研究Research
主軸：以人工智慧解決天線設計中的不可微分與二值化最佳化問題。完整內容請見各專案詳情頁。Focus: applying AI to the non-differentiability and binarized-optimization challenges of antenna design. See each project page for the full write-up.
學術論文Publication
海報發表Poster
已接收Accepted
A Data-Free Patch Antenna Generation Framework via Two-Stage Gradient Exploration and BiScaleNorm
- 作者Author
- 吳維文 (Wei-Wen Wu) 等et al.
- 會議Venue
- URSI General Assembly and Scientific Symposium (URSI GASS 2026)
- 地點 / 時間Location / Date
- 克拉科夫，波蘭 · 2026 年 8 月Kraków, Poland · August 2026
- 場次Session
- B10 — Machine learning and optimization techniques in electromagnetics
碩士論文M.S. Thesis
基於 ACP 與連通性損失之貼片天線生成式反向設計ACP with Connectivity Loss for Generative Inverse Design of Patch Antennas
一套無需資料集、無需 GPU 的線上深度學習框架，自動生成可製造的 28 GHz 貼片天線金屬圖樣。A data-free, GPU-free online deep-learning framework that generates manufacturable 28 GHz patch-antenna metal patterns.
- 三項機制：ACP（自適應循環策略）、SC Loss（圖譜連通損失 · Fiedler 值）、DLF（動態損失過濾），解決局部解、金屬孤島與樣本效率問題。Three mechanisms — ACP (adaptive cyclical policy), SC Loss (spectral connectivity / Fiedler value), and DLF (dynamic loss filter) — address local optima, metal islands, and sample efficiency.
- 全系統響應損失 0.99 dB、饋入連通性 18.89% → 61.98%（HFSS 驗證）；28 GHz · RO4003C · 純 CPU 無 GPU。Full-system response loss 0.99 dB, feed reachability 18.89% → 61.98% (HFSS-validated); 28 GHz · RO4003C · CPU-only, no GPU.
大學專題Undergraduate Capstone
機械手臂視覺抓取系統研發Research & Development of a Robotic-Arm Visual Grasping System
結合 YOLOv7 與 OpenCV 的六軸機械手臂視覺抓取，同時取得物體的類別、位置與角度。A six-axis robotic-arm grasping system fusing YOLOv7 and OpenCV to obtain an object's class, position, and angle at once.
- YOLOv7 偵測（類別＋位置）+ OpenCV 最小外接矩形（角度）+ RealSense D435 深度，動態 FOV 校正後以比例控制與逆向運動學驅動 AR3 六軸手臂。YOLOv7 detection (class+position) + OpenCV min-area rectangle (angle) + RealSense D435 depth; dynamic FOV calibration, then proportional control and inverse kinematics drive the AR3 arm.
- 能正確分離相鄰／重疊物體並連續處理；詳情頁含實機 Demo 影片。Correctly separates adjacent/overlapping objects and handles them sequentially; the detail page includes a live demo video.
參與計畫Funded Projects
- 具適應性之主被動式上肢外骨骼機器人系統開發及其於復健之應用Adaptive active-passive upper-limb exoskeleton robot system for rehabilitation
- 調控電磁環境之可重構智慧面雛形平台開發Reconfigurable intelligent surface (RIS) prototype platform for electromagnetic-environment control
- 知識定義與需求工程技術於智慧型系統非功能性需求塑模之研究Knowledge-definition and requirements-engineering methods for modeling non-functional requirements of intelligent systems