Development of A Suggestive System to Aid Hand Crafting of Behavior Trees

Behavior trees (BTs) are a common automation approach in game development and robotics. Implementing efficient BTs and foreseeing situations requires experience and experiments. To overcome such difficulties, we propose a suggestive system to aid handcrafting of BTs by suggesting expansions of incomplete sub-trees based on a set of collected data. Our model uses recursive auto-encoders to transform trees into a vectorized space and find adjacent trees to the incomplete subject. We indicate the performance of our approach by illustrating expanded trees and assessing their quality. The results show that expanded trees have significantly higher quality than the initially collected trees.

کلیدواژه ها:

Behavior Trees ، Learning from Demonstration ، Recursive Auto-Encoders.