Many more crappy projects I worked on long ago

Long time ago experiment

Long time ago experiment

Sketcher draws a given image as an artist would.

This project tackles the problem of controlling a physics-based biped that can be subject to external pushes and terrain disturbances. Various machine learning regression techniques are used in an imitation learning setting to learn a mapping from state to joint torques, using the SIMBICON controller as a reference. A reduced state representation is used that only contains infomation about the overall speed of the body, and pressures at the feet. The performance of several machine learning regression algorithms is compared on this task, and we discuss the limitations of this general approach.

This project tackles the problems of real-time object detection, localization, and tracking. More specifically, we locate a Rubik's Cube in a video stream and extract the sticker colors with minimal help from the user. The edge based approach we take is invarient to various camera settings and does not assume a particular color scheme for the cube.

A really fun project from a long time ago. The agents have inputs, outputs and a neural network that forwards between the two. The agents can be herbivores (and eat plant food) and carnivores (that eat other agents) and also reproduce. This may look like amateur crazy person production but this is actually quite a sophisticated simulation that I wish I had more time for, and many people have contributed to the project.

A long time ago I was really into evolution stuff. I'm perfectly aware that this generally has negative connotations :)

Fun project

Fun project - uses Monte Carlo estimation of inter-reflection.

Many real-world phenomena can be described by volume densities rather than boundary surfaces. Such phenomena are often referred to as participating media. Examples include smoke, clouds, fire and mixing fluids. The computer graphics community has recently begun to explore techniques for capturing these time-evolving systems by reconstructing the densities in a 3D volume. In this work we consider the problem of recovering the volume density of smoke. Coded patterns of light that result from temporal dithering of a DLP projector are directed into the volume of space occupied by smoke, and the scattered light is observed by a high speed camera. By considering the physics of the underlying planar Mie scattering process, we are able to reconstruct the density of the smoke in a three-dimensional volume.

Wanted to write one from scratch

Not that exciting

Works quite nicely with a very simple elipsoidal potential model.

I can't remember why I made this anymore. Seems rather useless.

I implemented this while I was taking my second Quantum Mechanics class. It visualizes the amplitude of a wave function corresponding to a 1D particle as it evolves in time according to a discretization of the time-dependent Schrodinger's equation. It supports arbitrary potential functions.

I wanted to see what happens...

Choon was a physics shoot-em-up game I developed with two friends for the game competition TOJAM 3 (2008). I wrote most of the physics and AI parts: particle engine, explosions, and mice that aim to kill.

Scale is a physics/logic/arcade game I developed with a friend for the competition TOJAM 4 (2009). We won the first place for best use of Theme. (The theme was "scale")

Progger is a physics/logic game I developed with a friend for the UofT game programming competition. It was quite the hit!