Hello, I'm Boxuan Ye

You can also call me Stephen if you wish

I work across robotics, computer vision and embodied AI—bridging learning-based perception with verifiable, safe planning. My current research builds certified safety layers for UAV navigation in dynamic crowds; past work spans mixed-reality teleoperation, open-vocabulary 6D pose estimation, and vision–language–action policies for humanoid manipulation.

Working Experiences

Robotics · AI · Design Engineering

My work sits at the intersection of robot navigation, manipulation, and perception. I build systems that understand objects, predict human motion, and come with guarantees about how they behave. Recent work spans a certified safety layer for quadrotors flying near pedestrians (Imperial, with Dr David Boyle), humanoid teleoperation and vision–language–action deployment at Midea Group, and embodied-intelligence research at KUKA Robotics.

  • 🛸 Certified safe go-around navigation for UAVs in dynamic crowds
  • 🤖 Mixed-reality teleoperation & π0.5 VLA policies for humanoid robots
  • 👁️ Open-vocabulary 6D pose estimation pipelines

Interfaces

Soft technology & calm tools

I’m interested in how technical systems can feel gentle—how robotics, sensing, and datasets can be shaped as quiet, legible tools rather than noisy complexity. My work often focuses on interaction behaviours: micro-feedback, spatial clarity, and the visual organisation of diagrams and control surfaces.

I prototype with a mixture of embedded controllers, RGB-D cameras, motion sensors, and computational pipelines, supported by clean editorial UI layers for debugging, teleoperation, and documentation.

Python
C++
PyTorch
OpenCV
Open3D
ROS / ROS2
MuJoCo
Isaac Gym
Blender
Fusion 360

Projects

Selected works & experiments

Mixed-Reality Humanoid Robot Teleoperation
Robotics · 2025

Mixed-Reality Humanoid Robot Teleoperation

Apple Vision Pro-driven control system translating human gesture into humanoid behavior. Real-time inverse kinematics, depth-based collision avoidance, and haptic feedback loops for intuitive robot manipulation.

View Project
One-Shot 6D Pose Estimation
Computer Vision · 2025

One-Shot 6D Pose Estimation

Open-vocabulary object understanding from a single demonstration. Synthetic dataset generation in Blender, real-time tracking refinement via Kalman filters, enabling robots to manipulate novel objects without retraining.

View Project
Nature Kit — Children's Devices
Industrial Design · 2025

Nature Kit — Children's Devices

Teachable behavior systems for children aged 0–6. Embedding sensor feedback, soft material interfaces, and habit formation through calm interaction patterns that grow with the child.

View Project

Publications

Peer-reviewed & in preparation

In preparation · IEEE RA-L / ICRA 2027

  • Certified Safe Go-Around for UAV Navigation in Dynamic Crowds
  • Ye, B., & Boyle, D. — first-author manuscript

IEEE Robotics and Automation Letters · 2026

  • Bridging the Depth Gap: Adaptive Scene-Instance Alignment for Training-Free Depth Refinement in Robotic Manipulation Scenes
  • Hou, P., Zhang, Y., Ye, B., & Wu, Y. · doi:10.1109/LRA.2026.3673899

Engineering Applications of AI · 2025

  • A Lightweight Network for Category-Level Open-Vocabulary Object Pose Estimation with Enhanced Cross Implicit Space Transformation
  • Hou, P., Zhang, Y., Zhou, W., Ye, B., & Wu, Y. · Vol. 155, 110890 · doi:10.1016/j.engappai.2025.110890

Education Background

Robotics, interaction, soft infrastructures

My ongoing work moves between engineering and narrative systems, exploring how tools can be designed with clarity, emotional distance, and precision—similar to the logic of editorial design.

2023 - 2027

MEng Design Engineering · First Class (Expected)

Imperial College London · Dyson School of Design Engineering

2020 - 2023

High School

Focus on Mathematics, Physics & Computing

Research Interests

A living archive

I maintain a small archive of experiments, diagrams, code notes, dataset runs, and speculative sketches—updated whenever a result feels worth capturing.

  • 🛸 Certified motion planning for aerial robots in dynamic human environments
  • 🧠 Vision–language–action policies & open-vocabulary object reasoning
  • 👁️ Open-vocabulary perception & 6D pose estimation
  • 🛰️ Multi-sensor perception: RGB-D, stereo, and depth-aware safety layers

Beyond Work

Music & inspirations

Band & production

  • Bassist in a Chinese rock band and a JPOP band
  • Write songs and basslines with performance 10+

Practice playlists

  • Vulfpeck — Beastly
  • Stevie Wonder — Sir Duke

Recent listens

  • Radiohead · in Rainbows
  • Deca Joins · 浴室