Dino Ratcliffe impresses in 'Doom AI deathmatch'

Posted 23rd September 2016

Dino Ratcliffe created one of the standout bots in the Visual Doom AI competition held at this year's IEEE Computational Intelligence and Games conference in Santorini.

The Visual Doom AI competition challenged the entrants to create an AI that could 'effectively play Doom using only raw visual input'. This was a perfect challenge for Dino whose research focuses on teaching AI agents transferable skills for gameplaying. Dino's used asynchronous advantage actor-critic in the development of his AI and despite being able to complete only 40% of the bot training required due to a power supply failure, his bot Clyde achieved third place on track 1 of the competition. 

Dino was a single person team competing against teams from digital industry powerhouses such as Intel and Facebook, so his third place achievement is even more impressive. 

Dino's next plan is to focus on creating an AI that can learn to play 2D platformers with the ultimate goal of teaching an AI to learn to play Mario.

Dino is currently in his second year on the IGGI PhD programme and is based at the University of Essex.