Projects

Development of Multi-Purpose Unmanned Ground Vehicle (Firefighting)
Sponsor:  Hyundai Rotem
Duration: 2025-2026

This project supports the development of a multi-purpose unmanned ground vehicle (UGV) for firefighting applications. We focus on designing and implementing control algorithms for both forward and reverse driving, as well as developing autonomous navigation systems optimized for a six-wheeled vehicle platform. The system is being tailored for robust operation in hazardous fire-response environments.

Development of Integrated Control System for Smoke Control Equipment in Accessory Rooms
Sponsor:  Confidential
Duration: 2025-2026

This project develops an integrated control system for the smoke‐control equipment installed in residential accessory rooms (e.g., utility closets). Under varying door–open/close conditions, it coordinates supply‐air fans and dampers with the exhaust (outflow) system to maintain a target differential pressure and achieve required smoke‐exhaust velocity.

Development of Transformer-Based Time-Series Factory Yield Prediction Algorithm
Sponsor:  LG Display (AI/BigData Optimization Team)
Duration: 2023-2024

This project implements a Transformer-based time-series forecasting framework to predict overall production yield in complex, multi-step manufacturing processes (e.g., semiconductor and display fabrication). By modeling intermediate inspections, tests, and defect occurrences across each process stage, the algorithm provides an estimate of final yield before completion. Ongoing work focuses on improving data consistency and incorporating diverse, unforeseen event data to further enhance prediction robustness.

Development of AI-Based Welding Defect Classification Algorithm for Display Manufacturing (2023 LG Awards)
Sponsor:  LG Display (AI/BigData Optimization Team)
Duration: 2022-2023

This project develops an AI-driven classification framework to automatically inspect micrometer-scale welds in display production using camera imagery. Manual OK/NG assessment introduced human error and high labor costs, jeopardizing downstream processes. We trained a classification model on a curated dataset of weld images, achieving over 99% accuracy in defect detection. By automating the inspection step, the solution delivers significant cost savings in quality control and markedly reduces overall defect rates.

Development of a Tree-Based Machine Learning Model for PCB Specification and Price Prediction (2022 LG Display CFO Awards)
Sponsor:  LG Display (AI/BigData Optimization Team)
Duration: 2020-2021

This project develops a tree-based machine learning framework that, given a bill of materials, predicts key PCB parameters—layer count, board size, and board type—and computes an estimated procurement price. Trained on historical development data, the model achieves over 95% accuracy in price estimation. By enabling cost forecasting from preliminary schematics, this tool empowers engineers and buyers to negotiate better prices and optimize design choices before detailed layouts are completed.

Development of multi-objective re-entry and re-exit trajectory optimization for space-shuttle surveillance missions
Sponsor: Lockheed Martin, UK
Duration: 2016-2017

This project presents a multi-objective trajectory optimization framework for spaceplanes performing suborbit-orbit-suborbit surveillance missions. The proposed method integrates Fuzzy Satisfactory Goal Programming (FSGP) and a Radau Pseudospectral Method (RPM) to address conflicting objectives such as minimizing fuel consumption and total heat flux while maximizing terminal mass. Simulation results demonstrate the feasibility and efficiency of the proposed re-entry and re-exit trajectory for future reusable space platforms.

Development of multi-agent routing and scheduling for UTM(UAS Traffic Management)
Collaborator: SESAR
Duration: 2016-2018

This research presents a novel flight planning algorithm for Unmanned Aircraft Systems (UAS) operating in route network-based airspace. The algorithm aims to minimize flight time for each UAS while ensuring safe separation throughout the mission. It combines a decentralized inner loop, where each UAS computes its shortest path, with a centralized outer loop that allocates flights sequentially. The method effectively handles multi-agent routing and scheduling in complex airspace environments, supporting future UTM operations.

Separation-Compliant Multi-Flight Routing and Scheduling in Terminal Manoeuvring Areas Using Time-Based Airspace Graphs
Collaborator: SESAR
Duration: 2016-2018

This study proposes a routing and scheduling algorithm for multiple flights operating in a Terminal Manoeuvring Area (TMA). Using a time-weighted airspace graph, the algorithm generates separation-compliant flight paths and speed profiles under speed constraints. A First Come First Served logic assigns arrival sequences. Through a case study with EUROCONTROL data, the approach shows promise as a decision support tool for ATCOs to enhance runway throughput and operational efficiency.

User-Centric Charging Station Placement via Game-Theoretic and Clustering-Based Optimization
Collaborator: Volvo GTT, Swedish Electromobility Centre
Duration: 2018-2020

This project introduces a decentralized game-theoretic framework, called k-GRAPE, for optimal placement of electric vehicle charging stations. The algorithm accounts for user preferences and station crowdedness, aiming to maximize overall user utility. Combining k-means clustering with strategic decision-making, the approach guarantees at least 50% optimality and outperforms prior methods by 17%. Its effectiveness is validated through real-world data experiments and comparative analysis.

Reinforcement Learning-Based Personalized Pricing Strategy for Competitive EV Charging Markets
Collaborator: Volvo GTT, Swedish Electromobility Centre
Duration: 2018-2020

This project proposes a Personalized Dynamic Pricing Policy (PeDP) for fast electric vehicle charging stations using reinforcement learning. A multi-agent simulation environment models strategic EV user behavior, considering price, waiting time, and state of charge. A Q-learning algorithm optimizes pricing to maximize station revenue. Results reveal the impact of waiting time, the advantage of adaptive pricing, and potential misuse of privacy-preserved information in competitive EV charging markets.

Can AI Abuse Personal Information?
Collaborator: Swedish Electromobility Centre
Duration: 2018-2020

This project explores the ethical implications of AI-driven dynamic pricing for fast EV charging stations. A model-free reinforcement learning agent is trained to maximize revenue based on user behavior data. Simulations reveal that, without constraints, AI can learn to exploit private user information. To prevent such misuse, the study proposes privacy-aware design principles and policy guidelines, ensuring fair and responsible AI deployment in intelligent transportation systems.

Reinforcement Learning-Based Charging Scheduling for A Fleet of Autonomous Electrified Trucks
Collaborator: Volvo GTT, Swedish Electromobility Centre
Duration: 2018-2020

This project develops a fully unmanned fleet of VERA electric trucks operating 24/7 to maximize delivery throughput. A multi-agent simulation—fed by real-world driving data—models strategic charging decisions, waiting times, charging durations, and speed control. Hybrid rule-based and reinforcement-learning algorithms optimize policies for charging and routing. Results to date demonstrate safe, efficient operation and validate the approach in experimental trials.