Back
精读 Training-Time Action Conditioning:用训练时模拟推理延迟和 action prefix 条件化,去掉 RTC 的推理时 inpainting 开销。
paper deep dive
real-time chunking
vla
flow matching