There have been labs at my uni in non-deep RL or evolutionary robotics researching curiosity rewards a few years back, but I've not seen something like this unt

There have been labs at my uni in non-deep RL or evolutionary robotics researching curiosity rewards a few years back, but I've not seen something like this until this year. Is it just scale and more powerful NNs that is making curiosity work well now? The Montezuma and this Mario video look like a serious breakthrough to me (non-expert).