Back to Course
Reinforcement Learning: Agents
Module 10 of 11
10. Inverse RL (IRL)
1. The Reward Problem
Designing a perfect Reward Function is hard. (The "King Midas" problem). If you reward "Speed", the car crashes.
2. Learning from Humans
Instead of guessing the reward, learn it from expert demonstrations. Observing a human driver entails inferring what they are trying to optimize (Safety + Speed).