_reward = 0
         self.node_terminal_state_reached = False
         while not self.node_terminal_state_reached:
-            observation, reward, terminal, info = self.low_level_step_current_node(
-            )
+            observation, reward, terminal, info = self.low_level_step_current_node()
             if visualize_low_level_steps:
                 self.env.render()
             total_reward += reward

options/simple_intersection/maneuver_base.py

@@ -29,7 +29,7 @@ class ManeuverBase(EpisodicEnvBase):
     # _extra_action_weights_flag = True); note that a cost is defined
     # as a negative reward, so a cost will be summed up to the reward
     # with subtraction.
-    _cost_weights = (1.0 * 1e-3, 1.0 * 1e-3, 0.25 * 1e-3, 1.0 * 1e-3,
+    _cost_weights = (10.0 * 1e-3, 10.0 * 1e-3, 0.25 * 1e-3, 1.0 * 1e-3,
                      100.0 * 1e-3, 0.1 * 1e-3, 0.25 * 1e-3, 0.1 * 1e-3)
 
     _extra_r_terminal = None
@@ -292,9 +292,8 @@ class ManeuverBase(EpisodicEnvBase):
         raise NotImplemented(self.__class__.__name__ +
                              ".generate_learning_scenario is not implemented.")
 
-    def generate_validation_scenario(
-            self
-    ):  # Override this method in the subclass if some customization is needed.
+    # Override this method in the subclass if some customization is needed.
+    def generate_validation_scenario(self):
         self.generate_learning_scenario()
         self._enable_low_level_training_properties = False