Discrete and continuous, probabilistic anticipation for autonomous robots in urban environments

Abstract

This paper explores representations for capturing the anticipation of other objects by an autonomous robot in an urban environment. Predictive Gaussian mixture models are proposed due to their ability to probabilistically capture continuous and discrete obstacle behavior; the predictive system uses the probabilistic output of a tracking system (current obstacle location), and map (with lanes and intersections). The probabilistic tracking and anticipated motion are integrated into an optimized path planner. This paper explores various levels of model abstraction to understand how complex these predictive models must be in order to create a more robust path planning algorithm.