Jianming Hu

Power grid large language models (LLMs) require the efficient acquisition of dense regulatory knowledge while preserving lightweight deployment. We propose a three-stage framework for the efficient training of power grid LLMs via token-adaptive continual pretraining and orthogonal subspace on-policy self-distillation. First, Zone of Proximal Development (ZPD)-guided curriculum selection ranks domain texts using prerequisite knowledge, uncertainty, paraphrase stability, and explanation-induced learning gains, prioritizing samples that are both learnable and informative. Second, Privileged-View Self-Distillation Continual Pretraining (PVD-CPT) estimates token-level knowledge value from asymmetric student–teacher views, strengthens critical numerical, conditional, and normative tokens, and anneals auxiliary weighting and distillation during full-data coverage training. Third, Dual-Teacher Subspace-Guided On-Policy Self-Distillation (DSG-OPSD) evaluates weak and strong views of a frozen teacher along fresh constrained greedy trajectories, estimates a periodically refreshed foundational-correction subspace in the centered full-vocabulary logit space, and retains orthogonal rationale-and-error guidance through spectral-energy and output-uncertainty gating. On a 1024-question single-answer multiple-choice power-grid standards benchmark, where the overall score is question-level accuracy, PVD-CPT improves the score from 19.63 to 32.71. The same DSG-OPSD-trained checkpoint scores 43.26 under direct-answer inference and 49.41 under reasoning inference; the additional 6.15 points are therefore attributable to the inference protocol. On this target-standard benchmark, 49.41 is higher than 44.63 for the strongest evaluated external baseline, Qwen3.5-397B-A17B. Because the external models were evaluated without task-specific adaptation to the 50-standard source collection, this comparison positions target-domain specialization under asymmetric training exposure rather than establishing general-purpose or unseen-standard superiority. The results demonstrate effective domain knowledge acquisition and answer-policy refinement without additional inference-time teachers or auxiliary inputs.

Coordinated control of multiple intersections remains a major challenge in urban traffic systems due to complex local–global dependencies. To address this issue, the MPNNLight framework is proposed, a message-passing graph neural network for cooperative multi-intersection traffic signal control. The core innovation of this work is the proposed Attention-Wavelet Spatial Transformer (AWSformer), a message computation module that combines self-attention mechanisms and 2D discrete wavelet transform (2D-DWT) to adaptively compute inter-agent messages across spatial and frequency domains. This design enables efficient extraction of multi-scale dependencies and enhances coordination among intersections. Experiments on diverse datasets demonstrate that MPNNLight achieves superior performance and efficiency, compared with existing GNN-based traffic signal control methods.

Testing and evaluation are critical to the development and deployment of autonomous vehicles (AVs). Given the rarity of safety-critical events such as crashes, millions of tests are typically needed to accurately assess AV safety performance. Although techniques like importance sampling can accelerate this process, it usually still requires too many tests for field testing. This severely hinders the testing and evaluation process, especially for third-party testers and governmental bodies with very limited testing budgets. The rapid development cycles of AV technology further exacerbate this challenge. To fill this research gap, this paper introduces the few-shot testing (FST) problem and proposes a methodological framework to tackle it. As the testing budget is very limited, usually smaller than 100, the FST method transforms the testing scenario generation problem from probabilistic sampling to deterministic optimization, reducing the uncertainty of testing results. To optimize the selection of testing scenarios, a cross-attention similarity mechanism is proposed to extract the information of AV’s testing scenario space. This allows iterative searches for scenarios with the smallest evaluation error, ensuring precise testing within budget constraints. Experimental results in cut-in scenarios demonstrate the effectiveness of the FST method, significantly enhancing accuracy and enabling efficient, precise AV testing.

