Get the latest tech news
Solving Path of Exile Item Crafting with Reinforcement Learning
Background #PoE (Path of Exile) is an ARPG similar to Diablo 2/3/4 or Last Epoch.
But in our case the branching factor is due to stochasticity in the environment, which we must take an expectation over, meaning that we cannot easily prune away paths like we could when choosing an action. This function assigns a value to each state-action pair, representing the expected sum of rewards when taking action \(a\) in state \(s\) and following the optimal policy thereafter. The crafting process is rather simple, but in the second step the optimal action is to use a “Cannot Roll Attack Mods” metamod in combination with an Orb of Annulment, which wouldn’t be immediately obvious to many players.
Or read this on Hacker News
