Original article was published on Artificial Intelligence on Medium
Fun Artificial Intelligence Powered Video Games And Simulations You Should Try Now
And all of them are free!
“A Superpower”, coined by none other than Andrew Ng himself, AI nowadays can do just about anything. There is absolutely no limit to what it can create or where it can be applied. Here, we’ll be opening a small window for you in the application of deep learning in gaming and see how the world of reinforcement learning can influence this industry in the near future.
With that being said, let us now explore some interesting games and simulations which use AI in their key gameplay or integral to the mechanics.
- In this game (or rather a simulation) the player uses virtual joints, bones, and muscles to build creatures that are only limited by your imagination and see if the underlying algorithms can train your creature to perform tasks like walking, jumping, running, etc.
- It is based on a combination of a neural network and a genetic algorithm that can enable your creatures to “learn” and improve at their given tasks all on their own. As seen in the gif, one can start from scratch or use a pre-trained model like the frogger. You can view the entire learning process in real-time.
- The game does take time to reach a stage where it has high functionality but that is what is likable as it proves two things, the real world scenario and its base of machine learning.
- One can also create a very basic model of the MIT Cheetah. A real game-changer to such an application would be a 3D node environment. The possibilities will be endless.
You can try this game here- https://keiwan.itch.io/evolution
- Another fantastic game/simulation built upon AI. As the name suggests it is built upon the NEAT Algorithm which is a form of Neuroevolution. Simply put, the algorithm is based off again on the terms of a genetic algorithm focused on artificial neural networks.
- The basic objective is to make the cars learn to navigate the track, all by themselves.
- This simulation evolves quite fast! We did have our doubts, in the beginning, thinking it was a trick but the option of adding obstacles cleared all of them. Unlike Evolution, there is an option to tweak the viewing of the learning paradigm. Options such as the sensor or number of vehicles really give on good insight as to how the game and the algorithm is functioning.
- All code is available on the mentioned website.
You can play Neat Race here- https://oxygenium.itch.io/neat-race
- The approach to this game is very simple. An AI cockroach runs away from you, and you’ll have to catch it! This game was specifically chosen to show the prowess of reinforcement learning. As claimed by the developer, the cockroaches were not coded in any way except the Unity3D Machine Learning tool.
- This goes to show how far the reach of machine learning is. The cockroaches avoid the trap with a very quick response rate which actually makes it a good game to spend a couple of minutes. One can notice how the cockroach has learned to stay at the bottom left corner until the user approaches it.
- It also creates a good standpoint for someone to develop games in this fashion. As commented by an individual, a scoring system can be set to improve the game further.
You can try it out here- https://fangh.itch.io/cafard
Catch the box
The basic overview of the game is a direct catch and escape play. The AI agents try to catch and escape from each other. You also have an option to play directly with the AI.
Here is the link to a more detailed explanation, as provided by the developer-
- Once that is realized the game does quite well and is actually quite fun to explore as there are many options within the game. These options let one explore as to how a game such as this can be developed!
- The game is again based on the Unity 3D Machine Learning Tool.
You can try it out here- https://epsi.itch.io/ai-job-catch-the-box-
Talons Wrought of Steel
- Here, you fight with AI warriors in real-time in a fully-fledged 3d environment.
- This action-packed game is a prototype based on Q-learning. Q values or action values are stored in 2D arrays and are reiterated into different values based on the learning which takes place. There is a lot of scope for improvement, but this does show a good proof of concept of what can be built.
- The game as such is great, has all the basic action moves with the theme suiting it well. On one hand, every game has the concept of increasing hardness level by level. On the other hand, we made sure we weren’t attacked by continuously moving backward and attacking. By the third level, the AI made sure it didn’t stop moving front, thereby learning our pattern. By looking at the change of weights it looks like the machine learning paradigm is working, but it may vary person to person!
The technology present is all these games are quite complex. The foundation and experimentation with such games can give us insights that can ultimately help us build robust programs for tech like driverless cars and large scale automation. For example, Google’s Deepmind AI has formulated a method to see a pattern in learning to walk on its own! Such methods can then be tested with real-life robots. With Reinforcement Learning, a lot can be achieved and a lot can also be understood on how machines learn to do something.
Just imagine the day when someone creates a game with the combination of virtual reality and reinforcement learning!