Bayesian intention inference for trajectory prediction with an unknown goal destination

Contextual cues can provide a rich source of information for robots that operate in the presence of other agents such as people, animals, vehicles and fellow robots. We are interested in context, in the form of the behavioural intent of an agent, for enhanced trajectory prediction. We present a Bayesian framework that estimates both the intended goal destination and future trajectory of a mobile agent moving among multiple static obstacles. Our method is based on multi-modal hypotheses of the intended goal, and is focused primarily on the long-term trajectory of the agent. We propose a computationally efficient solution and demonstrate its behaviour in a pedestrian scenario with a real-world data set. Results show the benefits of our method in comparison to traditional trajectory prediction methods and illustrate the feasibility of integration with higher-level planning algorithms.