IJCAI Preview | Can a computer play Angry Birds better than you? Let this competition tell you the answer.

Many people may not realize that Angry Birds can actually be a very brain-burning game.

One thing everyone knows is that its simulation of physical properties is very realistic - this is one of the reasons why it is so popular. The weight, shape, placement of objects such as birds, pigs, bricks, the relative speed of birds and bricks when the collision occurs, the angle of incidence, etc. will change the result of the entire collision. Many times, the building on the side of the pigs will have one or more structural weaknesses. If you successfully hit the structural weakness, the whole house will fall down like in the real world. At this time, the player watches the pigs explode into scores one by one, and the sense of accomplishment is really great.

However, as the number of levels gradually increased, the production team significantly weakened the visibility of structural weaknesses and increased the difficulty of hitting them. Some overall structural weaknesses will only appear after some small weaknesses are destroyed. At this time, relying on the pitiful few birds given by the production team to players is far from enough.

In fact, this level is quite easy, because most of the materials are wood. As the difficulty of the level increases, can you guarantee that you can complete it in one try every time? , Image source, cnetnews.com

However, the result of each launch is still predictable. If you understand this, you will understand why this game can be very brain-burning. If the production team requires players to complete the game in one try, for example, if they fail once, they can only try again after one day. The difficulty of this game will increase instantly. If you want to know what the players will become at that time, you can refer to a famous submarine simulation game: Hunter Submarine. It is said that many hardcore players will pause the game and count on a piece of draft paper for half an hour before launching in order to avoid torpedoes from being fired into the air.

But Rovio obviously doesn't want Angry Birds players to play so hard, after all, this is a casual game, so they hide all these complex parameters from players, and let us break through each castle only by intuition. In fact, we can really have fun in this way, but as the level difficulty increases, we rely more and more on luck to pass the level.

But - you know what I mean - for AI, this game is a completely different world. AI does not have "intuition" yet, so if you want AI to play Angry Birds, it must know every physics formula in the game that affects the results of collisions, and use these precise calculations to know what kind of bird will produce what kind of result when it is shot at what angle and force. It also needs to plan for the future and know where to attack first and where to attack next.

So the question is, can AI do this?

If you are also curious about the results, then I would like to say that we are lucky because we are not the only ones who are curious about this. In order to explore the possibilities of AI and promote the development of AI, a group of people established an annual "Angry Birds AI Competition" in 2012. The original 2012 competition was held under ACAI (Australian Conference on Artificial Intelligence). Later, most of the time, including this year, the competition was held under the famous artificial intelligence conference IJCAI (International Joint Conference on Artificial Intelligence). The competition includes three projects:

1. Contestants from all over the world showcase their AIs for playing Angry Birds. The highest requirement for this AI is that it can play better than the best human players when faced with a level it has never seen before.

2. The competition will hold a seminar to discuss Angry Birds AI.

3. Host an Angry Birds game between humans and robots to see who wins.

In fact, it is very difficult for AI to play Angry Birds. Although it is easy for it to calculate all the data, it may not understand what the data means. The AI ​​playing this game must know the position of all objects and the interaction between them. Their weight, solidity, how to hit them so that they fall down, how to launch the bird to hit them (sometimes it involves rebound). And the most important strategy: how to plan the direction of each bird to destroy all the pigs before the opportunity runs out. These are common sense for humans, but it is a big project to teach AI.

Angry Birds from the perspective of AI, Image source, AIBirds.org

However, the last mentioned strategy is obviously the most difficult and the most important part of the game. In order to allow the contestants to focus on the development of game strategies, the organizing committee of the competition simply provided a computer vision component - a module that allows AI to easily understand the location, score and other attributes of objects; a component that can help AI calculate how to launch the bird to achieve the goal after deciding how to hit the position; and a component that helps AI complete the game actions and realize screenshots.

This year's Angry Birds AI Challenge will be held all day on July 14 at IJCAI, and the Angry Birds Man vs. Machine Challenge will be held on Friday the 15th. How smart has AI become at playing games? We will continue to pay attention to this competition and other highlights of IJCAI, so stay tuned!