Google DeepMind AI learns how to do parkour

Researchers around the world are constantly working to improve artificial intelligence and allow the tech to perform a greater number of tasks. Now, interestingly enough, a team at Google’s DeepMind has gotten AI to learn how to do parkour. The findings come from the “Emergence of Locomotion Behaviours in Rich Environments” report, which aims to prove how reinforcement learning (RL) can be used to teach an AI agent how to navigate a complex environment.

Agents having fun in Parkour! Cool paper from colleagues at DeepMind https://t.co/X0PwKXrQ2M pic.twitter.com/yMT6XCNv45

— Oriol Vinyals (@OriolVinyalsML) July 10, 2017

In DeepMind’s tests, the AI was placed in a terrain filled with platforms, drops and ledges, among other obstacles. The faster the AI traversed the terrain, the greater the rewards it would receive. As a result, the stick-figured AI — which somewhat depressingly lacked arms or a head — was incentivized to learn how to run, jump and climb his way through the environment.

The full report can be viewed here.

Image credit: Flickr – Ben Nuttall

Via: Engadget