carla-simulator / imitation-learning

Repository to store conditional imitation learning based AI that runs on CARLA.
MIT License
433 stars 124 forks source link

Carla detailed settings #60

Open JunningHuang opened 5 years ago

JunningHuang commented 5 years ago

Can you provide detailed carla settings? @nsubiron @felipecode @nishanthjois Including weather, NPC car number, pedestrian number etc...

Without the detailed settings in Carla, it's hard to reproduce your work

JunningHuang commented 5 years ago

BTW, I have load your model to test in Town01, but I find that the car didn't stop when the traffic light turns red... Can you tell me the reason? @felipecode @nsubiron @nishanthjois

markus-hinsche commented 5 years ago

@Johnnyswong It is not stopping in the training data recorded, therefore it is not what the model learned to do.

SunFishlike commented 5 years ago

Your work is truly attractive and I am trying to reproduce the benchmark. However, I see no NPC vehicles when I run the benchmark. Is thit normal?

As I find that each episode will reset the client with default CarlaSettings, and the default CarlaSettings at least have 20 NPC vehicles (but I see None). Anyway, I tried to solve this, but I have no idea now.

@felipecode @nsubiron

Default setting from the Carla 0.8.2: class CarlaSettings(object): """ The CarlaSettings object controls the settings of an episode. The str method retrieves an str with a CarlaSettings.ini file contents. """

def __init__(self, **kwargs):
    # [CARLA/Server]
    self.SynchronousMode = True
    self.SendNonPlayerAgentsInfo = False
    # [CARLA/QualitySettings]
    self.QualityLevel = 'Epic'
    # [CARLA/LevelSettings]
    self.PlayerVehicle = None
    self.NumberOfVehicles = 20
    self.NumberOfPedestrians = 30
    self.WeatherId = 1
    self.SeedVehicles = None
    self.SeedPedestrians = None
    self.set(**kwargs)
    self._sensors = []