Tsinghua Yao Class Instructor Mengdi Xu: Scaling Alone is Not Enough for General Robots
量子位 · wechat · 2026-08-29
Tsinghua IIIS Assistant Professor Mengdi Xu has become a head teacher for the 2026 Yao Class. With a PhD from CMU under Ding Zhao, her research focuses on robot learning and safety. Xu argues that building general robots cannot rely solely on the Scaling Law, as real-world tasks are infinite and cannot be fully covered by training sets. In her dissertation, she proposed that robots must possess adaptation capabilities, including rapid adjustment via few-shot demonstrations (Prompting Decision Transformer), discovering shared structures across tasks, and leveraging LLM knowledge for physical problems (RoboTool). Her vision for "next-generation intelligence" aims to enable robots to adapt quickly in unfamiliar environments beyond their training data.
More from Companies & People
- Japanese Official Tests Tesla, Wayve, and Waymo: Distinct Strategies in Autonomy — alexgkendall · 2026-08-29
- ML Engineering Guide: Build 1-2 End-to-End Capstone Projects — Al_Grigor · 2026-08-29
- Free ML Zoomcamp 2026: From Regression to K8s Deployment — Al_Grigor · 2026-08-29
- Indie strategy: Ship loads vs. Focus on one product — kylegawley · 2026-08-29
- Suno AI Leak Reveals MERT25 Stack and YouTube Music Scraping — bdsqlsz · 2026-08-29
- SN33 Test: Verified Data is Scarce, One-Shot Full-Stack App Gen — markjeffrey · 2026-08-29