For an arbitrary genome, you can use BEDOPS directly:

$ binSize=50
$ sort-bed signal.bedGraph | awk -vOFS="\t" '{ print $1, $2, $3, ".", $4 }' - > signal.bed
$ bedops --merge signal.bed | bedops --chop ${binSize} - > bins.bed 
$ bedmap --echo --echo-map-score bins.bed signal.bed > answer.bed

Replace the binSize as desired.

For a specific genome where you are doing comparisons between files with different signal windows, you can use BEDOPS with UCSC Kent utilities:

$ binSize=50
$ sort-bed signal.bedGraph | awk -vOFS="\t" '{ print $1, $2, $3, ".", $4 }' - > signal.bed
$ fetchChromSizes hg38 | awk -vOFS="\t" '($1!~/_/){ print $1, "0", $2 }' | sort-bed - > hg38.bed
$ bedops --chop ${binSize} hg38.bed | bedmap --echo --max --prec 3 - signal.bed > answer.bed

We use --max --prec 3 in place of --echo-map-score so that the score reported is the maximum score over the bin, in the generic case where a bin might happen to straddle two signals. You might pick a bin size larger than one of your signal's windows, for instance, or a bin size that does not divide signal windows evenly.

Other signal/score operators are available in bedmap, like --mean, --sum, etc. See bedmap --help or the documentation for more detail.

You can then repeat this for multiple signal.bedGraph files, and get score values over the same bins, regardless of what windows are in any particular signal.bedGraph.

I would follow this approach (if not python).

1. Generate 50bp windows using bedtools makewindows function
2. Convert bedGraph to bigwig using bedGraphToBigwig
3. Use bed file from step-1 and bigwig from step-2 with multiBigwigSummary from deeptools or bigWigAverageOverBed