The future motion of surrounding vehicles is key for autonomous vehicles to make safe decisions. However, motion prediction is a challenging task and may fail when encountering unseen data. In this study, an ensemble graph neural network-based motion prediction model is first established to predict the future motion of surrounding vehicles. Then, the designed prediction model is trained and tested via Next Generation Simulation (NGSIM) datasets. Specifically, the motion prediction not only predicts the future motion of surrounding vehicles but also outputs the risk of the prediction results. In addition, an uncertainty-aware motion planning framework is then developed based on model predictive control techniques, which integrates both prediction results and corresponding risk information. Finally, simulation results demonstrate the effectiveness of the proposed method.
Uncertainty-Aware Motion Planning for Autonomous Driving on Highway
28.10.2022
5132425 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
LIMITED VISIBILITY AND UNCERTAINTY AWARE MOTION PLANNING FOR AUTOMATED DRIVING
British Library Conference Proceedings | 2018
|