Comment by vervez

There's also recent Bayesian Optimal Experimental Design methods that allow you to directly design experiments using gradient ascent. Not sure how it compares with BayesOpt on your problem, though.