Toward a Better IPA Experience for a Connected Vehicle by Means of Usage Prediction

Recently, a connected vehicle comes into wide use. A connected vehicle enables seamless access from outside of the vehicle. A user can easily control the vehicle remotely as if it is a home appliance device. IPA (Intelligent Personal Assistants) or smart speaker adoption is growing rapidly these years. We have already released an IPA application for remote control of a vehicle to enhance accessibility of our connected vehicle. Since the launch, we found significant drop of users from using application, which is reportedly caused by social or mental obstacles to use IPA in some researches. To mitigate these obstacles, we propose proactive suggestion by IPA. We apply a vehicle usage prediction model on our connected vehicle data. The prediction enable spontaneous suggestion by IPA, therefore the user doesn't need to speak wake- up word. In our model, firstly LDA automatically discover usage patterns and clustering method on the latent space. The result suggests some user segments which represent habitual vehicle usage. The prediction is performed by dynamic Bayesian networks for each user segment. An evaluation with actual service data confirms applicability of the prediction model and potential of IPA log itself can also assist vehicle usage prediction.