AI4G-7-Planning Lecture Notes PDF
Document Details
Uploaded by ThrilledDenver
LMU München
2024
Matthias Schubert
Tags
Summary
These lecture notes cover deterministic planning in the context of artificial intelligence for games. They discuss deterministic problems within games, sequential planning, and assume static and fully-observable environments. The notes focus on maximizing cumulative rewards or minimizing cumulative costs.
Full Transcript
Lecture Notes on Artificial Intelligence for Games Summer Semester 2024 Deterministic Planning Script © 2020 Matthias Schubert Deterministic Problems in Games very few deterministic games are interesting solving subtasks for AI Engines: –...
Lecture Notes on Artificial Intelligence for Games Summer Semester 2024 Deterministic Planning Script © 2020 Matthias Schubert Deterministic Problems in Games very few deterministic games are interesting solving subtasks for AI Engines: – find exits of a building – route to a target approximate and heuristic solutions for non-deterministic problems often rely on making the problem deterministic and then solve the deterministic solution example: assume that the opponent uses the identical policy as the player in Go or chess. Artificial Intelligence for Games 2 Deterministic Sequential Planning assume : static and fully-observable environments set of states S = {s1,..,sn} Set of actions A(s) for each state sÎS reward function R: R(s) (if negative = cost function) transition function T: S´A => S: t(s,a) = s’ goal: maximize cumulated rewards (minimize cumulated cost) cumulated reward/cost of the episode: ∑$!"# 𝛾 ! 𝑟! with 0 < 𝛾 ≤ 1 (𝛾=1: all rewards count the same, 𝛾
