Abstract
Prediction of surrounding vehicles accurately is an essential prerequisite for safe autonomous driving. Trajectory prediction methods can be classified into physics-, maneuver-, or learning-based methods. Learning-based methods have been studied extensively in recent years because it effectively exploits the road information and interactions among vehicles. However, learning-based methods perform poorly in unseen environments that were not considered during training and provide unreasonable results such as inconsistent trajectories according to road geometry. In this paper, to address this problem, a hybrid model that combines a learning-based model with physics- and maneuver-based models according to their uncertainties is proposed. The deep ensemble technique is also used to estimate the uncertainty of the learning-based method. Because the deep ensemble tends to show a large variance in unseen environments, this method is used to determine whether to use a hybrid model. The proposed method is trained and validated using the Lyft l5 dataset, the real environment vehicle driving data containing several types of intersections.
Highlights
Rapid advances have been made in Advanced driver assistance systems (ADAS) over several years, and systems such as adaptive cruise control (ACC) and lane keeping assistance (LKA), have already been effected in mass-produced vehicles
Because the lead vehicle can increase the accuracy of future trajectory predictions by providing a reference to the future velocity profile, the maneuver-based model improves the prediction performance by using an ACC strategy when there is a lead vehicle
WORKS In this study, a hybrid model integrating learning, physics, and maneuver-based models is proposed for the trajectory prediction
Summary
Rapid advances have been made in Advanced driver assistance systems (ADAS) over several years, and systems such as adaptive cruise control (ACC) and lane keeping assistance (LKA), have already been effected in mass-produced vehicles. Learning-based methods, compared to maneuver- and physics-based methods, make it relatively easier to consider road shape information and interactions among vehicles. When driving in a new environment that was not observed during training, or when the density of surrounding vehicles changes, the deep learning network may receive input data with a different distribution from the training data. This increase the likelihood of the network predicting an unexpected future trajectory, which can lead to accidents due to undesirable planning. An integrated model that combines the trajectory results from physics-, maneuver-, and learning-based method is proposed to improve trajectory prediction performance in various environments. The deep ensemble method is utilized to tackle new environments by estimating the uncertainty of the learning-based model. Performance evaluation is conducted on the Lyft l5 dataset, which is collected under real-world vehicle driving conditions, and includes various road environments such as intersections
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
