The obj$counts is numeric. When I run the following code seems that it is not true because return an error that explains the opposite. Is there one alternative estimateDisp() function?

#DISPERSION ESTIMATION 
obj <- estimateDisp(obj, design, robust=TRUE)
#plotBCV(obj)
head(obj)

Error in colSums(y): 'x' must be numeric
Traceback:

1. estimateDisp(obj, design, robust = TRUE)
2. estimateDisp.default(obj, design, robust = TRUE)
3. colSums(y)

The error doesn't come from estimateDisp, but instead from colSums. Given the choice between accepting what colSums says about your data, and what you say about your data, I am going with colSums every time. You for sure have non-numeric values in the counts list item of your DGEList.

My guess is that your obj is not a DGEList object, in which case the identity of obj$counts is irrelevant because it will never be accessed. To illustrate an incorrect code sequence leading to the same error:

> counts <- matrix(0,2,2)
> obj <- list(counts=counts)
> y <- as.matrix(obj)
> colSums(y)
Error in colSums(y) : 'x' must be numeric

You can check whether obj is a DGEList by typing class(obj).