AI dev blog (SFPI)

  • Karakanlud
    11th Feb 2016 Member 1 Permalink

    Hey guys!

     

       In this thread, I would like to make a development diary for my projects in the SFPI (Search For Powder Intelligence) challenge.

     

    About SFPI: This challenge revolves around life and evolution. We're trying to build bots that mimic real life creatures in the sense of interactions between the being and its environment.

     

    I'm making this diary for the developmetnt because most of the steps in it don't result in new saves, which makes following the progress quite hard. 

     

     

    Current state & save history:

     

    pre-SFPI: This was connected to a previous attempt to make brainlike structures in tpt: the "brain" unit in this save is technically a randomizer, which has a temp/mercury based pseudo-memory (basically, the inputs previously sparked have a higher chance to fire the "cell"). It is connected to a system of ARAY-TSNS neurons (the credits for this idea go to Schmolendevice). These form an allow-block system, which operates the frmebot through a spark separator. 

     

     (pls no h8 for the hectic design)

     

    The simplier form of a wall avoider was the "bishop" bot. It always moves diagonally, and when it detects a wall, it changes the direction of movement on that axis to the opposite.

     

     

    SFPI Project 1.: Multi-cell creatures.

     

    The next goal was making the bots self contained. My first attempt was making a multi-cell bot. The operations are placed in separate cells, which are connected by wifis. These proved to be quite unstable, and require a large number of wifi temps, so it's  only good for creating a single multi-cell robot.

     

    My most advanced version of these bots:

     

     

     

    Current project: Single Cell Robots

     

    This project focuses on stuffing all the controls into a single cell. I use a gravity-mediated system to function as the pseudobrain. 

     

    First type:

     

     

    It's quite compact, but uses wifi, so you have to adjust the temps to make copies move independently. They're also made to avoid each other.

     

    The same type, made to detect each other, thus avoiding collisions:

     

     

    Current stable prototype:

     

    1919945View Save 1919945

     

    These are stable, completely wifiless versions of the previous bots, so if you copy one of them, the copied robot will act on its own.

     

    The only backside is that its brain mechanism is based on attraction rather than repulsion. This means that if the bot is, for example, stuck between two opposing walls, the bot will collide into one of them.

     

    The red one has a better transmission to the pstns, so it can move faster. Also thanks to ScipioCarian for testing.

     

    Current dev.:

     

    I'm trying to make the repulsion-based brain working. I'm also planning to implement a (possibly optional) eating mechanism too. Most of the work is hopefully done, but I'm still calibrating it, because it's way more unstable now.

     

     

    Edited once by Karakanlud. Last: 11th Feb 2016
  • Karakanlud
    16th Feb 2016 Member 0 Permalink

    Finally found a promising design for the repelbot. Strangely it works since the update, so the previous breaks were probably caused by a bug.