PhD Student · University College London
I am a PhD student at University College London (UCL), supervised by Zhibin Li. I received my MEng in Computing (AI & Machine Learning) from Imperial College London with First-class Honours.
My research interests lie in combining multimodal foundation models with low-latency visuomotor policies to build robust robot systems that operate reliably in the real world. I am also interested in world models and foundation models grounded in physical interaction, particularly how prediction and action can be learned together.
How can robots connect broad multimodal knowledge with responsive control, and learn predictive models through physical interaction?
Combining multimodal foundation models with low-latency visuomotor policies, with an emphasis on responsive closed-loop behaviour and reliable deployment on real robots.
Exploring how world models and foundation models can learn from physical interaction, and how predictive representations can support robot learning, planning, and control.
Ongoing · UCL PhD
A teleoperation and data-collection system for a 12-DoF tactile dexterous hand (XHAND1), combining hand tracking, VR wrist pose, egocentric RGB, and fingertip tactile sensing. The goal is an open visuo-tactile dataset for learning contact-rich dexterous manipulation.
Ongoing · UCL PhD
Recurrent-depth VLAs improve as their action head iterates more, but every iteration adds latency to a real-time control loop. We rectify the recurrent core by distillation so that one or two iterations reach what the full loop converges to, evaluated closed-loop on CALVIN, LIBERO, and SimplerEnv.
UCL Research Assistant · April – October 2026 · Under review at ICRA 2027
RAI closes the loop between imitation learning and deployment: each round executes the current policy faster, records its rollouts with targeted human corrections, and trains the next policy on the accumulated experience. The multimodal flow-matching policy combines vision, proprioception, and estimated contact forces.
During my research assistantship at UCL, my work on the project spanned real-robot teleoperation, data collection and conversion, and real-time chunked policy deployment, connecting 30 Hz policy actions to 200 Hz arm control.
At Shanghai Jiao Tong University, I built the teleoperation, data collection, and evaluation stack for a LinkerHand O6 + xArm6 platform. This work grounds the real-robot side of my current interests, especially contact-rich data collection, dexterous action spaces, and deployment-time evaluation.
My MEng thesis built a closed-loop system where an LLM planner decomposes natural-language commands into a sub-task DAG and a behaviour-tree controller orchestrates perception, policy execution, and automatic recovery. The project achieved 100% success on pick-and-place and a 2.4× improvement over a code-as-policy baseline.
I built and maintain an open-source leaderboard comparing Vision-Language-Action models across eight benchmarks, including LIBERO, CALVIN, Meta-World, RoboChallenge, and RoboTwin 2.0, alongside a dexterous-manipulation track. This project helps me systematically compare architectures, training data sources, and post-training strategies across the current VLA landscape.
Supervised by Zhibin Li. Research interests in multimodal foundation models, low-latency visuomotor policies, world models, and learning through physical interaction.
Research on foundation models and real-world robot learning. Completed the RAI project during this period, developing robot control, data collection, policy training, and evaluation infrastructure.
Research with Professor Bo Zhao on Vision-Language-Action models for dexterous manipulation and VLA benchmarking (Evo-SOTA). Built teleoperation, data collection, training, and evaluation pipelines for a LinkerHand O6 + xArm6 platform.
Developed agentic capabilities for Westlaw Deep Research Agent and owned guardrails for query-intent classification and policy routing. Benchmarked and fine-tuned small language models for guardrail classification.
First-class Honours. Thesis on multimodal imitation learning for tabletop manipulation. Coursework in deep learning, computer vision, reinforcement learning, NLP, and robotics.
Built LLM-based pipelines for automated knowledge-graph construction on large-scale GPU clusters. Acknowledged in a Findings of ACL 2025 paper.
Outside research, I spend most of my time climbing. I summited Kilimanjaro and played on Imperial College's Men's 1st Basketball team.
