Evaluating Pre-trained Predictor Models of Pedestrian Destinations for a Voice Guidance Robot
Asami Ohta, Satoshi Okano, Nobuto Matsuhira, Yuka Kato · 2019
In recent year, there has been increasing interest in communication robots, and a variety of services including voice guidance are expected for such robots. For providing those services, state estimation of robot users is required. From the background, we have been studying a method to predict the walking direction of a pedestrian who heads toward a robot in order to provide suitable voice guidance to him/her by a communication robot installed at the reception desk of a public facility. In this paper, we verify the effectiveness of the proposed method by using actual observed data. Here, we measure pedestrian trajectories using a laser range scanner installed on a tripod and predict the branching direction using pre-trained predictor models by a machine learning algorithm. In this paper, we generated two predictor models using an open dataset of pedestrian trajectories in a shopping mall. By conducting evaluation experiments using the models, we found out that one model can predict the direction with practical accuracy but the accuracy of another one is not sufficient. The result shows that using robust and adequate predictor models are important for our target system.