Project Pages: ICRA 2026 Overcoming Imperfect Kinematics in Surgical Robotics Through Sim-to-Real Visuomotor Learning
KD

Kaizhong Deng

PhD Candidate · Robotics & AI for Healthcare

The Hamlyn Centre for Robotic Surgery, Imperial College London

Expected graduation: October 2026

Open to Work · Available from Oct 2026 · Research Scientist & Research Engineer · Robotics
Embodied AI Medical Robotics Imitation Learning Computer Vision Deep Learning Visual Servoing
k.deng21@imperial.ac.uk kaizhong-deng

About Me

TL;DR

I specialize in autonomous surgical robotics integrating computer vision, robot learning, and control, with extensive real-robot experience in medical applications.

I'm a final-year PhD candidate (expected October 2026) at Imperial College London's Hamlyn Centre for Robotic Surgery, developing autonomous robotic systems that push the boundaries of surgical precision and safety. My research bridges computer vision, imitation learning, deep reinforcement learning, and robotics control to enable robots to perform complex medical instrument manipulation tasks.

With a track record that includes first-author publications at ICRA, experience integrating cutting-edge Vision-Language-Action models at Kinisi Robotics, and hands-on work with platforms from da Vinci surgical robots to KUKA systems, I specialise in translating AI advances into practical robotic solutions. My work on sim-to-real learning and visual servoing has achieved human-level performance in medical sampling tasks.

I'm actively seeking opportunities for full-time robotics research engineer positions starting Autumn 2026, passionate about advancing robotics and AI systems that solve real-world challenges and bridge the gap between research and deployment.

If you are interested in introducing robotics and AI technology into Device Art or any form of art expression — please feel free to reach out!

Open to discussing industry opportunities and research collaborations in AI and robotics — feel free to connect!

Education

PhD in Robotics and AI for Healthcare

Imperial College London — The Hamlyn Centre for Robotic Surgery

Oct 2022 – Oct 2026

Research topic: Autonomous Robotic System for Medical Instrument Manipulation

Thesis: Robotic-Assisted Diffuse Reflectance Spectroscopy Sampling for Tissue Assessment: From Model-Driven Control to Learned Policies

Medical Robotics Computer Vision Imitation Learning Visual Servoing Deep Learning

London, UK

MRes in Medical Robotics & Image-guided Intervention

Imperial College London — The Hamlyn Centre for Robotic Surgery

Oct 2021 – Oct 2022

Thesis: Imitation Learning for Robotic Radio Sensor Manipulation

Overall: 78.1% Rank: Top 10% Thesis: 79.8% Distinction

London, UK

BEng in Mechatronics and Robotic Systems

University of Liverpool (2+2 with Xi'an Jiaotong-Liverpool University)

Oct 2017 – Jun 2021

Final year project: Deep learning segmentation tools for lung X-ray images (COVID-19)

GPA: 3.92 Rank: 1st / 60 Thesis: 85% First Class

Liverpool, UK & Suzhou, China

Research Experience

End-to-End Learning for Autonomous Surgical Instrument Manipulation

2025 – Present

An ongoing research project developing a complete autonomous robotic learning pipeline for surgical instrument manipulation, integrating multimodal sensing with advanced generative policy architectures. Demonstrated significant real-world performance improvements on a clinical surgical robot platform. Details withheld pending publication.

Imitation Learning Diffusion Policy World Models Multimodal Sensing da Vinci Surgical Robot Computer Vision Visual Servoing Spectroscopy

Sim-to-Real Learning from Demonstration for Imperfect Kinematics

2024 – 2025
  • Employed Reinforcement Learning to train a robust policy in simulation with domain randomisation.
  • Proposed teacher-student training to distil policy with privilege information and extend perception capabilities.
  • Improved spatial generalisation for uncovered workspace by 81%.
Published at ICRA 2026 — Co-1st Author

Reinforcement Learning for Robotic-Assisted Radio Tracer Detection

2023 – 2024
  • Developed an adaptive scanning method to enhance initial detection robustness.
  • Designed a deep reinforcement learning framework with reward shaping to optimise radiotracer detection precision and operational efficiency.
  • Validated on simulated and real surgical robot systems, demonstrating an 80% success rate.
Published at ICRA 2025 — 2nd Author

Autonomous Robotic-Assisted Spectroscopy Sampling System

2022 – 2024
  • Collaborated across interdisciplinary teams (clinicians, engineers, researchers) to translate clinical needs into technical solutions.
  • Designed a complete robotic system integrating a KUKA arm with multiple sensors.
  • Proposed a vision-based precise contact state estimation module for millimetre-level control.
  • Minimised variance to human-level intra-observer variance as gold-standard manual sampling.
Published at ICRA 2025 — 1st Author

Industry Experience

Robotics Research Engineer — Intern

Robot Learning Team, Kinisi Robotics — Bristol & London, UK

March – September 2025
  • Integrated and trained Vision-Language-Action (VLA) models including PI0, PI0FAST, and GR00T for autonomous warehouse manipulation tasks in simulation and on physical robots.
  • Developed simulation environments for benchmarking bimanual humanoid robotics policies.
  • Optimised ML policy training pipelines and evaluation infrastructure.
VLA Models PI0 / PI0FAST GR00T Isaac Sim Bimanual Robotics Policy Learning

Technical Skills

Core tools and methods I use across surgical robotics, robot learning, and computer vision.

