This is the same as Roland's solution except using the stringr package. When working with strings I'd recommend it though as the interface is more intuitive.
library(stringr)
d <- str_replace(junk$percent.error, pattern="%", "")
junk$percent.error <- as.numeric(d)/100