Siheng Yean engineer’s notebook

Robotics · control · embedded intelligence

Siheng Ye

Between signals
& the physical world.

I’m a final-year Electrical and Electronic Engineering student at the University of Nottingham Ningbo China. I work where sensing, software, and machines meet.

Explore selected work Public CV
Portrait of Siheng Ye
葉嗣恒EEE · UNNC

Observe. Build. Understand.

On this page01 Selected work02 Research03 A little contextNingbo · 2026

01 / Selected work

Selected projects

From reading a lift panel to running a small language model on a Raspberry Pi. Research prototypes and earlier engineering projects, with the evidence behind them.

01

Robotics & perception

Robot–Elevator Interaction

Experimental robot platform from Paper 105 Figure 1, with the RoArm-M3 arm, wrist camera, RGB-D camera, Jetson computer, laptop and mobile cart
Paper 105 · Fig. 1 / experimental platformFull figure ↗

Teaching a robot
to read the lift.

Recognising a button is one problem. Choosing the right one, locating it in 3D, and approaching it safely is another. This project connects those steps in a staged, guarded workflow.

RGB-D · OpenCV · calibration · robot arm

My contribution: staged perception, RGB-D localisation, calibration, and guarded arm integration within a collaborative research project.

Open the project notes

The pipeline combines button detection, HOG+SVM classification, semantic target selection, RGB-D localisation, calibration, and guarded robot-arm interaction.

Evaluation: 0.9963 macro-F1 on 379 labelled crops; 85.1% offline request-level target selection accuracy. The two qualitative demonstrations are separate from evidence of generalised deployment.

IHCI 2026 · Paper 105 · accepted for oral presentation

Project overview Slides & poster Paper · PDF Code
02

Edge intelligence

Raspberry Pi 5 Engineering Assistant

Hardware & Ask/Quiz workflowsFull Paper 106 Figure 1: Raspberry Pi 5 prototype hardware in panel A, and the Ask and Quiz workflows in panel B
Paper 106 · Fig. 1 / hardware & workflowsFull figure ↗

A small machine.
A traceable answer.

A compact language model, course-material retrieval, and calculation tools. The aim: keep an answer close to the evidence that produced it.

Python · Qwen3-1.7B · RAG · Raspberry Pi 5

My contribution: quantised-model deployment, retrieval and calculation tools, the inspectable web workflow, and configuration comparisons.

Open the project notes

A 4-bit Qwen3-1.7B deployment pairs retrieval with deterministic calculation tools. The web prototype separates generated answers, retrieved sources, and tool outputs.

Evaluation: seven configurations compared across 96 EEE questions in an offline GPU reference evaluation. Pi 5 CPU observations are reported separately, using the same quantised model artefact; reference scores are not Pi device results.

IHCI 2026 · Paper 106 · accepted for oral presentation

Project overview Slides & poster Paper · PDF Code
03

Human–robot interaction

Bangbang V2

Raspberry Pi 5 vision workflowBangbang 2.0 vision workflow on Raspberry Pi 5, showing the shared MediaPipe front end and alternative action-unit and CNN/ONNX recognition paths
Bangbang 2.0 / vision workflow · project graphic prepared for LinkedIn The vision node used 16 GB RAM; the 8 GB label in the original graphic is incorrect.Full image ↗

Keeping the human
in the loop.

Facial and bilingual speech cues for an avatar robot. Developed the Raspberry Pi 5 vision path, while keeping the operator responsible for the final decision.

MediaPipe · ONNX · edge vision · teleoperation

My contribution: the facial-perception pipeline and Pi 5 vision integration within the collaborative teleoperation system.

Open the project notes

The dual-Pi teleoperation system combines a MediaPipe FaceMesh front end with complementary CNN and interpretable FACS/AU routes, bilingual speech, and a browser interface.

AI outputs are advisory. The human operator retains final decision responsibility.

Two distinct conference papers study the shared platform: facial perception and bilingual speech (HRFEST Paper 230), and human-in-the-loop edge/cloud LLM decision support (IHCI Paper 99).

HRFEST 2026 · Paper 230 · accepted for presentation
IHCI 2026 · Paper 99 · accepted for oral presentation

Project overview Slides & poster FURP 2026 poster (PDF) Code

Also on the workbench

04 / Power electronics

Closed-loop forward converter

A team-built 50 W two-switch forward converter. The work pairs PLECS modelling and custom magnetic components with Type III compensation and closed-loop hardware measurements.

