img

GameAI


AI for learning how to play simple games using Neuro Evolution

img

Holograms


A device that is able to project large holograms onto the viewer's eyes

img

Homework Bot


A machine that does my homework

img

Squishy


A 3D printer that prints in fabric

img

PaperThin


Flexible and Foldable displays using projection mapping

img

Drishti


Visual-Aural Aid for Visually impaired

img

Sandals


A mission to near-space using a Raspberry Pi

img

Amateur Rocketry


An attempt to hit the skies using PVC Rockets

img

LiteDub


Live Synced Subtitles in the movie hall

img

Guitar Pickup


Homemade electric guitar!

img

ShockedUp


Interactive device for realtime electrical simulations

img

Yoda


AI friendly robot, that maps and navigates its environment

img

Groove


Artificially Intelligent DJ that manages and mixes

img

Jarvis


AI Assistant that does whatever you say

img

Blink!


App that forces user to blink using eye tracking

img

Captcha Breaker


An attempt to break login captcha

GameAI


Simple games, especially the ones that go viral have always prompted me to try and automate them.

This is a general program that learns to play it, through a process akin to biological evolution. The algorithm is just given the position of the obstacles and is told to maximize the game score. It then learns how to play on its own with the human out of the loop.

The general algorithm is based off an algorithm called NEAT (Neuroevolution of Augmenting Topologies), in which a generation of neural networks starts off randomly and the individual genomes breed and sometimes mutate. Only the fittest networks (genomes) are chosen to breed the next generation.

It innovatively uses speciation, where new mutations are grouped together to let the ideas fully develop before they are discarded.

Next steps? Using Convolutional Networks + NEAT to play the game just on raw pixels.