I've tried, casually, quite a bit of this. With generally poor results. Often this ends up similar what I have seen as classic RL problems with seeking reward in sparse environments. Unless you give the agent an a-priori goal outside the world it tends to not do anything interesting. Theoretically there ought to be ways around this such as herarchical approaches, but those tend to scale in complexity faster than my limited resources can keep up.

It's very neat though.

Edit: I really should watch the video before commenting. Looks like it goes into lots of these problems itself, and they got much further than me.

I think you would have to go back to more classic genetic algorithm/pet training methods: Iterate through increasing "distance" to the goal. Initially make the goal simple/immediately available, then slowly "distance" them from the goal, with more obstacles/etc introduced between, as you go.

For a homing pigeon, this would be literal, driving further and further each time you release them.

For an agent that you train to walk, this would be flat surface to past some line, then introduce speed reward, then non-flat surfaces, then obstacles, then curves, etc [1].

[1] "AI Warehouse" channel is all about training AI for light gaming tasks: https://www.youtube.com/watch?v=L_4BPjLBF4E