Collaborative Evolution of 3D Models

Editor's Notes

• #3: We will first look at the motivation behind this work, which is creativity in design, and supporting collaboration in the conceptual design phase. Our solution, is the use of a computational model of creative design based on collaborative interactive genetic algorithms. To test our model, we conducted user studies where participants were given a task, and they used our model with and without collaboration to assess how the use of our collaborative model affected creativity. The task was 3D modeling, manipulating 3D models for a video game. I’ll be discussing the results and our most important findings. Let’s begin by looking at the motivation behind this work.
• #4: Design is an iterative, exploration process. Within this process, creativity has the potential to occur when a designer purposely shifts the focus of the search space. That is, goal oriented shifts while brainstorming, during exploration, are crucial to creativity. Several factors contribute to this. For example, the understanding of the problem changing over time or by considering additional variables which may result in interesting solutions. What we propose is a computational model of creative design that supports these shifts in the focus of the search space, in order to enhance creative design.
• #5: Our computational model of creative design is based on collaborative interactive genetic algorithms. In our model, we support the goal-oriented shifts by adding variables to the search space. Variables are added to the search space by the sharing of solutions amongst users.
• #6: Our case study is 3D modeling. Here you see the three models, a blue female model – like in a space suit, a green ninja, and a white robot. These are the models we used in our experiments. We use vertex programs to create transformations of 3D models. A vertex program applies a mathematical equation on a per vertex basis. So each equation, whatever it may be, sin(x), x+10, cos of a time variable. These equations are applied to each vertex of a 3D model, and that’s how you can create transformations of 3D models. How do we generate these equations? The genetic algorithm generates these equations, and the user never sees the equations, the user only sees how the equations transform the models. So the way we test our model in our experiments is as follows.
• #7: It looks like this. On the left there are the user’s own solutions, consisting of transformations for the x and z coordinates. On the right the user can see solutions from his/her peer, consisting of transformations for the y and z coordinates. So by selecting and injecting a solution into his/her genetic algorithm, the user adds the missing variable, and is able to evolve transformations for the x, y, and z coordinates. So now let’s take a closer look at our experimental setup. In our experiments, users use this interface to create transformations individually and collaboratively. Individually means simply using a genetic algorithm, without viewing solutions from peers and without the case injection and without expanding the search space. So the user sees 18 solutions from his/her GA. During collaboration, the user sees exactly this. 9 that belong to him/her, and 9 that belong to peers, and where the user can inject solutions from his/her peers.
• #8: We conducted our user studies using an alternating baseline setup. Here, our baseline is creating transformations individually – without collaboration. And we compare that to creating transformations collaboratively. Each user completed 6 trials in this order, individual trial followed by collaborative trial followed by individual trial, and so on. Once we evaluate the solutions generated for each of these trials, the goal is to show that the creativity score associated with these trials, zig-zags as illustrated here. We want to show that using collaboration, improves the creativity score, in this case low being better.
• #18: Scores for creativity are highly volatile. Novelty, surprise, value (usefulness)