DAgger

The algorithm for DAgger is given as follows:

1. Initialize an empty dataset
2. Initialize a policy
3. For iterations i = 1 to N:
   1. Create a policy .
   2. Generate a trajectory using the policy .
   3. Create a dataset by collecting states visited by the policy and the actions of those states provided by the expert . Thus, .
   4. Aggregate the dataset as .
   5. Train a classifier on the updated dataset and extract a new policy .