In AI Arena, your NFT is powered by AI.
What does that mean, exactly?
It means the initial generations are powered by a type of Machine Learning (ML) model called neural networks¹.
Your NFT may look like a fun, simple cartoon character. But it’s not simple! Beneath the surface, there is a complex, intricate, and layered machine.
These layers are the Skin, Frame & Core, which together makeup your NFT character.
The Skin is the outer layer of the NFT, giving it a distinct visual appearance. There are 8 categories of randomized attributes which make up the appearance. Each attribute category has a rarity associated with it.
8️⃣ Categories of Physical Attributes
The Skin is generated via a DNA sequence created by Chainlink VFR.
This DNA sequence is used to extract the physical attributes. Each generation of fighters has a different set of attributes with a given DNA sequence.
4️⃣ Levels of Rarity
Each physical attribute has a rarity associated with it. There are a total of 4 rarities: Bronze, Silver, Gold, and Diamond.
We use VFR to ensure the minting process is fair and transparent, where no one (including the team) is able to know the rarity of the character ahead of time.
The Frame is the NFT’s robot body. The metal alloy that makes up the Frame determines the fighter’s distinct strengths and weaknesses - or Battle Attributes. In addition, each robot body is endowed with a special Elemental Power.
Frames are built using a metal alloy, composed of Titanium and Black Zirconium. Each Frame has a randomized composition of these metals. The table below illustrates the spectrum and tradeoffs.
Speed / Agility
Based on the metal composition for the Frame, each NFT falls into one of three different weight classes:
Striker ➡️ Lightweight. Relatively more Titanium.
Scrapper ➡️ Middleweight. Balanced Titanium and Zirconium.
Slugger ➡️ Heavyweight. Relatively more Zirconium.
It is important to note that battle attributes are endowments that do not change. Therefore, think about training as learning to fight given the potential that a fighter is "born with".
🏋️ Weight - The relative composition of the metal alloy determines a fighter’s weight in the game. A fighter’s weight is the primary determinant of its other relative strength and weaknesses (i.e. all other battle attributes are a function of weight). Additionally, it is used to calculate how far the fighter moves when being knocked back.
💪 Power - The higher the power, the more damage a fighter can inflict on its opponent.
🛡️ Defense - The higher the defense, the less a fighter will get damaged for a given attack. This also means that they will experience less knockback.
💨 Speed - The higher the speed, the more distance a fighter can cover over a period of time.
🎯 Range -The longer the range, the more distance the fighter’s projectile attacks can cover.
🏃♂️Stamina - Each fighter has a stamina meter that gets used up when they perform certain attacks.²
In addition to standard battle attributes, each NFT is endowed with special powers based on their element.
There are three featured elements, each with corresponding elemental moves. These moves have relative strengths and weaknesses depending on pairing.
There are currently two main types of NFTs.
🤖 AR-X Bots - Most NFTs are AR-X Bots (short for ArenaX Bots). They are the standard class of NFTs battling in the Arena.
📺 Dendroids - Dendroids are a more exclusive class of NFTs. They have the ability to incorporate other metaverse assets (e.g. NFTs from other projects) into the Arena! These are very rare and more difficult to obtain. You’ll have to follow the story to find out how to get access to one of these!
We showcase the first generation Dendroids below. New designs will be released slowly over time.
It is important to note that Dendroids do not have any performance or attribute advantages over AR-X Bots. The difference is purely cosmetic.
The Core is the distinct energy source that powers the NFT. This is where the AI is contained.
The current generation of Cores are a type of Machine Learning models called neural networks. These neural networks have two components, the architecture (think: nature), and the parameters (think: nurture).
➡️ Architecture. The architecture of the model is stored on IPFS, and it determines the method for which states are transformed to actions. We use a feedforward neural network for the model architecture. It is helpful to think of the architecture as the framing for the machine learning model.
➡️ Parameters. The parameters of the model are also stored on IPFS. The parameters can be thought of as the “intelligence” that the model acquires through training. Each time a fighter is trained and the user chooses to save the progress, the parameters are updated on IPFS, and the new hash is updated on Ethereum.
Multiple models can share the same architecture, but all models will have unique parameters.
- In the future we will introduce the capabilities for researchers to supply different types of AI models to the game.
- Details on how stamina will be incorporated is still being finalized.