Usually you would calculate the cumulative density function and then use this to convert uniform random numbers to your chosen distribution.
However, I don't think you can analytically integrate this pdf.
The other method is to draw random points from the x-y box surrounding the pdf, and reject the ones that are above the pdf. However this is tricky since your limits are +/- infinity. But you could do it with a smaller x range if you knew the x value at which the pdf essentially goes to 0. Like this.
# plot the pdf
xlim <- 2 # by trial and error -2, 2 contains most of the curve
pdfx <- seq(-xlim, xlim, 0.001)
pdfy <- 2^0.25*gamma(0.75)/pi*exp(-pdfx^4/2)
plot(pdfx, pdfy)
# generate enough points
n <- 1000
x <- (runif(n*2)*2-1)*xlim # twice as many is enough
y <- runif(n*2)*max(pdfy)
myy <- 2^0.25*gamma(0.75)/pi*exp(-x^4/2) # calc pdf at x
i <- which(y < myy)[1:n] # find first n pairs which are valid
points(x[i], y[i]) # plot them

Created on 2020-11-09 by the reprex package (v0.3.0)