Hunter Hale

Engineering

Some coding on DASSIE interface working on knocking out mantis bugs

DEACCON work – Continue python implementation. Currently has some bugs on splitting that need to be cleared.

Science

Picked up a bunch of feedback on DASSIE while at AIIDE

Random

Attended AIIDE conference

Engineering
-Implemented a headless design of DEACCON that will adapt to any control system(CGUL UI or the temp one I built) to generate decomps
-Implemented OBJ loading
-Implementing 2d/3d versions of the algorithms(still working on this) – Collision detection and the branching conditionals that control growth and the edge/contour following are now the primary foci
-Implemented a 2D viewer that will easily switch over to 3D for testing by replacing one panel object
-Implemented SPSS exporter
-Generated poster for AIIDE currently stored on mac desktop under my account. For future reference anytime I need any publication graphics I should make them in omnigraffle that application is awesome. Currently in the process of upgrading poster with the latest improvements.
-I am currently also expanding and improving my slides for my presentation on DEACCON on wednesday.

Science
I have been looking at how the 3D expansions work and how to look at the collision detections using some modeling software. I believe I have figured out something interesting.

In 2D the negative space growing regions are composed of edges which are defined by vertexes. The vertexes are the only thing that can be subdivied(though really they are being promoted into an edge). In 3D I originally thought that I could just convert vertexes into faces if the vertexes collide with objects. However this will not be sufficient. Based on my modeling it looks like the edges and the vertexs of regions will need to be able to convert to faces in some cases.

I also think I see a nice decomposition evaluation metric for 3D decomps. Certain region formations (for example those shorter then a person) are damaging to the ability of agents to utilize the affordances provided by terrain/structures. I think we could look for situations like this in comparing decomp qualities. I know in the included picture showing the stairs decomps that the affordance problem there can be solved by allowing gravity to drop the seeds to the ground plane, but I imagine that there are other affordance issues that the algorithm will not be able to provide a consistent solution for. I think deaccon 3d could solve these problems by doing multiple decomps and then evaluating with a metric to pick the most usable one.

Paper Targets(Proposed):

Flairs – 1) A DASSIE paper of some sort possibly the arch stuff
2) A DEACCON dynamic rebuilding in real time paper(2D with stated 3D implications) also the technique presented would be generalizable to HM or SFV or most other forms of decomp methods. Since this isn’t DARPA related I should be able to submit it at the actual FLAIRS deadline
FDG – 1) The DASSIE Taxonomy stuff? I don’t know how good of a fit this stuff would be for this conference. However the timing of when things will most likely be ready fits a lot better for this conference than FLAIRs since more work has been done on the interface than on the taxonomy and that study is still ongoing.

Conditional UI interfaces

I have also done some thinking about how to do the conditional UI interface tool. Most conditional interface I have seen assume the binding order is in the order give. Most also allow the more advanced user to enter parens to define how exactly the bindings work. I think a basic user level tool that assuming binding from first to last would be acceptable for the basic user. Then on demand we display a tool that gives a more detailed binding order and allows the user to adjust it. Something like below.

Other Admin Stuff:
Almost done filling out my phd progress report form.