An agar.io bot that tries to compete with humans

Beat humans in the agar.io game by having the biggest score.

Input: A sequence of the last N frames with:

• positions and types of:
  • at most f nearest instances of type food
  • at most p nearest instances of type player
  • at most m nearest instances of type mine

Initially N = 1, f = 1, p = 1, m = 0 will be used. If the results appear to be promising, these values will be changed.

Values outlined with green are distances to food, black - blobs.

N = 1, f = 2, p = 3, m = 0.

Output:

• direction
• action:
  • move
  • split - move and split
  • feed - move and feed

Initially only move action will be allowed.

During the evolution score will be assigned based on:

• time lived
• food consumed
• enemies consumed

Genetic programming is going to be used.

The programs are going to be represented as Lisp code.

The first implementation should be as simple as possible.

An example program:

(if (< (distance (mine 0))
       (distance (food 0)))
    (+ 180 (direction (mine 0)))
    (direction (food 0)))

To choose a subexpression to mutate, we can walk it recursively, construct a list of (references to) the individual expressions and choose one of them to mutate.

An example of such a list:

((if (< <...>) <...>)
 (< <...>)
 (distance (mine 0))
 (distance (food 0))
 (+ 180 <...>)
 180
 (direction (mine 0))
 (direction (food 0)))

Symbols:

• (distance (mine i))
• (direction (mine i))
• (size (mine i))
• 0, 5, ..., 355

Operations:

• (if test then else)
• (+ a b)
• (- a)
• (< a b)

• research existing solutions
  • Existing agar.io bots
  • Websocket-based agar.io clients
  • Open-source agar.io servers
• give a presentation
• play on external server
• play on internal server
• compare results from internal and external training
  • quality
  • speed
• compare the results
  • ai vs pro vs amateur
• give a presentation again
• add teamplay
• profit