My layman understanding is 1. big, practical size robotic arms are expensive and 2. also very slow, making real training expensive and time consuming. And 3, simulations aren't high-fidelity enough to be fully sufficient for training, but they definitely can help.
Tbf, this is Google we're talking about. They routinely pay a single developer more per year than a few of these would cost to produce. They could make hundreds without a second thought.