Large vision-language models (LVLMs) have demonstrated remarkable capabilities in autonomous driving scene understanding tasks. However, these models occasionally generate hallucinatory texts, resulting in descriptions that seem reasonable but do not correspond to the image. To address this challenge, this paper proposes HELTA, a training-free data annotation method used in traffic scenarios, which is designed to support the offline generation of high-quality semantic datasets without hallucinations. Specifically, HELTA employs a cross-checking mechanism to filter entities and directly extracts critical objects from the given image, enriching the descriptive text. Experimental results on the POPE benchmark demonstrate that HELTA improves the F1-score of the Mini-InternVL-4B and mPLUG-Owl3 models by 12.58% and 4.28%, respectively. Additionally, qualitative results using images collected in open campus scene further highlight the practical applicability of the proposed method. Compared with the GPT-4o model, HELTA achieves comparable descriptive performance while significantly reducing costs. Finally, two high-quality semantic understanding datasets, CODA_desc and nuScenes_desc, are created for traffic scenarios to support future research. The codes and datasets are publicly available athttps://github.com/fjq-tongji/HELTA

Reinforcement learning (RL) has become a key method for decision-making in autonomous vehicles, particularly in tasks like path planning and obstacle avoidance. However, RL often struggles with sparse or delayed reward signals, which can hinder learning. A promising solution is incorporating a teacher-student framework, where the agent learns from a teacher. While these methods can accelerate learning, they risk sub-optimal outcomes if the teacher's guidance is flawed. This paper introduces Double Entropy Reinforcement Learning (DERL), which enables the student to initially rely on the teacher and gradually shift towards independent exploration as it surpasses the teacher's performance. Experiments show that DERL outperforms existing teacher-student algorithms in learning efficiency and effectiveness, reducing reliance on imperfect guidance.

In complex real-world traffic environments, autonomous vehicles (AVs) need to interact with other traffic participants while making real-time and safety-critical decisions accordingly. The unpredictability of human behaviors poses significant challenges, particularly in dynamic scenarios, such as multi-lane highways and unsignalized T-intersections. To address this gap, we design a bi-level interaction decision-making algorithm (BIDA) that integrates interactive Monte Carlo tree search (MCTS) with deep reinforcement learning (DRL), aiming to enhance interaction rationality, efficiency and safety of AVs in dynamic key traffic scenarios. Specifically, we adopt three types of DRL algorithms to construct a reliable value network and policy network, which guide the online deduction process of interactive MCTS by assisting in value update and node selection. Then, a dynamic trajectory planner and a trajectory tracking controller are designed and implemented in CARLA to ensure smooth execution of planned maneuvers. Experimental evaluations demonstrate that our BIDA not only enhances interactive deduction and reduces computational costs, but also outperforms other latest benchmarks, which exhibits superior safety, efficiency and interaction rationality under varying traffic conditions.

3D multi-object tracking is an important component of the perception module in autonomous driving systems. Due to the limitations of a single sensor, tracking methods based on either LiDAR or cameras always have certain deficiencies. Fusion-based tracking methods have received increasing attention. However, existing fusion-based tracking methods often underutilize image information, ignore the respective effects of appearance information and 2D detection results, and lack further analysis on the simultaneous use of both. This paper proposes a novel camera-LiDAR fusion tracking framework that primarily relies on the motion model using 3D objects. It fully leverages the appearance information and 2D detection results simultaneously from images and introduces three modules to reduce the number of false positive samples, false negative samples and ID switches, respectively. Besides, the entire tracking process does not require global processing and achieves online tracking. The proposed method achieves competitive results on the KITTI tracking dataset with 78.50% HOTA. Compared with EagerMOT using the same 3D and 2D detectors, the HOTA metric improved by 4.11%. Code is available on https://github.com/zengwz/FusionTrack.

This paper proposed a timescale graph-parallel (GP) computation method to solve the real-time optimization problem of nonlinear predictive energy-saving control, thus to realize the implementation of MPC on vehicle on-board controllers. The proposed scheme consists of two parts: forward prediction of the objective function and backpropagation of the partial differential function, both of which can be calculated in parallel. Thus, compared with traditional serial solution method for optimization problems, the timescale graph-parallel computation method can utilize the computing resources of the controller fully. In this paper, firstly, based on the characteristics of commercial vehicles, a mixed integral optimal control problem (MIOCP) was constructed. Then, a detailed timescale graph-parallel computation algorithm was derived for the MIOCP. Finally, GP and Pontryagin’s Minimum Principle (PMP) algorithms were applied on the predefined road for the simulation of the prediction of energy-saving control for commercial vehicles. The simulation results showed that compared with PMP, the maximum iteration number, average iteration number, single longest solution time, and single average solution time of the proposed GP decreased by 60%, 64.28%, 89.53%, and 93.56%, respectively. In addition, GP can also improve fuel efficiency by 1.55% without sacrificing much power performance.

Real-time and high-performance semantic segmentation is a crucial task in the scene understanding of autonomous vehicles. This paper focuses on this issue and proposes a transformer and convolutional neural networks (CNN) hybrid encoder-decoder structure SegTransConv. Firstly, we present a four-stage hierarchical encoder, and the feature extractor in each stage is composed of two transformer layers and CNN modules in series. In this way, the encoder better exploits the global contexts of the input and expands the receptive fields. In the U-shape decoder, the feature maps are upsampled through the proposed feature enhancement upsampling module (FE_Up). Then the knowledge distillation strategy is leveraged to improve the model performance under the guidance of the teacher network STDCNet. Finally, a novel evaluation metric is designed to comprehensively assess the accuracy, speed, floating-point operations (FLOPs), and parameters of real-time segmentation methods. Extensive experiments on two public datasets and self-collected images have evaluated the effectiveness of our method. SegTransConv-A and SegTransConv-B obtain 72.8% and 73.0% mIoU, respectively, at the inference speed of 68.0 FPS with an input resolution of$1024\times 512$.

This study aims to develop an optimal cruise controller that can automatically adapt to individual car-following style. First, the adaptive cruise control (ACC) problem is formulated as a linear quadratic optimal control, and an optimal control law containing the longitudinal acceleration of the target vehicle is derived. Then, a certain number of individual car-following styles are predefined on the basis of the proposed optimal cruise controller. Thereafter, a car-following style learning algorithm is proposed to quantify the closeness of the predefined individual car-following style to the specific driver, and a proper style is thus determined for the specific driver by using this learning algorithm. On the basis of the learned car-following style, the proposed optimal cruise controller can adapt itself to individual car-following style. Finally, the proposed self-learning optimal cruise controller is evaluated through simulation and experimental tests. Results show that the control behavior of the proposed self-learning optimal controller is closer to that of the human driver than that of a factory-installed ACC.