PLECS · magnetics · Type III compensation

Design notes & original figures ↗
Complete two-switch forward converter power-stage schematic extracted from Group 07's EEEE2074 report, Figure 1
EEEE2074 report · Fig. 1 / power stageFull figure ↗

06 / Sensing & signal processing

Doppler radar speed detector

A team-built detector connecting an HB100 radar and analogue front end to STM32 FFT processing and a CPLD display. Oscilloscope-led debugging traced the timing and communication signals.

HB100 · STM32 · CMSIS-DSP · Verilog

Design notes & original figures ↗
Complete hardware prototype and wiring photograph from the Doppler Radar Group 02 presentation, Figure 2-4b, showing the STM32 board and breadboard connections
Group 02 slides · Fig. 2-4b / prototype wiringFull photograph ↗

07 / Earlier embedded systems

Robotic vehicle coursework

From optical-sensor line following and PID tuning to a Raspberry Pi vision prototype. An early team project connecting sensing, communication and motion control.

Arduino · Raspberry Pi · OpenCV · PID

Design notes & original figures ↗
Course vehicle hardware from the original viva slides
EEEE1039 viva slides / line-following hardware · Full image ↗

08 / Prototype in development

LEKSI

A handheld interactive prototype, currently in development.

AI-generated concept illustration of LEKSI, a handheld device with a simple pixel expression; not the final prototype
AI-generated concept illustration; not the final prototype. Based on the Seeed Wio Terminal.
Earlier study notes · hydropower & optical computing ↗

02 / Research

Research record

ORCID record

Conference papers and presentation materials. Proceedings publication and indexing are recorded separately.

IHCI
2026 / 105

Robot–Elevator Interaction: From Physical Panel Recognition to Activation

Panel recognition, target selection, and guarded physical interaction.

Abstract & authors Official programme Paper · PDF Code

Author’s version · September 2026. Publisher version and DOI not yet confirmed. Session: 29 October, 15:00–16:00 · SS4 · IEB 124 · Ningbo (UTC+8).

Accepted
Oral presentation
IHCI
2026 / 106

A Lightweight Test and Training Platform Utilizing a Raspberry Pi 5: AI-retrieval, Tools and a Compact LLM

Seven configurations compared on 96 EEE questions, with GPU-reference results and a separate Raspberry Pi 5 run.

Overview & authors Official programme Paper · PDF Code

September 2026 author’s version available. Publisher version and DOI not yet confirmed. Scheduled session: 30 October, 15:30–16:30 · SS2 · IEB 124 · Ningbo (UTC+8).

Accepted
Oral presentation
HRFEST
2026 / 230

Bangbang V2: On-Device Facial Emotion and Bilingual Speech Recognition for a Raspberry Pi-Based Tele-Operated Avatar Robot

Facial and speech cues with human-led teleoperation.

Siheng Ye · second author of nine.

Overview & authors Code

Accepted for presentation; proceedings publication and indexing are not yet confirmed.

Accepted
Presentation
IHCI
2026 / 99

Bangbang 2.0: Human-in-the-Loop Edge-Cloud LLM Decision Support for Inclusive Service Robots

Human-led selection and review of edge/cloud LLM suggestions for a telepresence operator.

Siheng Ye · fourth author of eight.

Paper details & authors Official programme Slides · PDF BibTeX

Accepted for oral presentation. Official programme: 30 October 2026, 15:30–16:30 · SS2 · IEB 124 · Ningbo (UTC+8). Proceedings publication and DOI remain unconfirmed.

Accepted
Oral presentation

The Robot–Elevator Interaction and Pi Engineering Assistant papers are shared for scholarly reading, subject to Springer Nature’s sharing terms.

03 / A little context

Learning to see
the whole system.

葉嗣恒 Siheng Ye

My starting point is electrical and electronic engineering. My next questions lie in robotics, control, and the intelligence we can put close to a machine.

I’m completing a BEng (Hons) at the University of Nottingham Ningbo China, with graduation expected in July 2027. I’m preparing for postgraduate study across embedded systems, control and automation, cyber-physical systems, and edge AI.

My current final-year project is a simulation-based study of digital-delay compensation and prediction robustness in PMSM predictive torque control. The work is exploratory; hardware validation remains ahead.

Across these projects, I care about what can be measured, what can be explained, and where a person should retain control.

Explorer of words & worlds. Outside engineering, I enjoy reading, learning languages and travelling.

I’m also looking for internships in embedded systems, robotics, control, electronics testing, and applied AI. During term, I can discuss 2–3 on-site visits per week near UNNC, with hours arranged around classes.

Selected final-year modules · 2026/27
EEEE3124
Internet of Things
EEEE3066
Control Systems Design
EEEE3076
Robotics, Dynamics & Control
EEEE3110
Sensing Systems & Signal Processing