|
|
|
|
|
Tim Lethbridge's Master's Thesis Research: Perceived Animate Motion by Simple Deterministic Rules of Inter-Object BehaviourAfter completing my undergraduate studies at the University of New Brunswick, I stayed at the same university to pursue my Master's. It took me a year of studying (and a summer of teaching) before I settled on my research topic. I originally thought I would work in pattern recognition with Dana Wasson (who later became Dean of the Faculty of Computer Science). However, I was attacted into the area of computer animation and studied under Colin Ware. The premise of my research was as follows: You can create very interesting looking behaviour on a computer screen, using a few simple squares and circles as visual elements, and a few rules that describe how these objects mutually interact with each other. In fact, the behaviours so elicited can evoke powerful feelings in the viewer that what they are observing is alive. The interaction rules are placed in a matrix (whose rows and columns represent the objects). Each object thus responds to every other object. We call the cells of the matrix 'partial responses'. And we give the term 'behaviour function' to the sum of the partial responses of an object to all other objects. Interesting special cases are where an object responds to itself, and where an object does nothing (i.e. represents an obstacle). At each animation time step, a new position is calculated for each object, based on its behaviour function. Over time, therefore, the objects move around the screen. What gives rise to the interesting behaviour is the feedback from one time step to the next. Feedback is a key ingredient of what are called 'nonlinear dymanic' or 'chaotic' systems. And in fact, the lifelike behaviours we observe are examples of 'strange attractors' in chaoes theory. As with chaotic system, objects on the screen in my MSc research would tend not to repeat themselves exactly (i.e. they would not follow the same path) but they would nevertheless appear to the viewer to be doing a simular kind of thing on an on-going basis (i.e. a bird never flies through the same space twice, but it has the same flying appearance every day). These systems are also sensitive to initial conditions: Staring a system on one state would result in a completely different sequence of events from a the same system started after moving just one object just one pixel. It was relatively easy to create 'physical' behaviours like pushing, pulling bouncing and orbiting. It was also not too difficult to create behaviours such as 'chasing', 'insect flying', 'catarpillar exploring' etc. An important class of behaviour was 'articulation' whereby several objects are joined together in order to work as a whole (e.g. segments of a catarpillar). This work is an early example of what is now known as 'artificial life', although to the best of my knowledge it has not been cited as such. The thesis is available as a pdf file: Perceived Animate Motion by Simple Deterministic Rules of Inter Object Behaviour You may also read the following journal paper: You could also see this book chapter: I was originally expecting to start PhD studies directly after completing my masters degree. To this end, I had obtained a postgraduate scholarship from the Canadian Natural Sciences and Engineering Research Council (NSERC). In fact, when the then-director of graduate studies in the UNB computer science department, Uday Gujar, saw my work, he offered to let me stay for an extra year and convert my program into a PhD program. However, just before that offer was made I received a job offer from Bell-Northern Research in Ottawa. Weighing all factors, I decided to defer my scholarship and work for a few years before continuing with my PhD studies. Whether this was a good decision I will never know. I would have had a PhD a good deal earlier, but at the same time I would have plunged into academia without the same degree of industrial experience. [Back to my main biography page]
|