It's an exciting time to be involved in artificial intelligence! New algorithms are surfacing after decades of research, modern hardware is making a huge impact, and companies/developers are taking it more seriously than ever. One of those emerging algorithms is Monte-Carlo Tree Search, or MCTS for short.
This algorithm was discovered in 2006, and its application to board games has been extremely succesful so far. Now MCTS is making its way into AAA video games and simulations too. Over the next few days, we'll be digging into MCTS in depth as we release The AI Sandbox™.
Launching in 2014-8-27 12:00:00 GMT+00:00...
Yes, it's finally launch week! Thanks to those of you who helped with testing and feedback in our Community Alpha, and everyone else for your patience. To celebrate, join us as we dig further into MCTS over the course of these (free) videos and tutorials.
1. MCTS for Strategic AI
2. Observing MCTS in Action
3. Applying MCTS Real-Time
You'll learn how MCTS was applied in TOTAL WAR: ROME II, what the algorithm looks like in action, and insights how to apply it to realtime simulations. (Click for details!)
Based on a presentation at the Game/AI Conference 2014 by Tim Gosling and Piotr Andruszkiewicz, TOTAL WAR: ROME 2 uses multiple instancest of Monte-Carlo Tree Search for its campaign AI. It's the first documented use of MCTS in the AAA games industry — and it's exciting!
There are a variety of details we'll be digging into in our video:
You can access the audio/video recording from Tim and Piotr's talk, as well as our analysis. We'll send it over by email, along with the rest of the videos from this week's pre-launch!
It's hard to fully grasp the Monte-Carlo Tree Search algorithm from text descriptions. Where the tree is expanded? Which states are randomly explored? How is the reward calculated and propagated? Seeing these things in action not only helps understand the algorithm, but also makes it easier to solve specific problems.
The figure above was generated within The AI Sandbox™, and is one of the (surprisingly) few visualizations of a real MCTS tree running on a real problem! In this case, the algorithm is responsible for coordinating multiple agents to explore a building.
As part of pre-launch week, we're releasing animated versions of these trees that you can browse around and watch as they grow. It's like seeing MCTS think about how to solve a problem — and it's fascinating!
While Monte-Carlo Tree Search has found great success in board games, applications to real-time simulations or games are still rare. There has been some academic research on the topic, but it's not yet found its way out of the incubator yet.
As part of this pre-launch week for The AI Sandbox™, we're going to be broadcasting our experiments applying MCTS to a real-time situation. In particular, it'll involve a Capture The Flag-style scenario, where multiple teams of bots have to secure the enemy flag and return it to their base.
It's a complex problem in rarely explored territory, so unexpected things may happen! What's clear, however, is that we'll all learn something new :-) Make sure to tune in to the stream to be part of it.
This is going to be an exciting week as we dig into Monte-Carlo Tree Search further. Join us on this amazing journey as we release (free) videos/tutorials on the topic, and stay tuned as we prepare to finally launch The AI Sandbox™!
NOTE: Due to a power surge and outage that occured during the weekend, we had to reschedule launch day to Wednesday, August 20th. We can work better with our computers powered!
Enter your primary email address, and we'll send you everything as it becomes available this week:
NOTE: Make sure you click the confirmation link once you receive our email...