low_level_policy_main module

class low_level_policy_main.ManeuverEvaluateCallback(maneuver)

Bases: rl.callbacks.Callback

on_episode_end(episode, logs={})

Called at end of each episode

on_train_end(logs=None)
low_level_policy_main.evaluate_low_level_policy(maneuver, pretrained=False, nb_episodes_for_eval=100)
low_level_policy_main.low_level_policy_testing(maneuver, pretrained=False, visualize=True, nb_episodes_for_test=20)
low_level_policy_main.low_level_policy_training(maneuver, nb_steps, RL_method='DDPG', load_weights=False, training=True, testing=True, visualize=False, nb_episodes_for_test=10, tensorboard=False, without_ltl=False)

Do RL of the low-level policy of the given maneuver and test it.

Parameters:
  • maneuver – the name of the maneuver defined in config.json (e.g., ‘default’).
  • nb_steps – the number of steps to perform RL.
  • RL_method – either DDPG or PPO2.
  • load_weights – True if the pre-learned NN weights are loaded (for initializations of NNs).
  • training – True to enable training.
  • testing – True to enable testing.
  • visualize – True to see the graphical outputs during training.
  • nb_episodes_for_test – the number of episodes for testing.