Modern automotive systems have been equipped with a highly increasing number of onboard computer vision hardware and software, which are considered to be beneficial for achieving eco-driving. This article combines computer vision and deep reinforcement learning (DRL) to improve the fuel economy of hybrid electric vehicles. The proposed method is capable of autonomously learning the optimal control policy from visual inputs. The state-of-the-art convolutional neural networks-based object detection method is utilized to extract available visual information from onboard cameras. The detected visual information is used as a state input for a continuous DRL model to output energy management strategies. To evaluate the proposed method, we construct 100 km real city and highway driving cycles, in which visual information is incorporated. The results show that the DRL-based system with visual information consumes 4.3-8.8% less fuel compared with the one without visual information, and the proposed method achieves 96.5% fuel economy of the global optimum-dynamic programming.

Energy management technologies have significant potential to optimize electric vehicle performance and support global energy sustainability. However, despite extensive research, their real-world application remains limited due to reliance on simulations, which often fail to bridge the gap between theory and practice. This study introduces a real-world data-driven energy management framework based on offline reinforcement learning. By leveraging electric vehicle operation data, the proposed approach eliminates the need for manually designed rules or reliance on high-fidelity simulations. It integrates seamlessly into existing frameworks, enhancing performance after deployment. The method is tested on fuel cell electric vehicles, optimizing energy consumption and reducing system degradation. Real-world data from an electric vehicle monitoring system in China validate its effectiveness. The results demonstrate that the proposed method consistently achieves superior performance under diverse conditions. Notably, with increasing data availability, performance improves significantly, from 88% to 98.6% of the theoretical optimum after two updates. Training on over 60 million kilometers of data enables the learning agent to generalize across previously unseen and corner-case scenarios. These findings highlight the potential of data-driven methods to enhance energy efficiency and vehicle longevity through large-scale vehicle data utilization.

The vehicle dynamics model is a fundamental prerequisite for advanced software development for intelligent vehicles. This incites the need for accurate mathematical modeling to match the driving dynamics of real vehicles as closely as possible. However, an accurate vehicle model has a variety of parameters that often rely on massive real-vehicle testing to identify and calibrate. This process is a laborious and tedious task for automotive engineers. In this paper, we introduce APOVD, Automatic Parameter Optimization of Vehicle Dynamics, a Bayesian optimization framework that can search the abundant vehicle parameters automatically and efficiently. APOVD inherits the reliability and interpretability of physics-based vehicle models while enjoying the benefits of data-driven methods, that is, the ability to adapt and improve from the data. First, an 8-Degree-of-Freedom dynamics model is developed for a four-wheel independent drive (4WID) electric vehicle. Then, APOVD is used to tune the vehicle model parameters to close the gap between the real vehicle and the simulated vehicle model. Finally, the modeling accuracy of different parameters, various vehicle configurations, and different optimizers is compared in real driving data and CarSim-based simulation data. In the experiments, Bayesian optimization provided accurate vehicle parameters (more than 90% reduction in error) and effectively corrected for incorrect parameters.

Accurate spatial-temporal traffic flow forecasting is essential for helping traffic managers take control measures and drivers to choose the optimal travel routes. Recently, graph convolutional networks (GCNs) have been widely used in traffic flow prediction owing to their powerful ability to capture spatial-temporal dependencies. However, designing the spatial-temporal graph adjacency matrix, which is essential to the success of GCNs remains an open question. This paper proposes a GCN-based traffic flow forecasting method that reconstructs the binary adjacency matrix via tensor decomposition. We first reformulate the spatial-temporal fusion graph adjacency matrix into a three-way adjacency tensor. Then, we use Tucker decomposition to reconstruct the adjacency tensor, encoding more informative and global spatial-temporal dependencies. Finally, we propose multiple Spatial-temporal Tensor Graph Convolution layers that assemble a Spatial-temporal Synchronous Graph Convolutional module for localized spatial-temporal correlations learning and a Dilated Convolution module for global correlations learning in parallel. This enables the comprehensive spatial-temporal dependencies of the road network to be aggregated and learned. Experimental results on four open-access datasets demonstrate that the proposed model outperforms state-of-the-art approaches in terms of prediction performances.

In recent years, the pervasive deployment and progression of autonomous driving technology have engendered heightened demands, particularly within the intricate campus and surrounding environments frequently traversed by autonomous delivery vehicles, such as automated food delivery and courier services. Accurately predicting pedestrian trajectories is paramount in the realm of autonomous driving. In the face of complex scenarios within campus and surrounding environments, traditional pedestrian trajectory prediction methods have failed to achieve satisfactory results. To address this challenge systematically, this paper employs a digital twin methodology to establish a novel dataset, denoted as the vulnerable pedestrian trajectory prediction dataset (VPT), grounded in the authentic road network structures of six campuses and their environs. This paper proposed a UTD-PTP trajectory prediction framework based on the diffusion model, which seeks to forecast pedestrian trajectories in settings characterized by heightened pedestrian traffic, disorderliness, and irregularities. Importantly, the applicability of our proposed methodology extends beyond campus environments, showcasing commendable performance on standard autonomous driving datasets. Experimental results reveal an average enhancement of 0.03 in ADE and 0.05 in FDE on publicly available datasets. On the VPT dataset, our method demonstrates substantial improvements of 0.12 in ADE and 0.38 in FDE relative to the baseline model. Overall, our proposed method exhibits superiority in pedestrian trajectory prediction models, substantially reinforcing confidence in the safety of vulnerable road users in autonomous driving.