Jim Fan Introduces VFlash for Faster Physical AI Inference

Jim Fan introduced VFlash, a model that predicts fixed-length geometric outputs such as object locations and robot trajectories in parallel, and said its paper and source code are open. He reports up to 12× faster action decoding and a 5.1× overall robot inference speedup on edge GPUs, including Jetson Thor and DGX Spark. Fan also reports up to 61% lower 3D trajectory error than the vision-language model backbone. These are reported results for VFlash; the figures describe different measures of speed and accuracy.

AI 根据报道生成 · 57 分钟前更新

还没有足够的连续观测数据，暂不绘制趋势。