Realistic traffic simulation is critical for ensuring the safety and reliability of autonomous vehicles (AVs), especially in complex and diverse urban traffic environments. However, existing data-driven simulators face two key challenges: a limited focus on modeling dense, heterogeneous interactions at urban intersections - which are prevalent, crucial, and practically significant in countries like China, featuring diverse agents including motorized vehicles (MVs), non-motorized vehicles (NMVs), and pedestrians - and the inherent difficulty in robustly learning high-dimensional joint distributions for such high-density scenes, often leading to mode collapse and long-term simulation instability. We introduce City Crossings Dataset (CiCross), a large-scale dataset collected from a real-world urban intersection, uniquely capturing dense, heterogeneous multi-agent interactions, particularly with a substantial proportion of MVs, NMVs and pedestrians. Based on this dataset, we propose IntersectioNDE (Intersection Naturalistic Driving Environment), a data-driven simulator tailored for complex urban intersection scenarios. Its core component is the Interaction Decoupling Strategy (IDS), a training paradigm that learns compositional dynamics from agent subsets, enabling the marginal-to-joint simulation. Integrated into a scene-aware Transformer network with specialized training techniques, IDS significantly enhances simulation robustness and long-term stability for modeling heterogeneous interactions. Experiments on CiCross show that IntersectioNDE outperforms baseline methods in simulation fidelity, stability, and its ability to replicate complex, distribution-level urban traffic dynamics.

Research on the longitudinal control of mixed platoons has garnered significant attention, particularly addressing the uncertainties of human-driven vehicles (HDVs) and the computational inefficiency of centralized control for large platoons. This paper presents a novel distributed control strategy for multiple mixed platoons, utilizing a flexible model structure to account for dynamic uncertainties and lateral disturbances. A distributed control framework is introduced, where each platoon’s states and constraints are coupled with neighboring platoons, improving scalability and resilience in large-scale mixed traffic environments. Each platoon’s dynamic uncertainties are characterized by convex hulls in the subsystem model. A distributed data-driven input mapping method is proposed, where subsystems dynamically map the process data linearly from both local and neighboring platoons to local control law, eliminating the need for pre-collected datasets or offline training. To reduce iterations and communication requirements, a distributed implementation algorithm is proposed by designing subsystem basic laws with robust model predict control, then the control problem is solved online using convex quadratic programming, enabling fast computation of control actions and enhancing adaptability to real-time traffic variations. Additionally, a theoretical analysis guarantees the stability and optimality of the system, providing conditions that assist in subsystem parameter design for practical implementation. Simulation results validate the effectiveness and efficiency of the proposed approach, particularly in comparison to traditional methods that rely on pre-collected data.

The assessment of safety performance plays a pivotal role in the development and deployment of connected and automated vehicles (CAVs). A common approach involves designing testing scenarios based on prior knowledge of CAVs (e.g., surrogate models), conducting tests in these scenarios, and subsequently evaluating CAVs’ safety performances. However, substantial differences between CAVs and the prior knowledge can significantly diminish the evaluation efficiency. In response to this issue, existing studies predominantly concentrate on the adaptive design of testing scenarios during the CAV testing process. Yet, these methods have limitations in their applicability to high-dimensional scenarios. To overcome this challenge, we develop an adaptive testing environment that bolsters evaluation robustness by incorporating multiple surrogate models and optimizing the combination coefficients of these surrogate models to enhance evaluation efficiency. We formulate the optimization problem as a regression task utilizing quadratic programming. To efficiently obtain the regression target via reinforcement learning, we propose the dense reinforcement learning method and devise a new adaptive policy with high sample efficiency. Essentially, our approach centers on learning the values of critical scenes displaying substantial surrogate-to-real gaps. The effectiveness of our method is validated in high-dimensional overtaking scenarios, demonstrating that our approach achieves notable evaluation efficiency.

Testing and evaluation are critical to the development and deployment of autonomous vehicles (AVs). Given the rarity of safety-critical events such as crashes, millions of tests are typically needed to accurately assess AV safety performance. Although techniques like importance sampling can accelerate this process, it usually still requires too many numbers of tests for field testing. This severely hinders the testing and evaluation process, especially for third-party testers and governmental bodies with very limited testing budgets. The rapid development cycles of AV technology further exacerbate this challenge. To fill this research gap, this paper introduces the few-shot testing (FST) problem and proposes a methodological framework to tackle it. As the testing budget is very limited, usually smaller than 100, the FST method transforms the testing scenario generation problem from probabilistic sampling to deterministic optimization, reducing the uncertainty of testing results. To optimize the selection of testing scenarios, a cross-attention similarity mechanism is proposed to learn to extract the information of AV's testing scenario space. This allows iterative searches for scenarios with the smallest evaluation error, ensuring precise testing within budget constraints. Experimental results in cut-in scenarios demonstrate the effectiveness of the FST method, significantly enhancing accuracy and enabling efficient, precise AV testing.