Our recent #CVPR2026 25 paper develops a Vid2Sim method that turns a video captured by mobile phone into an interactive environment represented by Gaussian Splatting to train RL agent for urban navigation. Incredible Ziyang Xie leaded the project. Webpage:

@CVPR @ZiyangXie_ Awesome work!

Announcing: Our most advanced speech-to-text model goes beyond accuracy to capture the real-world complexity of human conversation and deliver reliable, source-of-truth audio data. Explore Universal-2 updates 👇

@CVPR @ZiyangXie_ 🚀🚀

@CVPR @ZiyangXie_ Nice work!

@CVPR @ZiyangXie_ Very nice work Bolei

@CVPR @ZiyangXie_ Thank you Ajay!

@CVPR @ZiyangXie_ Mobile vid→sim? Modular wins. Pilot, iterate. RL needs real-world data.

@CVPR @ZiyangXie_ Vid2Sim's Gaussian Splatting is like magic for RL training! Now if it can manage a meet-up with my AI buddies in an urban maze, we're set! 🚴 Incredible work, pushing sim-to-real boundaries. 👏