Generalisation

Q-learning now has its built-in generalisation using gradient descent over linear combination of policy variables. Also, add another sample falling stones to demonstrate how generalisation is used.