Lecture on Nov 16, 2009.

Guest lecture by Maneesh Agrawala, UC Berkeley.


  • Required

    • Pictorial and verbal tools for conveying routes, Lee & Tversky (pdf)

    • Rendering effective routemaps, Agrawala & Stolte (pdf)

    • Identification and validation of cognitive design principles for automated generation of assembly instructions, Heiser et al. (html)

  • Optional

    • Designing effective step-by-step assembly instructions, Agrawala et al. (html)

  • Demonstrations


fxchen wrote:

Agrawala's research serves to inspire visualizations that correlate closely with hacks we perform but don't think about in our daily lives. LineDrive is awesome. I have a number of small notecard sized maps drawn identically to this style in my car. It's really unfortunate Microsoft's map search is not very intuitive. The clunky search interface really hinders my drive to switch from googling my maps.

Finally, I thought y'all would enjoy this, it's peripherally related. Eisenhower Interstate System in the style of H.C. Beck’s London Underground Diagram: http://www.flickr.com/photos/senexprime/4055072020/sizes/o/

jqle09 wrote:

The LineDrive maps were really neat. I think it would be really nice if the GPS systems in cars worked like this. Although while driving, the local view around the car seems to work well already, and eliminates the need the need to have the whole route mapped out at one time like in a LineDrive map.

I especially liked the described methodology of identifying design principles then approaching a visualization problem. Developing systematic reasons for why any visualization works better than others seems like the most important aspect of creating a good visualization. Also I would have really liked to be a part of the assembly instruction experiments; assembling the table looked like it would be fun.

jieun5 wrote:

Getting lost or making a wrong turn while following driving directions occur very frequently, and unless we have a GPS system that "recalculates" the direction for us, it's often stressful to get back on track. I wonder how LineDrive compares with a more traditional in-proportion, non-simplified map in helping the user find his way back in these situations... I feel that there is a trade-off between simplicity and abstraction (found in LineDrive) and details and information that are "to-scale" found in traditional maps.

cabryant wrote:

I will echo jqle09 in my appreciation of the approach employed by Agrawala et al in developing an automated system for generating assembly instructions. Their methods reminded me of Grounded Theory, a form of qualitative research that generates hypotheses/theories from sampled data (cf. the "scientific method"). I also found their findings compelling (that their auto-generated diagrams promoted more rapid assembly times and reduced errors), but I would like to see followup studies that reduce confounding variables between the subject, factory, and computer generated diagrams. E.g.:

  • Running the same final experiment but with similarly colored hand-drawn and exploded diagrams.
  • Running the same final experiment but with exploded diagrams color-coded to imply a sequence of assembly.
  • Running another series of experiments in which subjects are asked to "create step-by-step visual instructions for a competent assembler" (rather than "write instructions using text, diagrams, or a combination so that a novice assembler could easily and efficiently assemble the TV stand").

On a side note, I enjoyed the lecture segment regarding interactive, three-dimensional renderings of building layout. As an avid museum-goer, I immediately thought of several structures that could benefit from this form of visualization (particularly given that many museum structures are works of art in their own right that are difficult to appreciate without a global perspective). For instance, I would love to see a version of Wright's Guggenheim Museum

bowenli wrote:

I agree with Jieun in that traditional maps, even point to point directions from an online map service are often useful for people who need additional cues about the location. It's useful to see what other streets and landmarks are in the area in case of missing a turn or simply not sticking to the plan exactly.

One area that I find interesting about LineDrive is that it provides very minimal information to the user and is basically codifies the step by step instructions in a visual way. I think this approach will lend itself much better to giving directions around landmarks. If you consider people's behaviors, often directions are given in the form of "turn left at the 7-11" or "there will be a large statue on the right hand side". Landmark cues like these seem better suited for the LineDrive system than a traditional map with highlighted route (similar to fig3 in Tversky).

