Learning Domain Knowledge for Classical Planning
||Learning Domain Knowledge for Classical Planning|
||Simon Ståhlberg <email@example.com>|
||2020-10-26 – 2021-11-01|
A typical solver for classical planning considers each problem in isolation. That is, the solver have no prior knowledge about the domain of the problem. This leads to inefficiencies when the solver is given many problems in the same domain since it will likely perform the same mistakes several times. In this project we look at learning domain specific knowledge from example problems that can be reused for any problem in the domain. The extracted knowledge can be used to improve existing solvers and enable them to handle larger problems. Furthermore, the knowledge is also understandable for humans. This is not typical for machine learning, where the result is normally seen as a black box.