Yanming
Whole-body robotic dexterity / Embodied RSI
I am currently at HKU MMLab, mentored by Prof. Ping Luo. Previously, I worked with Shanghai AI Lab, Shanghai Jiao Tong University, and Qizhi Institute, mentored by Prof. Yao Mu.
Research vision
Towards human-level dexterity
I am interested in how a robot learns to solve a task that its current skills cannot handle. My focus is dexterous manipulation: using physical feedback to discover a first successful strategy, then turning that strategy into a skill that transfers to new objects and situations.
I study embodied recursive self-improvement through self-harness: an agent revises the code that coordinates its tools, memory, and skills, then tests those changes through interaction. I want to connect this process with learning from real and synthetic human data. Successful trajectories could train better action models; those models could, in turn, support exploration of harder tasks. My long-term goal is to understand what makes this cycle reliable and scalable toward human-level dexterity.
Current projectPhysicalRSI ->
Contact
“The sea advances imperceptibly and without a sound. Nothing seems to break; nothing moves. The water is so far away that one can barely hear it… Yet it eventually surrounds the resistant substance, which little by little becomes a peninsula, then an island, then an islet, and is finally submerged in turn, as though dissolved in the ocean stretching as far as the eye can see.”
Récoltes et Semailles · English rendering