Visual DOOM AI competition results

Funny but I think that this competition is a blatant case for not using deep learning for everything. Apparently the software can have access to the depthbuffer. That makes the "defend the center" exercise extremely trivial and should make aiming and shooting much better than the demo we see.

I think that a scripted AI will be much better than a blind deep-learnt one. A level of deep-learning may add a good layer of intelligence, but I think that building it on some optimal bricks would be a good start.

I'd love to see this happening with self driving cars versus race drivers in an endurance race