AGen: Adaptable Generative Prediction Networks for Autonomous Driving

摘要: In highly interactive driving scenarios, accurate prediction of other road participants is critical for safe and efficient navigation autonomous cars. Prediction challenging due to the difficulty in modeling various behavior, or learning such a model. The model should be reflect individual differences. Imitation methods, as parameter sharing generative adversarial imitation (PS-GAIL), are able learn models. However, learned models average out When used predict trajectories vehicles, these biased. This paper introduces an adaptable framework (AGen), which performs online adaptation offline recover differences better prediction. particular, we combine recursive least square algorithm (RLS-PAA) with from PS-GAIL. RLS-PAA has analytical solutions adapt every single vehicle efficiently online. proposed method reduce root mean squared error 2.5 s time window by 60%, compared