Xiao Li

Massachusetts Institute of Technology, Shanghai Jiao Tong University

Papers

Total Citations

271

H-Index

About

Xiao Li is a robotics and artificial intelligence researcher whose work spans reinforcement learning, safe robot control, and language-model-driven navigation. Li's influential 2017 paper on reinforcement learning with temporal logic rewards, which has garnered over 160 citations, addressed a fundamental challenge in robotics: designing reward functions capable of encoding complex, real-world behavioral constraints beyond simple heuristics. By leveraging formal logic specifications, Li helped establish a more principled framework for training autonomous systems on sophisticated tasks. Building on this foundation, Li's 2023 work introduced BarrierNet, an innovative approach integrating differentiable control barrier functions into neural network architectures, earning nearly 100 citations and offering end-to-end trainable safety guarantees for critical robotic applications. Most recently, Li has turned attention toward accessible, efficient deployment of language models in robotics, with FASTNav exploring fine-tuned small language models for multi-point navigation tasks suited to edge computing environments. Collectively, Li's research reflects a consistent drive to make autonomous robots not only more capable and adaptable, but rigorously safe — contributions that have meaningfully shaped how the field approaches the intersection of machine learning, formal methods, and real-world robotic deployment.