I am an undergraduate in Software Engineering at Wuhan University. My work runs on three parallel tracks: visual representation under appearance variation, small-model structured scientific reasoning, and multi-agent systems for real tasks. I care more about methods that turn into reproducible, analyzable, deployable systems than isolated benchmark gains.

• 2026.04 — TPAMI appearance-variation ReID survey §2–§8 drafted with 231 references; §1 introduction and literature alignment still iterating.
• 2026.03 — Submitted Reliability-Aware 3D Geometric Injection for Universal Person Re-identification to ECCV 2026.
• 2026.03 — PEARL mainline pipeline ready: graph_spec IR, validator, compiler, issue-list repair, and joint-score evaluation are in place; targeting AAAI submission.
• 2026.02 — SmartPhone Battery: The Moment of Depletion earned MCM 2026 Finalist Award with dual-branch ECM-EKF + PINN approach.
• 2025.11 — Built an A2A-protocol multi-agent system on Holos with routing, validation, fallback, and observability for end-to-end task demos.

• Status: ECCV 2026 submission (2025.10 – 2026.03)
• Focus: Universal ReID under occlusion, clothing change, and cross-modality, with monocular 3D geometry treated as conditionally useful evidence rather than an always-trusted auxiliary.
• Contributions: Proposed UniGeo, decoupled SMPL parameters into global shape and joint topology, introduced a Reliability-Aware Gate, and designed adaptive residual fusion between 2D visual features and 3D structural features.
• Evidence: Unified training, comparisons, ablations, gate-correlation analysis, fusion-complexity checks, and perturbation-stability verification across 9 benchmarks.

• Status: Ongoing, target submission to AAAI (2026.03 – present)
• Framework: Positioned as a complete framework for scientific reasoning graph extraction, organized around "generation-verification decoupling" with training-free data construction layer and student SFT layer.
• Contributions: Designed graph_spec intermediate representation and two-stage semantic-closure workflow in training-free layer; introduced LoRA training pipeline with curriculum learning (nodes → reasoning units → graph_spec → DOT) for small models like Qwen3.5-4B in student SFT layer.
• Training plan: Curriculum learning with high-quality supervision from PEARL-produced graphs targeting small-model scientific reasoning graph generation.
• Progress: Phase 1 mainline code ready; Gate B local metrics passed; main experiments in progress.

• Status: Ongoing, TPAMI survey preparation (2026.03 – present)
• Framework: Reconstructed the field through a three-branch taxonomy of short-term / long-term / compound appearance variation and reorganized existing methods through a robustness vs. generalization lens.
• Progress: Completed §2–§8 main text and 231 references; current work focuses on evaluation blind spots and metric transferability under appearance variation.

• Worked on software architecture for a modular intelligent hardware system and a dual-bus power-communication platform.
• Implemented module coordination, bus arbitration, hot-plug, and bus-merging control logic; supported three PCB iterations and 24V industrial-environment adaptation.
• Achieved 98.5% communication success rate and about 40% efficiency gain over traditional solutions.
• Result: National First Prize (Top 6) in the National University IoT Design Competition and completion of a provincial innovation project.

• Designed and implemented the A4R (Ask for Refine) interaction scheme for ambiguous smartphone-agent requests.
• Built 150+ real interaction samples covering 6 domains and 16 apps; designed Guide, Discern, and Validate collaboration.
• On China Unicom real-world tasks, the system reached 64.68% completion rate and outperformed baselines such as MobileAgent-V2, MobA, and GUI-Explorer on key metrics.

• Built a complete logistics delivery system on Xiaomi Quadruped Development Platform for complex-environment scenarios.
• Decomposed delivery tasks into departure, pickup, path navigation, storage identification, unloading, return, and secondary delivery stages.
• Focused on state transitions, action timing coordination, and reducing failure propagation to improve closed-loop completion rate within time constraints.
• Result: National Third Prize in National University Computer System Capability Competition (Xiaomi Cup).

• Built a multi-agent system on Holos and A2A v0.3.0 for uncertain-requirement scenarios.
• Implemented GuideAgent, DiscernAgent, and 8 functional agents, covering requirement clarification, decomposition, execution, and verification.
• Completed protocol adaptation, SSE streaming response, task-fingerprint deduplication, LLM/rule fallback, and observability debugging for travel-planning and email-collaboration demos.

• Built a unified physical layer mapping screen, CPU/GPU, communication, and sensor loads to total power consumption, constrained by coulomb-counting dynamics and thermal equilibrium.
• Implemented dual-branch solvers: 2RC equivalent circuit model + Extended Kalman Filter (ECM-EKF) for real-time SOC estimation and drift correction; PINN with identical dynamics in loss function for nonlinear fitting and factor attribution.
• Validated on 8 heterogeneous data sources with over 120,000 time steps, achieving PINN SOC RMSE of 1.24% and overall TTE prediction MAPE of 8.7%.
• Result: MCM 2026 Finalist Award.

• Wuhan University, B.Eng. in Software Engineering — 2023.09 – 2027.06 (expected)
• GPA 3.74 / 4.0
• Core courses: Computer Vision, Frontier Technologies in Artificial Intelligence, Computer Networks, Operating Systems, Software Engineering, Software Quality Assurance and Testing, Database Systems, Object-Oriented Programming
• Research interests: visual representation under appearance variations · structured scientific reasoning with small models · multi-agent systems and real-world task systems

• National University IoT Design Competition (Huawei Cup) — National First Prize (Top 6)
• China Youth Sci-Tech Innovation Challenge — National Second Prize
• National University Computer System Capability Competition (Xiaomi Cup) — National Third Prize
• MCM 2026 — Finalist
• MCM 2025 — Honorable Mention
• CUMCM Hubei Division — Second Prize
• Zheng Geru First-Class Scholarship — ranked 1st in the grade-level defense and sole recipient
• Wuhan University First-Class Scholarship — twice
• Merit Student — twice
• Outstanding Student Cadre, Outstanding Youth League Branch Leader, Future College Outstanding Trainee

• Vice Minister of Publicity, Student Party Branch Secretaries Joint Council, Wuhan University (2024.09 – 2025.09)
• Supported undergraduate Party-building publicity work and contributed to 100+ published articles, including theoretical pieces and process optimization work.
• Project Team Member, Second Classroom Center, University Youth League Committee (2023.10 – 2024.09)
• Participated in platform quality monitoring, issue investigation, reporting, and fix follow-through for multiple platform issues.

• Programming Languages & Tools: Python, C/C++, Java, Git, MATLAB
• Computer Vision & 3D Modeling: Person Re-ID, SMPL body modeling, 3D Gaussian Splatting, experimental design, ablation analysis, academic writing
• LLMs & Multi-Agent Systems: orchestration, routing, validation, observability, and application development with GPT, Claude, and Gemini APIs
• Embedded Systems: embedded driver development, modular hardware protocol design, RISC-V / ESP32 dual-architecture platform development

• Email: 2023302143001@whu.edu.cn
• GitHub: BohanSu
• Website: BohanSu.github.io
• English Resume: CV PDF
• Chinese Resume: 中文简历 PDF