Path Planning Using Probability Tensor Flows

Abstract

Probability models are emerging as a promising framework to account for “intelligent” behavior. In this article, probability propagation is discussed to model agent s motion in potentially complex grids that include goals and obstacles. Tensor messages in the state-action space (due to grid structure, states are 2-D and the concomitant probability distributions are represented by 3-D arrays), propagated bi-directionally on a Markov chain, provide crucial information to guide the agent s decisions. The discussion is carried out with reference to a set of simulated grids and includes scenarios with multiple goals and multiple agents. The visualization of the tensor flow reveals interesting clues about how decisions are made by the agents. The emerging behaviors are very realistic and demonstrate great potential for the application of this framework to real environments.