See the Dictionary entry on read groups for more information about what they represent and why they're very important.
Note that the command line examples in this article have not yet been updated for GATK4. However, the principles they illustrate are valid.
As detailed in the FAQs about input requirements, GATK expects all read groups appearing in the read data to be specified in the file header, and will fail with an error if it does not find that information (whether there is no read group information in the file, or a subset of reads do not have read groups).
Typically you should add read group information when you perform the original alignments (with e.g. BWA, which has an option to do so). So what do you do if you forgot to do that, and you don't want to have to rerun BWA all over again?
You can use a Picard tool called AddOrReplaceReadGroups to add the missing information to your input file.
Here's an example:
# throws an error java -jar GenomeAnalysisTK.jar \ -T HaplotypeCaller \ -R reference.fasta \ -I reads_without_RG.bam \ -o output.vcf # fix the read groups java -jar picard.jar AddOrReplaceReadGroups \ I= reads_without_RG.bam \ O= reads_with_RG.bam \ SORT_ORDER=coordinate \ RGID=foo \ RGLB=bar \ RGPL=illumina \ RGSM=Sample1 \ CREATE_INDEX=True # runs without error java -jar GenomeAnalysisTK.jar \ -T HaplotypeCaller \ -R reference.fasta \ -I reads_with_RG.bam \ -o output.vcf
Note that if you don't know what information to put in the read groups, you should ask whoever performed the sequencing or provided the BAM to give you the metadata you need.