BYD AI Team Revealed for the First Time: HyWorldVLA Hybrid World Model Achieves SOTA on NAVSIM v1 With 90.59 PDMS, Marking BYD's Entry Into Autonomous Driving Foundation Models

BYD’s previously undisclosed AI team published its first paper on HyWorldVLA, a hybrid world model for autonomous driving.

The model combines pixel-level and latent-space prediction within a Vision-Language-Action framework and scored 90.59 PDMS on NAVSIM v1.

The paper found that both pixel and latent supervision are important for the best performance.

The result suggests BYD is building in-house autonomous driving foundation models, boosting its technical independence and EV AI ambitions.