Скачать или смотреть Apprenticeship Learning via Inverse Reinforcement Learning

Implementation of Apprenticeship Learning via Inverse Reinforcement Learning.

The green regions in the world are positive and the blue regions are negative (more green means more positive). We use a discount factor of 0.9.

The demonstrations are the execution of optimal policy using the reward function.

We use 10 demonstrations from different states in the grid world. We perform IRL, and find a policy which tries to control the system from each of these initial states. Each demonstration and learned behavior are indexed and shown at the top of the video.


Link to paper: http://ai.stanford.edu/~ang/papers/ic...

Credits to Prof. Abbeel and Prof. Andrew Ng for inventing the method