I felt the assembly project was well motivated but I have some hesitations about the method. The steps given by the system were optimized concurrently with the presentation. The paper didn't give a strong argument for doing this. I feel that the planning phase has certain criteria such as ease of assembly, or maximizing structural integrity, or building up smaller functional units before larger ones, etc. But it doesn't seem like the criteria of "we can show this" is as important as a purely mechanical criteria. In that sense, it doesn't really make sense to optimize both at the same time.

Also, one of the motivating examples given in the presentation was a grill that had multiple components. The examples shown in the paper are all monolithic structures. Does the system handle building up multiple subcomponents?

rnarayan wrote:

@jieun and @bowenli - good questions - in my opinion, the best system for recomputing directions if a turn is missed is any of the GPS systems with voice readouts. Garmin, TomTom, etc. continually use the present driver coords and automatically recalculate the new directions for instance from the wrong spot. Now, these turn-by-turn directions are available on the iPhone (for a pricey subscription) and lately I heard a free version is slated for Android mobiles. In general, these voice-enabled GPS systems have long obsoleted online maps for general purpose driving directions.

If only purely visual geospatial representations are considered, LineDrive does illustrate a couple of nifty design principles (such as local focus+context and redn of clutter) that is more of an afterthought or lacking in most of the early and several current map implementations. Having said that, at the present moment there are several map features and technologies that have long since solved these issues. For one thing, Google StreetView is fantastic when it comes to local navigation. Even without StreetView, Google gives axonometric projections for landmarks at higher zoom levels. Other map providers (CloudMade, etc.) provide rooftop and birdeye views as well. With a full map, it is also possible to get traffic encoded routes and even identify alternative routes.

Instead of a separate map like LineDrive, I asked in class about providing a bi-focal distortion for the local region on mouseovers. Although Dr. Agarwala's answer was correct (difficulty in properly merging the distorted and undistorted areas), I found that there are now newer map imaging techniques with tiles (BingMaps, etc.) that can make this acceptable (similar to http://www.flashxml.net/pyramid-gallery.html). Of course, this requires interaction. For a non-interactive method - if we further consider the use case scenario, a typical user needs expanded/scaled routes for only one end of the To/From location since she/he is likely already familiar with his starting surroundings. So, a single corner inset for the other local area at the higher zoom will likely suffice.

Lastly, maps like CoPilot (iPhone) have a SmartZoom capability that automagically increases zoom based on driving speed (so highways at cruising speed are at lower zooms than local areas).

rnarayan wrote:

Re: assembly instructions - applying planning algorithms (search, sequence determination, etc. ) from AI for this problem is an excellent choice in design - it may also have a sizeable commercial potential in having robots configured/programmed to automate these tasks for everyday items such as toys, appliances, etc. One important element missing in the section on Presentation in this paper (and talk) was the use of Animation for the same. For human assembly, there seems to be no better substitute than animating the (dis)assembly for fastest comprehension. Please see this video: http://www.youtube.com/watch?v=6-HiBDLVzYw (note: is a 9+min video - takes a while to load)

nmarrocc wrote:

Its a fascinating distinction between description and depiction. I use google maps on my ipod to navigate when driving some place new. I find that I often switch between the two modes. Usually I look at the map before I leave to get a feel for where I'm going and how the trip might go. Then I switch to the description mode, mostly because its faster to get the information I need, which is helpful when driving and looking at the screen at the same time. I imagine if I had a fancy program that could read to me that would work too, but maps will probably always be tough to look at while driving.

zdevito wrote:

Line drive reminded me of the Marguerite shuttle map (http://transportation.stanford.edu/pdf/2009-10-marguerite-map.pdf) I looked at while doing Assignment 3. When you first look at, it is immediately obvious that it is a map of Stanford, but only if you look at it closely do you realize that it is not at all to scale: both 280 and 101 are displayed on the map, many of the roads that the buses travel on have been straightened out , and the areas of interest in the center of the map are enlarged to make the routes easier to see.

Leave a comment