About me
I am a Robot Learning Engineer focused on enabling generalizable dexterous manipulation across diverse humanoid robot embodiments (arms/hands), tasks and natural human instructions.
Previously, I studied the Elite Master of Science program in Neural Engineering at the School of Computation, Information and Technology of the Technical University of Munich (TUM) where I was mentored by Prof. Gordon Cheng. My thesis, "Neural Feature Fields for Language-Based Dexterous Robotic Manipulation," focused on language based dexterous grasping and was done at TUM and Agile Robots SE. This work was supervised by Qian Feng and Prof. Alois C. Knoll, with co-advisorship from Jianxiang Feng.
In general, my work combines visual, tactile, proprioceptive and semantic understanding of the real world with Deep Learning to allow robots to perform dexterous manipulation based on human instructions. I have developed deep foundation models for generative robotic grasping trained on massive-scale synthetic datasets from simulation, then integrated them with LLMs and VLMs for prompt based one-shot grasping. I have also combined imitation learning with dexterous grasp optimization from 3D vision-language features, to enable language and demonstration driven few-shot grasping, and applied deep reinforcement learning to train in-hand manipulation policies.
What i'm doing
-
Dexterous Manipulation
Generalizable dexterous grasping across humanoid robot embodiments, guided by natural language instructions.
-
Robot Learning & Foundation Models
Generative grasping models trained on massive synthetic datasets, integrated with LLMs/VLMs, plus imitation and reinforcement learning.
-
3D Computer Vision & Perception
Object detection, instance segmentation, 3D pose estimation and scene reconstruction for robotic tasks.
-
Simulation & Sim-to-Real Deployment
Large scale robotic simulators, sim-to-real transfer, teleoperation, robot agent control and AI model deployment.
