OK I've figured it out anyway ... I would like to write my solution here for recording and any possible incomes.

What you need is several .loom files that contain velocity info, and a related seurat object which have already been integrated.

First you will need to read those .loom files (e.g. read.loom.matrices) as a list, each element of this list contains 3 elements :spliced, unspliced, and ambiguous. Then you need to change colnames of each sub element along with colnames of your seurat object. The merging process of this list could be done as a simple lapply function with do.call(cbind), after that you will get a new list which contains spliced, unspliced, and ambiguous dgmatirx.

For scVelo, use the as.Seurat(x = ldat) to convert the ldat list into the Seurat format (e.g., seu.new), then change the seu.new[["RNA"]] slot to the [["spliced"]] slot and set as the default assay slot with DefaultAssay function. Load what you will need (e.g., [["umap"]] or $seurat_clusters slot) into the seu.new object.

Finally, convert it into H5AD format with SaveH5Seurat and Convert function, which could be read by scanpy in python.

best, W