Papers
2
Total Citations
8
H-Index
1
About
Shuyuan Zhang is pioneering the intersection of embodied AI and edge computing for robotic manipulation. Her research focuses on developing vision-action models that overcome real-world constraints like occlusions and limited fields of view, as well as creating energy-efficient hardware for on-device learning. In her highly cited work, "Observe Then Act," Zhang introduces an asynchronous active vision-action model that enables robots to dynamically adjust their viewpoint before acting, directly addressing a critical bottleneck in passive observation systems. This work has already garnered 7 citations since its 2025 publication, signaling strong early impact. Complementing this algorithmic innovation, Zhang also leads hardware-software co-design, demonstrated by her development of a 94Hz inference, 7.4mJ/epoch fine-tune edge SoC for diffusion-based robot manipulation. This system, featuring speculative parallel inference and disturbance enhancement, achieves the rare feat of both low-latency inference and high-fidelity on-device fine-tuning, a major step toward truly autonomous, untethered robots. By bridging advanced perception models with efficient edge silicon, Zhang is laying the groundwork for the next generation of adaptive, real-world robotic systems.
Research Focus
Key Achievements
Top Papers
- 1
- 2