Xiaomi releases an integrated world model framework for reconstruction and generation, breaking mainstream benchmark performance records According to Beating monitoring, Xiaomi Auto has officially released a new framework for the Xiaomi EV World Model assisted driving world model, which achieves deep coupling between 3D reconstruction and video generation modules internally for the first time. In autonomous driving simulation, traditional techniques often separate reconstruction from generation. The reconstruction module can restore the scene but cannot predict changes, while the generation module can predict the future but is prone to distortion drift over long periods of time. The team proposed the JointWM architecture, which uses three-dimensional geometric structures as physical skeletons to anchor scenes, and then completes visual details and predicts unobserved areas through generation modules, breaking multiple best performance records in mainstream benchmarks such as Waymo and nuScenes. In terms of specific mechanism, the reconstruction module WorldRec abandons the traditional pixel by pixel paradigm and uses sparse 3D query points for scene representation, incrementally fusing into a cross view 4D Gaussian spatial skeleton to achieve fast reconstruction of 10 second videos. Based on the geometric priors provided by the reconstruction module, the WorldGen generation module is limited by the physical boundaries of the skeleton and is only responsible for generating reasonable light, shadow, and texture. For content beyond the boundaries of future frames and blind spots in the field of view, the generation module uses a two-stage temporal training and distribution matching distillation mechanism for physical prediction. The entire architecture achieves a single view generation speed of 0.19 seconds and a three view generation speed of 0.46 seconds on H20 GPU, and supports video generation for up to 1 minute. This scheme achieved a PSNR of 28.48 in Waymo reconstruction accuracy testing and maintained a leading position in nuScenes zero sample generalization. In terms of generation efficiency, the proposed approach is 5.6 times faster than the autoregressive baseline Epona, and its spatiotemporal coherence ranks among the top algorithms in its class. At present, the research results have been implemented in three major scenarios of Xiaomi cars, including delivering over 100000 high-quality synthesized data for perception model training, building a highly realistic closed-loop simulation environment to reproduce long tail road conditions, and launching an assisted driving school to guide users with generative videos. [Original link]