A Recursive Algorithm for Performance Analysis of Complex Stochastic Behavior Tree

Behavior trees are increasingly widely used for task descriptions, due to their advantages in modularity and reactivity. The performance analysis of behavior trees plays an important role in quantitative evaluations of tasks. When analyzing behavior tree performance using the theory of stochastic behavior tree, one of the most important steps is converting the stochastic behavior tree to corresponding Markov chain. However, when faced with complex behavior trees, the manual conversion is time consuming and err-prone. In this paper, we proposed a recursive algorithm which can convert arbitrarily complex scale and structure behavior tree to corresponding Markov chain, and record the corresponding transfer matrix. On this basis, key performance indicators such as success probability and failure probability can be calculated. This algorithm provides an operable method for performance analysis of complex stochastic behavior trees.