What does the expert policy describe?

jangirrishabh / toyCarIRL

Implementation of Inverse Reinforcement Learning Algorithm on a toy car in a 2D world problem, (Apprenticeship Learning via Inverse Reinforcement Learning Abbeel & Ng, 2004)

MIT License

172 stars 47 forks source link

What does the expert policy describe? #6

Closed matriculus closed 5 years ago

matriculus commented 5 years ago

I would like to know what those 8 values in expert policy define. I want to implement in another system for obstacle avoidance. I don't understand how the trajectory is provided to the irl agent.

jangirrishabh commented 5 years ago

Hello @pradeeshenator the values in the expert policy are Feature expectations which is the sum of discounted feature values. These feature values are defined manually for this particular agent, feel free to experiment with them. For more information check out the blog IRL/https://jangirrishabh.github.io/2016/07/09/virtual-car-IRL/