aagrawal05 / aiballpitswimmer Goto Github PK

Play around with attributes like ball size, player size, river (current flow of balls towards player), to see changes in behaviour from the agent.

Make it able to run faster in order to train faster and allow for new options to be implemented

Basic UI self-explanatory. See evolution simulator on youtube for reference on GUI between generations and development. For GUI for NEAT see the neataptic.js NEAT playground here for more information. Implement toggles and options, and restart buttons, using CSS

Bug with transparency.

Add dynamic scaling to ballCols and ballRows derived from chunkSize to ensure no gaps are created.

Connect up all the brain input layers and output layers w/ neataptic.js.

make best variable set to the correct best player - fix line 201, and fix transparency bug

All the agents interact with the shared balls, causing each individual's simulations to be to distorted, grid?

Find a working library for NEAT or implement it yourself. Good library is neataptic js --> https://wagenaartje.github.io/neataptic/

same problem of shared resources with the load new render bounds being shared by all agents - when to do so? highest max.x? avg?
Solutions:

• Grid base solution
  • if done this way, no more offset counter and moving platform, fixed camera of the whole grid, and the every object moves
    based on how the agent moves - relatively in each of the grid as opposed to the player moving through and everything
    shifting over (player stays in center)
• Brainstorm others

They should be derived form x position and do not collide (UI)

Add fluid dynamics instead of ball pit in a different branch, extra after finished with project

Try and fix choppy movements when balls teleport to the other sides on the edges of screen

Make agents not able to collide with each other but able to collide with everything else

See the ragdoll example in matter.js/examples/ragdoll.js

Take average of coords of all agent's "limbs" and see x difference from where he began, to calculate fitness provided. Function should tie in with the core update loop, in order to see how far the render frame is to the right.

Follow furthest player
Optimisation: check prevbest to not have to do n time

Remove walls and make a short platform for agent to "dive off of". Make ball pit "flow" (loop around to the front, see examples, implement with camera to center agent while balls flow from right and teleport to the left).

Make physics loop more efficient for a higher FPS, algorithmic improvements to increase overall efficiency.

Add the output functions to the agent definition, to allow the agent to access these functions as apart of the output matrix, when the AI is up and running.