Imitation Learning

Imitation learning techniques aim to mimic human behavior in a given task.

Learning rewards through demonstration.

Some applications:

  • Simulated highway driving
  • Aerial imagery-based navigation
  • Parking lot navigation
  • Quadrupled locomation

Inverse RL

Inverse RL is more about understanding the underlying reward structure by observing the behavior, such as why animals or humans move / behave in a certain way.

Behavioral Cloning

Treat this as a standard Supervised Learning problem.

Problem: Compounding errors.

Adversarial Imitation Learning (AIL)