Languages & Libraries
Python PyTorch NumPy OpenCV C++
Robot Learning
Imitation Learning (ACT, DAgger) Reinforcement Learning (PPO) Diffusion Policy World Models VLA Models (PI0 / GR00T) Transformers
Robotics & Control
Visual Servoing Visuomotor Control da Vinci Research Kit (dVRK) KUKA Teleoperation ROS
Simulation & Sim-to-Real
Isaac Sim Domain Randomisation Sim-to-Real Transfer Teacher–Student Distillation
Vision & Medical Optics
Computer Vision Keypoint Tracking Image Segmentation

Publications

1

Z. Yan, K. Deng, Z. J. Hu, G. P. Mylonas, D. S. Elson

"Overcoming Imperfect Kinematics in Surgical Robotics Through Sim-to-Real Visuomotor Learning"

ICRA 2026 IEEE International Conference on Robotics and Automation, 2026 — Accepted

Co-1st Author Surgical Robotics · Sim-to-Real · Visuomotor Learning Project Page
2

K. Deng, et al.

"Automatic Robotic-Assisted Diffuse Reflectance Spectroscopy Scanning System"

ICRA 2025 IEEE International Conference on Robotics and Automation, 2025

1st Author Medical Robotics · Spectroscopy
3

H. Zhang, K. Deng, et al.

"Hybrid Deep Reinforcement Learning for Radio Tracer Localisation in Robotic-Assisted Radioguided Surgery"

ICRA 2025 IEEE International Conference on Robotics and Automation, 2025

2nd Author Surgical Robotics · Reinforcement Learning
4

K. Deng, B. Huang, and D. S. Elson

"Deep imitation learning for automated drop-in gamma probe manipulation"

HSMR 2023 15th Hamlyn Symposium on Medical Robotics, Jun. 2023

1st Author Imitation Learning · Surgical Robotics
5

B. Dong, J. Chen, Z. Wang, K. Deng, et al.

"An intelligent robotic endoscope control system based on fusing natural language processing and vision models"

ICRA 2024 IEEE International Conference on Robotics and Automation, 2024, pp. 8180–8186

Endoscopy · NLP · Computer Vision
6

K. Deng, et al.

"TransBridge: A lightweight transformer for left ventricle segmentation in echocardiography"

MICCAI 2021 Workshop Simplifying Medical Ultrasound (ASMUS), MICCAI Workshop, 2021, pp. 63–72

1st Author Medical Imaging · Transformer · Segmentation

Mentoring

Supervised and provided hands-on research guidance for 10 MRes students at the Hamlyn Centre, Imperial College London, across thesis projects in medical robotics and robot learning.

2025–2026

Imitation Learning for Probe-Based Instrument Manipulation in Complex Surgical Environments

Imitation Learning Medical Robotics
2024–2025

Learning Visuomotor Skills Under Uncertain Hand-Eye Coordination on the Surgical Robot

Imitation Learning Computer Vision Robotic System
2024–2025

Imitation Learning for Shared Control for Surgical Task Automation

Imitation Learning Robotic System
2024–2025

Robotics Tumour Detection with Tactile Sensor System

Medical Robotics Robotic System
2024–2025

Robotic System for Optical Instrument Scanning

Robotic System Medical Robotics
2024–2025

Developing a Dual-Arm Soft Robot System for Learning Endoscopic Submucosal Dissection from Demonstration

Imitation Learning Robotic System Medical Robotics
2023–2024

Learning to Perform Hot-Spot Detection with Drop-In Gamma Probe: Enhance RL with Robust Initialisation from Primary Scanning

Reinforcement Learning Medical Robotics
2023–2024

Developing a Modular and Portable Surgical Robotic System for Automation Research

Robotic System Medical Robotics
2022–2023

Learning Optical Instrument Scanning Skills via Imitation Learning and Reinforcement Learning

Imitation Learning Reinforcement Learning
2022–2023

Learning Surgical Instrument Grasping Skills from Demonstrations for Robotic-Assisted Surgery

Imitation Learning Medical Robotics