Traffic-Aware Motion Prediction and Deep Reinforcement Learning-Based Energy Management for Autonomous Driving

Abstract

Autonomous driving requires accurate, traffic-aware motion prediction to offer various advantages over manual driving, including convenience, reduced traffic collisions, fuel savings and mobility efficiency. Understanding driver intentions and improving methods for predicting the trajectories of surrounding vehicles is crucial for adaptive traffic-aware autonomous driving. Predictive Energy Management Systems (EMS) and motion planning are integral components of autonomous driving. In this work, a unique application of iTransformer (inverted Transformer) neural network is explored to address the challenges of highway trajectory prediction with a focus on traffic awareness for autonomous driving in Battery Electric Vehicles (BEV). The focus of the iTransformer lies in predicting the speed of the ego vehicle, leveraging traffic-aware data to enhance motion prediction accuracy in complex scenarios. The training of the iTransformer model is done using the popular open-source NGSIM dataset. The performance of the proposed traffic-aware iTransformer model by comparing it against traffic-aware and non-traffic-aware state-of-the-art deep learning models, such as Long Short-Term Memory (LSTM) and Vanilla Transformer (VT). iTransformer improves the RMSE by 71.69% and 57.31% when compared against traffic-aware LSTM and VT, respectively. Similarly, in a non-traffic-aware environment, iTransformer outperforms LSTM and VT by 65.17% and 51.43%, respectively, showing its ability to learn robust motion patterns without traffic context. These predicted trajectories are integrated into a Deep Reinforcement Learning (DRL) based EMS that ensures the optimal utilization of energy resources by incorporating forecasts of future driving conditions. EMS is critical in balancing energy consumption across the vehicle's systems, including computing units, Heating, Ventilation, and Air Conditioning (HVAC). It outlines the design of state variables, action variables, and reward functions for energy optimization and describes the integration of real-time surrounding traffic data to enhance energy efficiency in dynamic driving scenarios. The EMS aims to extend battery life by improving the State of Health (SOH) and reducing operational costs. In summary, this research introduces the iTransformer network for the first time to address the challenges of highway trajectory prediction, with a focus on enhancing traffic awareness for autonomous BEVs. By leveraging real-time traffic data and predictive modeling, the iTransformer enhances motion planning and Deep Reinforcement Learning (DRL) based energy management strategies. This advancement will strengthen iTransformer for the development of sustainable intelligent transportation systems.