Ziyu Song

Lane-level high-precision localization is essential for autonomous driving, yet it remains highly challenging under Global Navigation Satellite System (GNSS) signal degradation and complex urban road conditions. To address these challenges, this paper proposes a unified low-cost localization framework that relies on a GNSS receiver, a MEMS-grade IMU, and a forward-facing camera. The proposed system integrates an Extended Kalman Particle Filter (EKPF), an enhanced Hidden Markov Model (HMM) based map-matching algorithm, and a vision-based lane-line-assisted correction module. The enhanced HMM incorporates a hierarchical road-to-lane search, an adaptive intersection-aware state transition model, and a multi-criteria lane-change detection mechanism to ensure reliable performance in dense urban environments. Building upon stable lane identification, the lane-line-assisted module utilizes camera-detected geometric features to correct lateral drift and improve robustness under GNSS degradation. Simulation and real-world experiments demonstrate that, while meeting real-time requirements, the proposed system achieves over 92% lane-level matching accuracy and maintains above 91% accuracy under GNSS degradation conditions with up to 4.0 m GNSS noise. The mean positioning error reaches 0.29 m under urban canyon conditions, confirming the framework's robustness, accuracy, and practicality for low-cost lane-level autonomous vehicle localization.

Autonomous driving in complex urban environments requires trajectory planning that balances safety, efficiency, and human-like behavior. Although imitation learning (IL) can capture expert driving patterns from large-scale demonstrations, existing IL-based planners still face challenges in safety-critical scenarios and long-tail traffic distributions. Meanwhile, optimization-based planners provide explicit constraint handling but are often separated from upstream learning modules, limiting their ability to jointly improve trajectory generation and planning feasibility. To address these issues, we propose a hybrid trajectory planning framework that integrates IL-based multimodal trajectory proposal with differentiable optimization. In the proposed framework, an IL backbone generates candidate ego trajectories and surrounding-agent predictions, while a differentiable optimizer refines the selected trajectory using multi-objective cost functions with learnable weights related to safety, efficiency, and comfort. This design enables optimization objectives and constraints to provide gradient feedback to the upstream planning network, improving the consistency between candidate generation and downstream planning objectives. In addition, we introduce a surrounding agent centric data augmentation strategy that reuses real-world trajectories of surrounding vehicles as additional expert demonstrations, thereby enriching complex interaction and long-tail scenarios without extra data collection. Closed-loop experiments on the nuPlan benchmark show that the proposed method achieves a composite score of 94.04, outperforming PLUTO’s 93.14 while using only 30% of the training data. The results demonstrate that the proposed framework improves closed-loop planning performance, trajectory feasibility, and data efficiency under complex urban driving scenarios.

Planning safe trajectories on slippery roads is a fundamental and challenging task, and with existing architectures, the trajectory passed to the tracking controller may be conservative or impossible to track. In this article, an enhanced motion planning strategy for safe automated lane change on slippery roads is proposed. The proposed strategy consists of a reference trajectory generator and trajectory optimization. The generator first establishes a lane change trajectory cluster based on a quintic polynomial, and safety envelopes based on vehicle dynamics are considered constraints to eliminate infeasible trajectories. Then, a cost function is designed to express inaccurate obstacle positions, irregular obstacle movements, and split friction risks and facilitate the generation of a reference trajectory. Finally, based on high-quality initial solutions, trajectory optimization is performed to improve the smoothness of the trajectories, and the safety of trajectories between adjacent discrete waypoints is ensured by increasing the size parameter. The proposed strategy is validated in near-natural simulations. The experimental results demonstrate that the strategy improves the stability of the generated trajectories and satisfies ride comfort requirements for slippery roads, the maximum steering wheel angle and lateral acceleration are reduced by 44%–55% and 62%, respectively, compared to those of the optimization-based method. Additional comparisons with learning-based methods show that our method yields smoother steering commands and consistently lower peak yaw rate and sideslip responses.

Accurate spatial reasoning and risk assessment from monocular video on consumer electronics platforms are prerequisites for safe decision-making in autonomous vehicles, yet general-purpose vision–language model (VLM) remains unreliable at lane-level localization and temporally grounded risk estimation. We present a deployment-oriented, spatially enhanced VLM that learns implicit 3D reconstructive information from monocular sequences via a reconstructive 3D tokenization pipeline and a spatial–visual fusion encoder built on MobileVLM. Concretely, the model fuses 3D reconstructive tokens with CLIP visual tokens through cross-attention to produce 3D-aware visual tokens, enabling lane-level localization and depth-aware reasoning about object orientation and inter-object relations. The unified decoder jointly generates scenario descriptions, object identities and positions, and converts spatial estimates into interpretable multi-level risk scores using a Time-to-Collision (TTC) mapping. Experimental results show that our system outperforms strong VLM baselines for multi-level risk classification on the nuScenes dataset. The generalization in CARLA and CoVLA further corroborates robust spatial reasoning and risk assessment. The findings indicate that coupling a lightweight VLM with monocular 3D cues via cross-attentional spatial–visual fusion yields accurate spatial localization and interpretable, transferable risk estimation that is robust under shift and practical for edge deployment in downstream tasks.

In smart-city deployments, vast spatiotemporal trajectory data are generated by consumer-electronics sensors, such as in-vehicle devices. Predicting the future movements of traffic participants in such complex environments is a fundamental task in autonomous driving. In this study, we proposed a trajectory prediction strategy based on attention mechanism and self-supervised learning. First, the agent and map features are extracted, and a set of searched candidate paths for the target agent is constructed based on prior knowledge. Leveraging these candidate paths, a feature selection mechanism is introduced to enable the model to adaptively select relevant feature vectors. This enhances the attention mechanism’s ability to model agent interactions. Furthermore, self-supervised learning is utilized to design auxiliary tasks, including path classification and agent interaction strength estimation, which further improve trajectory prediction performance. Extensive experiments on the nuScenes dataset demonstrate that the prediction model is competitive with the state-of-the-art models, reducing minADE by up to 32.6% and minFDE by 40.8% compared to the latest DICE model, while maintaining real-time inference with an average runtime of 14.9 ms per frame for 8 agents. Ablation studies indicate that each component is essential to ensure prediction performance. Finally, closed-loop Vehicle-in-the-Loop experiments validate the model’s ability to generate meaningful multimodal predictions and effectively integrate them into downstream planning tasks, the prediction-aware planner yields enhanced safety and comfort over existing baselines, and its low computational cost highlights strong potential for deployment on consumer electronics platforms, laying the groundwork for intelligent spatiotemporal analytics in smart-city traffic management.