Replication package for the article "STVR: Semantic teleoperation video reconstruction for low-bandwidth remote driving" (Journal of Intelligent and Connected Vehicles). It contains the code of the experiments (STVR, the evaluation, the four H.265 baselines, the packet-channel simulator and the closed-loop simulation), the inputs and intermediate outputs of the runs on the 150 nuScenes validation scenes, the CARLA assets the receiver draws, the records underlying Tables 2–4 and Figs. 1 and 3–7, scripts that regenerate the tables and figures, a runnable example and a self-check. README.md and the explanatory file give the instructions. Not included: nuScenes (https://www.nuscenes.org), the model weights (all public; fetch_weights.sh), and the closed-loop drive recordings and the 3D Gaussian splatting reconstruction made from them, which show pedestrians' faces and licence plates. Code: MIT License; files derived from nuScenes: CC BY-NC-SA 4.0; CARLA assets: CC BY 4.0.

This paper proposes ATH (Adversarial Trajectory Head), a framework for generating safety-critical scenarios to test autonomous driving systems. It addresses "who to attack" via Vision-Language Model (VLM) semantic reasoning over front-view images and structured states, and "how to move" through anchor-conditioned truncated diffusion synthesis. generator applies gradient-based guidance in trajectory coordinate space during DDIM denoising (4 steps), preserving kinematic feasibility while inducing collisions.  closed-loop causal danger scorer captures comfort disturbances beyond direct collisions. ATH distills guided-search experience by replacing natural anchors with NHTSA collision priors and retraining on high-danger trajectories, achieving one-pass generation  without online optimization. Experiments on NavSim with four planners (TransFuser, IDM, DiffusionDrive，PDM-Closed) demonstrate attack effectiveness (collision rate up to 69.50%), physical plausibility, and generation efficiency. Repository includes VLM selection, guided synthesis, danger scoring, and ATH training pipelines with configurations.

This reproduction package provides the complete code used in our paper, "Dangerous Driving Behavior Detection Method in Vehicle Intelligent Cockpit Based on TransWindow Multimodal Fusion" (Manuscript No.: JICV-2025-0093.R2). It includes the source code of the proposed TransWindow model, the baseline method used for comparison experiments, environment configuration, and data output images. This resource is of significant reference value to researchers and engineers who wish to reproduce this experiment and verify the research results.

Replication Package for “Choreographing CAV Formations on the Fly with Reflection”

All materials related to this study have been made available through ETS Data and an external public data repository for transparency and reproducibility purposes. The ETS Data replication package contains the main output results reported in the manuscript, the source code, and the corresponding computational environment configuration files. Due to its large size, the complete raw experimental dataset is hosted separately on Hugging Face and is publicly accessible at: https://huggingface.co/datasets/InterestingITS/smartphone_in_vehicle_multisensor