Our project develops a smart logistics autonomous driving system for trucks using a vision-language model (Mini-InternVL2-DA-DriveLM) integrated with the UC-win/Road simulator. A three-layer architecture—data perception, VLM reasoning, and control execution—transforms scene understanding into concise natural-language driving suggestions, then into throttle, brake, and steering commands. We implement prompt engineering, COM-based data capture, and real-time decision loops on modest hardware (RTX 4060 8GB). Fine-tuning with DriveGPT4-augmented datasets improves adaptability, safety, and interpretability across highways, urban roads, and ports. Scenario tests evaluate safety, compliance, comfort, and efficiency, demonstrating a feasible, explainable, and cost-effective path toward intelligent, unmanned freight transportation for future real-world deployments.

Our project aims to build a Digital Twin-based Smart Navigation Lock System that integrates fast 3D reconstruction, real-time ship tracking, and hazardous cargo ship detection to enhance the management of ship locks. The system uses a fast reconstruction algorithm to generate realistic 3D models of ships , while a YOLOv11-based tracking algorithm combined with AIS information ensures accurate, real-time positioning of ships during lock transit. Additionally, a ResNet-based classification model is employed to detect dangerous goods sjip and trigger visual warnings on a centralized display.

This project develops an intelligent digital twin system for gas stations. It utilizes YOLOv11 and Qwen2.5-VL for real-time vehicle tracking and anomaly detection, creating a synchronized 3D virtual replica via UC-win/Road. The system provides managers with a visual monitoring and early-warning platform, while a mobile app offers users services like real-time status checking, reservation, and contactless payment. It aims to enhance safety, optimize operational efficiency, and enable a smarter refueling experience.

Research Contents:
(1) Establish an EEG data acquisition platform tailored for drivers in autonomous vehicles, ensuring real-time signal capture.
(2) Construct a simulated environment replicating autonomous driving failures to provoke takeover scenarios.
(3) Collect pre-takeover EEG datasets and develop a machine learning algorithm for early intent recognition with 90% accuracy.
(4) Engineer an integrated brain-computer interface system enabling seamless software-hardware coordination for emergency vehicle control.

FlowSync is a dynamic urban traffic simulation platform built exclusively with UC‑win/Road and its SDK. The idea is to create an interactive environment where users can model, test, and optimize traffic flow in a simulated cityscape. By altering traffic parameters in real time, the simulator can demonstrate how various traffic management strategies perform under diverse conditions—from everyday congestion to emergency scenarios.

Based on the location information received from the UC-win/Road simulator, this project is a system that predicts the future location of moving objects and determines the risk of collision through an LSTM predictor. The prediction results are analyzed in voxel units, and the risk is calculated according to the degree of overlap and reflected in vehicle control. In addition, through object recognition, the risk is adjusted by applying weights to certain classes, such as children. It is a system in which a language model detects risk, explains the situation to the driver, and directly takes control when necessary.

This research leverages UC-win/Road as a simulation platform, integrating autonomous driving and Human-Machine Interface (HMI) technologies for the development of a digital twin system. High-fidelity Unmanned Aerial Vehicle observation data facilitates realistic traffic flow simulation. Scene fidelity is enhanced by integrating road-specific Building Information Models (BIM) and satellite imagery. Furthermore, autonomous driving plugins enabling networked communication, learning, decision-making, planning, and control were developed, alongside a LUA-scripted HMI system, constituting a next-generation digital twin framework.

Smart Motion Drive introduces an innovative interaction between people and autonomous vehicles by integrating motion capture technology with driving systems. Using Sony’s mocopi sensors attached to six body points, user motions such as raising a hand are translated into car control commands. The system supports two modes: Self-Driving Mode, enabling acceleration, turning, and stopping through gestures, and Auto-Driving Mode, allowing users to call or return a vehicle remotely. Implemented with Unity, Python, and UC-win/Road, the system empowers seniors and individuals with disabilities by promoting inclusive, joyful mobility within the Web 4.0 era, bridging digital and real-world transportation experiences.

As the automotive industry shifts to autonomous and electric vehicles, failure to respond to road obstacles has become a critical safety issue, not just a matter of discomfort. Current systems are limited by their reliance on non-real-time HD maps and a perception focus biased towards dynamic hazards over road surface details. Our system overcomes these limitations with a simulator-based, self-learning loop of ‘Perception → Classification → Control → Evaluation → Calibration’. This enables real-time detection and response without HD maps, ensuring safety and efficiency while continuously optimizing performance.

This project is a real-time safe driving score analysis system developed by integrating the UC-win/Road simulator with a Python-based dashboard. The system receives live vehicle dynamics data via UDP, including speed, acceleration, and yaw rate.
Our algorithms evaluate seven types of dangerous driving behaviors, such as speeding, sudden braking, and abrupt turns, based on criteria from official traffic safety agencies. The driver's score, starting at 100, is penalized for risky actions and rewarded for safe driving. All score changes are instantly visualized on a real-time graph, providing immediate, objective feedback to the driver.