There are two known issues with this tool (captured here: https://trello.com/c/plQYSCvS).
Perhaps one of these applies to your case?
This tool produces an error when the option "Genotype likelihoods calculation model to employ:" is set as "BOTH". When either of the two alternatives are selected individually instead, the job runs without issue. Occurs with both custom reference genomes or when using hg_g1k_v37.
Same exact error is reported for some datasets when this option is set to "INDEL" and the Rod file is the dbSNP vcf from the GATK bundle. It is not clear if this is a resource issue.
If so, then try to work-around for the first, and try an alternate reference file for the second.
Just a guess, but seemed worth sharing. Best, Jen, Galaxy team