Motion Planning of the Autonomous Vehicles with Multi-view Images and GRUs

Abstract

Motion planning is arguably the most significant module in autonomous driving. Image-based and LiDAR-based representations are very successful but costly. The proposed method considers three images from different angles to understand the ego vehicle's surroundings. The proposed approach decreases the occurrence of infractions, to reduce the violation of traffic signals. The proposed model generates a sequence of coordinates representing the waypoints in the predicted path of the vehicle for the upcoming few time steps and uses an inverse dynamic algorithm to derive the values of the driving parameters, like steering angle, throttle, and brake value, from the coordinates of the waypoints generated by the model. The CARLA simulator is used for data extraction to train and evaluate the model and simulate the motion planning. The proposed approach shows an improvement in the Route Completion and Driving Score metrics compared to the existing methods on the dataset of this simulator. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.