It's definitely finding a niche in robotic control. My lab just released a paper about running a policy trained in simulation in the real world on a bipedal robot.

Doesn't the Roomba use a form of RL?