Muyang Yan

I’m a senior in CS at Purdue researching Neuro-symbolic AI for robotics. I’m advised by Prof. Yexiang Xue and Prof. Zak Kingston at Purdue CS, and Prof. Katia Sycara at CMU’s Robotics Institute.

I want to build systems that can continually acquire and reason about both explicit and implicit knowledge to act in the real world alongside humans.

To this end, I study the use of neuro-symbolic abstractions to enable efficient planning, robust generalization, data-efficient learning, and verifiable behavior in robotic planning. In particular, I aim to develop abstractions that are aware of uncertainty in the real world, including that which arises from human decisions.

Outside of research, I love rock climbing, table-tennis, and music production.

Email  /  GitHub  /  Scholar  /  LinkedIn

Research

Online Data-Driven Goal Inference with Learned Action Schemas

Inferring the goals of agents when the underlying domain model is unknown by inferring a posterior over model hypotheses from data.

Using VLM Reasoning to Constrain Task and Motion Planning

Leverage the common-sense spatial reasoning capabilities of VLMs to prevent downward-refinement failures and accelerate planning.

Neuro-Symbolic Action Anticipation from a Single Image with Learned Probabilistic Rules

Learning probabilistic action preconditions from data through ILP to anticipate future human actions from single images.

Design and source code from Jon Barron's website