The proper functioning of connected and autonomous vehicles (CAVs) is crucial for the safety and efficiency of future intelligent transport systems. Meanwhile, transitioning to fully autonomous driving requires a long period of mixed autonomy traffic, including both CAVs and human-driven vehicles. Thus, collaborative decision-making technology for CAVs is essential to generate appropriate driving behaviors to enhance the safety and efficiency of mixed autonomy traffic. In recent years, deep reinforcement learning (DRL) methods have become an efficient way in solving decision-making problems. However, with the development of computing technology, graph reinforcement learning (GRL) methods have gradually demonstrated the large potential to further improve the decision-making performance of CAVs, especially in the area of accurately representing the mutual effects of vehicles and modeling dynamic traffic environments. To facilitate the development of GRL-based methods for autonomous driving, this paper proposes a review of GRL-based methods for the decision-making technologies of CAVs. Firstly, a generic GRL framework is proposed in the beginning to gain an overall understanding of the decision-making technology. Then, the GRL-based decision-making technologies are reviewed from the perspective of the construction methods of mixed autonomy traffic, methods for graph representation of the driving environment, and related works about graph neural networks (GNN) and DRL in the field of decision-making for autonomous driving. Moreover, validation methods are summarized to provide an efficient way to verify the performance of decision-making methods. Finally, challenges and future research directions of GRL-based decision-making methods are summarized.

As one of the main elements of reinforcement learning, the design of the reward function is often not given enough attention when reinforcement learning is used in concrete applications, which leads to unsatisfactory performances. In this study, a reward function matrix is proposed for training various decision-making modes with emphasis on decision-making styles and further emphasis on incentives and punishments. Additionally, we model a traffic scene via graph model to better represent the interaction between vehicles, and adopt the graph convolutional network (GCN) to extract the features of the graph structure to help the connected autonomous vehicles perform decision-making directly. Furthermore, we combine GCN with deep Q-learning and multi-step double deep Q-learning to train four decision-making modes, which are named the graph convolutional deep Q-network (GQN) and the multi-step double graph convolutional deep Q-network (MDGQN). In the simulation, the superiority of the reward function matrix is proved by comparing it with the baseline, and evaluation metrics are proposed to verify the performance differences among decision-making modes. Results show that the trained decision-making modes can satisfy various driving requirements, including task completion rate, safety requirements, comfort level, and completion efficiency, by adjusting the weight values in the reward function matrix. Finally, the decision-making modes trained by MDGQN had better performance in an uncertain highway exit scene than those trained by GQN.

In the autonomous driving process, the decision-making system is mainly used to provide macro-control instructions based on the information captured by the sensing system. Learning-based algorithms have apparent advantages in information processing and understanding for an increasingly complex driving environment. To incorporate the interactive information between agents in the environment into the decision-making process, this paper proposes a generalized single-vehicle-based graph neural network reinforcement learning algorithm (SGRL algorithm). The SGRL algorithm introduces graph convolution into the traditional deep neural network (DQN) algorithm, adopts the training method for a single agent, designs a more explicit incentive reward function, and significantly improves the dimension of the action space. The SGRL algorithm is compared with the traditional DQN algorithm (NGRL) and the multi-agent training algorithm (MGRL) in the highway ramp scenario. Results show that the SGRL algorithm has outstanding advantages in network convergence, decision-making effect, and training efficiency.

The decision-making algorithm serves as a fundamental component for advancing the level of autonomous driving. The end-to-end decision-making algorithm has a strong ability to process the original data, but it has grave uncertainty. However, other learning-based decision-making algorithms rely heavily on ideal state information and are entirely unsuitable for autonomous driving tasks in real-world scenarios with incomplete global information. Addressing this research gap, this paper proposes a stable hierarchical decision-making framework with images as the input. The first step of the framework is a model-based data encoder that converts the input image data into a fixed universal data format. Next is a state machine based on a time series Graph Convolutional Network (GCN), which is used to classify the current driving state. Finally, according to the state’s classification, the corresponding rule-based algorithm is selected for action generation. Through verification, the algorithm demonstrates the ability to perform autonomous driving tasks in different traffic scenarios without relying on global network information. Comparative experiments further confirm the effectiveness of the hierarchical framework, model-based image data encoder, and time series GCN.

The 4-D radar sensing provides rich measurements across Doppler, range, azimuth, and elevation dimensions, offering strong resilience in adverse environmental conditions. However, reconstructing accurate 3-D occupancy maps from 4-D radar tensors (4DRT) remains challenging due to low spatial resolution, nonuniform sampling, and artifacts caused by sidelobes and multipath reflections. This article proposes a diffusion-based framework for 3-D occupancy reconstruction that directly leverages the encoded structure of 4DRT data. The pipeline consists of three components: a feature-preserving dimensionality reduction module that produces Doppler-aware sparse descriptors from raw 4DRT; a hierarchical pillar-based representation that encodes vertical geometry using normalized height segments within a compact and structured spatial format; and a conditional diffusion model that iteratively denoises latent occupancy predictions. To encode the sparse and irregular 4DRT inputs, we design a hybrid condition encoder that combines convolutional layers with self-attention to extract both local and global contextual features. These are injected into a U-Net-based denoising network via cross-attention to generate dense and spatially consistent 3-D occupancy volumes. Extensive experiments on the Coloradar benchmark and real-world driving data demonstrate that the proposed method consistently outperforms the state-of-the-art baseline, improving intersection over union (IoU) from at most 3.3% to over 30% across diverse scenes, while reducing the Chamfer distance (CD) by more than $4\times $ and maintaining a single-frame inference latency of approximately 50 ms. The implementation is available at https://github.com/MoYuGit/Diffusion_radar_reconstruction

This paper proposes an adaptive visible-infrared (V-I) multi-constraint fused SLAM system for illumination robust localization. It organizes the two modalities into a primary/ auxiliary scheme: the primary stream is selected by feature-tracking stability with a hysteresis mechanism, while auxiliary features are continuously weighted according to their incremental spatial complementarity measured by the surprisal of the primary feature-density distribution. The resulting weighted visual factors are jointly optimized with IMU preintegration and vehicle-motion constraints (forward velocity and kinematic constraints with zero-velocity updates when applicable) in a tightly-coupled sliding-window framework to reduce drift under degraded vision. We further adopt an illumination-adaptive hierarchical relocalization pipeline to improve place recognition across day-night transitions. Experiments on public and custom datasets demonstrate consistently improved robustness and accuracy over single-modality and direct fusion baselines under both stable and varying illumination.