Summary
Zhutian Yang is a research scientist with a decade of experience at the intersection of robot learning, task-and-motion planning, and vision-language models, currently based in Mountain View and affiliated with Google DeepMind after a PhD at MIT. His work focuses on enabling household mobile manipulators to plan and act robustly in semantically rich, partially observable environments by fusing pre-trained VLMs with geometric planning and hierarchical replanning. He has a track record of translating research into impactful systems—PIGINet, a Transformer-based feasibility predictor developed during NVIDIA internships, cut planning runtimes by ~80% and demonstrated zero-shot generalization via CLIP-based visual encodings. Prior projects span robot skill chaining, interactive task learning with natural language, and real-world demos on Kinova platforms, highlighting both simulation and hardware expertise. Known for combining theoretical rigor with practical systems engineering, he often frames continuous constraint satisfaction and feasibility prediction as learned components inside classical planners. Outside core research, he brings product-minded optimization experience from early startup work that scaled an audio watermarking feature to thousands of users.
10 years of coding experience
1 year of employment as a software developer
Doctor of Philosophy - PhD Robot learning and planning, Doctor of Philosophy - PhD Robot learning and planning at Massachusetts Institute of Technology
Bachelor’s Degree Information Engineering and Media, Bachelor’s Degree Information Engineering and Media at Nanyang Technological University Singapore
English, Chinese, French