Current simulator features are fine for behaviour cloning.
However if you would like to train agents with algorithms that require interaction with environment you're out of luck - if agent get's stuck somewhere out of the road the environment will be useless without manual intervention.
Current simulator features are fine for behaviour cloning.
However if you would like to train agents with algorithms that require interaction with environment you're out of luck - if agent get's stuck somewhere out of the road the environment will be useless without manual intervention.