Zhou, Shuai

After wrapping up my ongoing projects at CMU around March 2nd, I am looking for internships focused on robot foundation models or any direction that may lead toward general-purpose robotics.

I was a robot planning guys and once drawn to combining neuro-symbolic planning with foundation models for long-horizon tasks, but ran into difficulties fitting classical TAMP structures into a framework that could hierarchically learn symbolic concepts while handling uncertainty and partial observability. If you know of work addressing this, or are interested in exploring it together, please reach out. More recently, I have been shifting my focus to purely learning-based approaches, such as VLAs or world models, to see how well they handle long-horizon reasoning and manipulation, and whether classical planning and control ideas at the high level can offer new perspectives.

Currently I am at CMU working on projects that with the two paradiam i mentioned ahead. One is pure learning-based manipulation for humanoid manipulation, or later cross-embodiment manipulation. And hierachical planning with learning policies, a multi-robot pushing manipulation with flow-matching sampler. Earlier, I developed algorithms for warehouse robot coordination.

I will be back at CMU Robotics Institute this Fall as an MSR, working on the direction that I mentioned above.