Building intelligent systems that connect perception, language, and action in the real world.
I work at the intersection of embodied AI, autonomous robotics, vision-language navigation, 3D perception, and intelligent manufacturing. My work spans research prototypes and deployable systems, from laser-scanned industrial inspection to multimodal decision making for outdoor robots.
Education
Ph.D. in Mechanical Engineering, Zhejiang University
| Area | What I build |
|---|---|
| Embodied intelligence | Grounded perception, language-conditioned planning, and action in physical environments |
| Robot navigation | Outdoor local navigation, visual-language decision making, and safe execution |
| 3D vision | Point-cloud synthesis, alignment, defect detection, and laser profilometer calibration |
| Intelligent manufacturing | Automated fiber-placement inspection and real-time industrial vision systems |
|
Multi-view pure-vision language navigation that connects camera observations, language goals, and continuous robot motion. Real-robot demonstrations cover cross-area navigation, semantic object search, access-gate passage, and elevator entry/exit, alongside closed-loop simulation evaluations. Pure Vision · VLN · Real-Robot Demos |
Localization-aided VLN for all-terrain industrial park patrol on the DEEP Robotics Lynx S10. The system combines semantic instructions, Goal Points, and single- or multi-view vision with high-fidelity Gaussian-scene training. Includes the system architecture, core inference interfaces, and 30 segment-level simulation replays. GOAI 2026 Runner-up · Algorithm Innovation Award |
|
POI-guided vision-language decision making for outdoor robot local navigation. The repository includes the public code slice and a visual project page with demos and presentation material. Python · VLM/VLN · Robotics |
Machine-learning and geometric methods for semi-automatic hand-eye calibration of laser profilometers, supporting accurate 3D inspection. Python · Point Clouds · Calibration |
|
An end-to-end text-to-video competition solution covering video selection, captioning, model fine-tuning, and evaluation. Python · Generative Models · Video |
Selected code and experiments are organized around reproducible perception, multimodal reasoning, and robot deployment workflows. Research prototypes · Reproducibility · Systems |
- A novel 3D laser scanning defect detection and measurement approach for automated fibre placement · Measurement Science and Technology, 2021
- An in-process inspection method integrating deep learning and classical algorithm for automated fiber placement · Composite Structures, 2022
- A novel hand-eye semi-automatic calibration process for laser profilometers using machine learning · Measurement, 2023
- A data-efficient framework for domain-adaptive laser point cloud synthesis in automated fiber placement defect detection · Advanced Engineering Informatics, 2025
- Integrated Real-Time Automated Fiber Placement Inspection System with High-Speed Point Cloud Alignment and Fusion · MEAI, 2025
- Addressing data scarcity in automated fiber placement defect detection via semantically controlled 3D synthesis · Composites Part B, 2026
- Zero- and Few-Shot Defect Detection in CFRP Manufacturing with Vision-Language Large Models · 2026
- A Vision-LiDAR Fusion Approach for Dynamic Obstacle Point Cloud Filtering in Mobile Robot Mapping · 2026
- POI-Guided Vision-Language Decision Making for Outdoor Robot Local Navigation — forthcoming
- Encoding Approach and Benchmark for Structured Laser-Scan Point Clouds in Automated Fiber Placement Inspection — under revision
- Distributed Spatiotemporal Cooperative Navigation for Ground Robot Swarms — accepted, pending publication
5th place — ModelScope-Sora Challenge
Competition solution and engineering notes →
For research or engineering collaboration: line_cliche@foxmail.com
Project videos, PPT walkthroughs, and interactive demonstrations are maintained on their respective project pages so each repository stays focused and reproducible.