An environment driven model of human navigation intention for mobile robots

In order to effectively do motion planning in environments inhabited by humans, robots must accurately predict human motion. A model of human navigation intent is one way to do this. This paper builds a probabilistic model of human navigation intent that is driven by the structure of the environment. Given functional places are combined with automatically extracted navigation way-points to build a model of navigation goals within the environment. Additionally a probabilistic model of human motion is proposed which is used to build arrays of probability distributions for a variety of path shapes. An on-robot laser range sensor is used to detect humans, and in conjunction with the human motion model, the probability distribution grids, and the navigation target map, is used to model human intention in the shared navigation space. The motion model's predictive capability is compared against a constant velocity model using real human motion data.