EducationSep 2021 – Jun 2025
BEng Robotics – First-Class HonoursBSc Computer Science
Ocean University of China & Heriot-Watt UniversityJoint programme · Dual degree
Robotics · Research & Engineering
Zeyuan Gao
I build perception and robot-learning systems across the full path from data collection and geometric reasoning to real-robot deployment.
Currently working on dual-arm manipulation and deployment-oriented evaluation, following the MRes Medical Robotics and Image-Guided Intervention programme at Imperial College London.
01 — Recent Work
Dual-arm Robot Learning / π0.5 Deployment
Integrated teleoperation, multi-view data capture, π0.5 inference and physical execution on two AgileX PiPER arms.
- Platform
- 2 × AgileX PiPER
- Data
- PIKA / UMI · LeRobot-compatible episodes
- Tasks
- Physical pick-and-place and towel manipulation
Deployment completed; controlled training comparisons remain in progress.
Explore this system ↗02
Research Trajectory
Education, roles and the systems developed along the way.
Education & Roles
EducationSep 2025 – Sep 2026
MRes Medical Robotics and Image-Guided Intervention
Imperial College London
RoleAug 2026 – Present
Embodied AI / Robot Learning InternLatencore AIDrag to explore · horizontal scroll · focus this region and use ← / →
03 — Explore Projects
Four systems. Different research questions.
Open a card for the problem, contribution and results.
Dual-arm Robot Learning / π0.5 Deployment
Aug 2026 – PresentIntegrated teleoperation, multi-view data capture, π0.5 inference and physical execution on two AgileX PiPER arms.
Problem, contribution & results
- Problem
- A successful checkpoint is only one component of a real dual-arm system. Demonstrations, calibration, camera streams, inference timing, trajectory handoff and hardware behaviour must form a reliable loop.
- My contribution
- Data-workflow integration, π0.5 deployment and physical-system diagnosis.
- Results
- Physical pick-and-place and towel demos
Zero-Shot 6-DoF Tool Pose Estimation
Large Needle Driver · RGB–Stereo–CAD Pipeline
Mar 2026 – Sep 2026Symmetry-aware geometric reranking raised top-1 ADD@5 mm from 44.11% to 57.75% across 968 valid pose outputs.
Problem, contribution & results
- Problem
- A foundation pose estimator can generate a valid candidate while ranking a geometrically inconsistent pose first. The object is symmetric, thin, partially occluded and evaluated at a strict 5 mm ADD threshold.
- My contribution
- Individual MRes research on geometric pose reranking and evaluation.
- Results
- Segmentation, depth and reranking figures
Clinical Multi-agent Text-to-SQL
Oct 2025 – Dec 2025Answered 17 of 20 questions correctly in the team’s project evaluation.
Problem, contribution & results
- Problem
- Translate clinical questions into schema-grounded SQL over MIMIC-IV, with validation, repair and safe abstention.
- My contribution
- Planner-agent design, LangGraph integration and failure-case evaluation.
- Results
- 17 / 20 team evaluation questions
Minecraft Voice Assistant Grounded in Live Game State
Jan 2025 – May 2025Connected speech, structured live game state and an LLM/API response to parallel voice and interface output.
Problem, contribution & results
- Problem
- Ground a voice assistant’s responses in structured, live Minecraft game state.
- My contribution
- Live-state integration, structured validation and responsiveness testing.
- Results
- Captioned in-game interaction demo
04 — Capabilities Across Projects
What I do across projects
All four projects remain accessible. View project cards ↑
05 — About
Research through systems that have to work.
I am interested in where model behaviour meets engineering constraints: how observations are collected, how actions reach hardware, and how failures guide the next experiment.