With VLAs and E2E everywhere, the modular self-driving stack feels a bit old-school, but still very much alive. Generative models make motion prediction surprisingly powerful. In this blog with @felix_yuxiang , we share the bitter lessons learned. 🧐 qiyan98.github.io/blog/2025/coge…
We’re excited to present TrajFlow at #IROS tomorrow — a conditional generative paradigm via flow matching for motion prediction, successfully demonstrated in self-driving applications.
🔗 Project page: traj-flow.github.io
💻 Code: github.com/DSL-Lab/TrajFl…
