⌖ HKG 22.3058° N, 114.1734° E EDITION NO. 0026 / 1000
DISPATCH Virtualization & Architecture →
張 張壽峰 / Shoufeng Zhang
ARCHIVE SPECIMEN · VOL. MMXXVI · EDITION NO. 0026 / 1000
SYSTEMS · STREETS · STORIES

Along the Way

I am drawn to beauty, spare elegance, the way a few symbols can hold an entire slice of reality. I am equally drawn to complexity, the patience of decomposing systems to their atoms, then the triumph of constructing something whole from first principles. Beyond, I observe. Photographing, reading, letting the inner lives of others resonate against my own.

Open to collaboration · University of Hong Kong
EXPLORE FOLIO ↓
壹 About
⌖ RESEARCH PROFILE // OVERVIEW · HKG 22.3°N

I'm a sophomore undergraduate student at the University of Hong Kong, under the Faculty of Engineering Elite Programme, majoring in Electronic Engineering with a minor in Computer Science. Prior to university, I studied at Beijing 101 High School.

I currently work on autonomous robotics. My broader interests span hardware-software co-design, edge deployment of intelligent systems, and the system-level questions that arise when autonomous systems leave controlled environments and meet the real world. I am drawn to research at the intersection of perception, planning, and the hardware constraints that make both hard.

「莫聽穿林打葉聲，何妨吟嘯且徐行。竹杖芒鞋輕勝馬，誰怕？一蓑煙雨任平生。」

貳 Education

The University of Hong Kong

2025.9 – Present

B.Eng., Engineering Elite Programme, Electronic Engineering (Minor: Computer Science)

GPACGPA: 3.96/4.3 (Major CGPA: 4.15/4.3)
Awards
  • 2025-26 The University of Hong Kong Entrance Scholarship
  • 2025-26 Martin Scholarship, St. Johns College, The University of Hong Kong
Courses
Data Structures & AlgorithmsComputer OrganizationLinear AlgebraMultivariable CalculusObject-Oriented Programming

Carnegie Mellon University

2024.7

Summer Session

GPAQPA: 4.0 / 4.0
Courses
Principles of Imperative Computation (C, invariants, memory management)Concepts of Mathematics (Discrete Math)

Beijing 101 High School

2022.9 – 2025.7

AP & IB Diploma Programme

叁 Experience & Works
01
2026.7 – Present · Undergraduate Research Assistant (Supervisor: Prof. Fu Zhang; Mentor: Siqi Liang)

MaRS Lab, HKU — Embodied AI & World Models

  • Formulating predictive world-action models and Vision-Language-Navigation (VLN) policies via SparseVideoNav (predictive video generation world model) for zero-shot navigation across complex, dynamic, unmapped environments.
  • Integrating real-time perception-action loops and predictive video navigation policies onto quadrupedal (Unitree Go2) and humanoid robotics platforms.
Embodied AIRoboticsWorld ModelsVLN
02
2025.12 – Present · Research Assistant (Part-time & Full-time Summer 2026; Advisor: Dr. Lei Yang)

HKU Innovation Wing — Autonomous Robotics & Spatial Perception

  • 3D Spatial Perception & Tracking GitHub: Built an end-to-end perception pipeline lifting open-vocabulary 2D detections into metric 3D bounding boxes via LiDAR–camera sensor fusion and frustum clustering; formulated metric spatio-temporal association to track 158 persistent infrastructure assets across 1,100+ paired inspection frames.
  • High-Performance C++ Engine: Re-architected mapping and grounding into a standalone compiled C++ pipeline decoupled from ROS2 message deserialization overhead. Slashed end-to-end dataset processing time from 30 minutes to under 1 minute (>30× speedup) and cut peak memory footprint from >32GB (resolving live OOM crashes) to <3.5GB across 1,100+ image pairs during operational metro station trials.
  • Autonomy & MTR Field Deployment: Integrated Nav2, FAST-LIO (LiDAR-inertial odometry), and scan-matching localization; eliminated odometry drift across featureless station floors and dynamic crowd occlusions during operational MTR field trials.
  • Museum Guide Robot (2025.12 – 2026.5): Built the full autonomy stack for a museum guide robot. Mitigated corridor drift and dynamic crowd occlusions by integrating FAST-LIO odometry, ICP localization, and Nav2 path planning.
Autonomous Robotics3D Spatial PerceptionHigh-Performance SystemsSLAM
03
2025.12 – 2026.5 · Undergraduate Research Assistant (Advisor: Prof. Hao Chen)

HKU Systems Lab — Multi-Agent Systems & Program Reasoning

  • RE-Agent: LLM-Enhanced Reverse Engineering Agent GitHub: Built an automated binary analysis agent translating expert human reverse-engineering workflows into a 3-phase automated pipeline (Triage & Hypothesis Generation, Context Assembly & Naming, Write-back & Refinement) grounded in Ghidra via Model Context Protocol (Ghidra MCP). Employs 5-layer non-linear BFS context expansion with tiktoken budget management, raising Exact Match from 7.8% to 71.2% (Token F1 from 28.4% to 74.1%, a 9.1× gain) across 172 stripped functions from 50 binaries.
  • INFRASCOPE: Vulnerability Variant Detection in AI Infra (arXiv:2605.20051, under review at USENIX Security 2027): Co-authored a reference-driven multi-agent framework that extracts transferable vulnerability semantics from known disclosures to detect variants across AI infrastructure. Analyzed 688 GitHub repositories and 251 disclosures, and evaluated across 20 production AI infra repositories, uncovering 20+ vulnerabilities with 11 vendor-acknowledged and 4 assigned CVEs.
Multi-Agent SystemsLLM ReasoningProgram Analysis
04
2024 · Project Leader

Hardware-Accelerated 1-Bit Quantization Using PyRTL

  • Designed a 1-bit MAC unit for a Systolic Array with PyRTL for ternary-weight XNOR-Net inference, achieving 95% area reduction and 76% critical path reduction compared to a standard full-precision matrix multiplication MAC unit.
PyRTLSystem DesignQuantization
05
2024 · Participant

Yishengyixin Program, Chinese Academy of Sciences

  • Explored computer architecture and systems. Designed circuits with Verilog/Vivado for RV32 instructions and built a simplified RV32I emulator in C.
VerilogVivadoCRISC-V
肆 Skills

Robotics & Perception

  • ROS2
  • FAST-LIO / LiDAR-Inertial Odometry
  • LiDAR-Camera Sensor Fusion
  • Point Cloud SLAM & State Estimation
  • Nav2 Path Planning & Navigation

Software & ML Systems

  • Modern C++ (C++17/20) & Python
  • PyTorch
  • Vision-Language Models (VLMs)

Hardware & Systems

  • Linux / Git / Docker
  • Verilog / SystemVerilog

Exploring

  • Embodied AI / World-Action Models
  • Vision-Language-Action (VLA) Navigation
  • High-Performance Autonomous Systems
伍 Contact

Feel free to reach out.

Emailzhangshoufeng@connect.hku.hkGitHub@shawnzsfRedNotePhotography & more