I assume that the R script in the Lambda has access to the AWS OS it's running on so theoretically it could use for example Clodyr's R library aws.s3 and fetch files from s3.
# get file as raw vector
get_object("mtcars.Rdata", bucket = "my_bucket")
Then you don't have to pass it thru Python and also you avoid rewriting your existing R code base. Or am I missing something? Is this not possible?