Super-Human Performance in Gran Turismo Sport Using Deep Reinforcement Learning

  • Thread starter Granadier
  • 1 comments
  • 931 views
Status
Not open for further replies.
1,336
United States
New York


This was shared with me recently and I wasn't able to find an existing thread about it. Some really awesome content about how reinforcement based AI learning was able to optimize a lap in GT Sport and squeeze every second out of it. Now the real-world monetary requirements for doing reinforcement learning likely puts this out of scope for Polyphony Digital to actually apply it to their in-game AI, but the experiment itself is interesting nonetheless.

What stands out though regarding the optimization of this lap time is this quote from TG.
"The AI drives very aggressively, but I think this is only possible through its precise actions. I could technically also drive the same trajectory, but in 999 out of 1000 cases, me trying that trajectory results in a crash which destroys my whole lap and I would have to start a new lap from scratch."
- TG

The white paper is linked to in the video description if you'd like to read more nerdy details.

Also, could this mystery TG be the one and only @Tidgney :lol:
 
Status
Not open for further replies.
Back