Identifying Design Principles

From CS294-10 Visualization Fa07

Jump to: navigation, search

Lecture on Oct 17, 2007

Slides

Contents

[edit] Readings

  • 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 Readings

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

Demonstrations

[edit] Ken-ichi - Oct 17, 2007 09:14:48 am

I love the idea of LineDrive, and would love to do some comparative tests. However, the inclusion of undistorted maps at only the start and end points of a route presumes that people don't get lost in the middle of their journey, which, at least for me, is certainly not always true. Maneesh, did you do any tests to see how much contextual information you could cram into a map before usability dropped down to the level of a conventional map?

[edit] Robin Held - Oct 19, 2007 03:29:12 pm

It seems like Google maps could really benefit from some of the analysis that went into LineDrive. Whenever I use google maps to generate directions and choose to print them, I find myself rescaling and repositions the small map windows they create for the start and end points of the voyage. Usually, they zoom in too much on the destination, so it's difficult to use the map to figure out the last few turns. Of course, as Maneesh mentioned in class, people typically rely on the written descriptions anyway. But if they're going to include a mini-map of the destination, it seems like more care should be put into its contents. Also, 90% of the time, I want directions from my home, so the minimap for the origin is unnecessary and wastes a lot of space. Googlemaps lets you turn off the origin minimap, but it may be useful to also allow you to rescale the destination and general minimaps.

[edit] Christopher Volz - Oct 19, 2007 11:50:54 pm

What I found particularly interesting about these readings is that the instructions for effective directions could be reduced to a relatively small toolset for both written directions and simple maps. It called out something that is intuitive but not necessarily obvious, which is that travel is essentially a simple repetitive process comprised almost entirely by forward movement and turns and effecitve directions can be boiled to just the essentials when required. To be fair, though, to the map makers who add all that extra information into the maps -- and I'm specicifically thinking about those big folding paper maps you get at gas stations -- all that extra information comes in very handy when your trip takes an unexpected turn, such as running out of gas, blowing a tire, or even just needing to find a restuarant for lunch during a long drive. That said, I agree, for most routine directions people are concerned about getting between A and B in a simple and fast a way as possible.

[edit] Mark Howison - Oct 21, 2007 03:54:46 pm

Although LineDrive maps provide the user with clear information on where to turn along a route by giving cross street names and distances, this isn't fail proof. Cross streets can be poorly marked and not all drivers think to measure distances with an odometer. Another way to encode the information for turns would be to provide a photograph of the intersection, possibly drawing on a data source like Google's Street View. A storyboard with photographs of each intersection could be provided with the map and text directions. Before implementing such a system, it would be a good idea to run a cognitive experiment to test for drivers' ability to map perceptual features of the photographs onto the real-life view. It may be that making sense of the photograph is highly dependent on aligning the angle of view of the photograph with that of the driver's view. Or that seasonal factors, such as snow or foliage, cause difficulty.

[edit] James Andrews - Oct 24, 2007 12:15:20 am

Did the automated generation of assembly instructions evaluation compare color, computer generated drawings to black and white instructions? If so, it's a strange thing to do; although the key point (that the instruction generated are of acceptable quality) is relatively unharmed, it seems to decrease trust in the results, and I imagine it wouldn't have been hard to make black-and-white computer instructions as well. I'd be curious how much the 'small' details of presentation (like color, correct perspective, etc) affect users.

[edit] David Jacobs - Oct 24, 2007 10:26:00 am

I think one of the problems with systems like LineDrive is the relative lack of contextual information. This is fine assuming nothing goes wrong, but Murphy's law makes things difficult. One of my scariest experiences driving around using written directions was getting lost around downtown San Diego (somewhere I've never been before). The freeway exit recommended by Google maps was closed, so I had to wander through the one-way streets until I found a road on the recommended route (and even then I wasn't 100% sure about my orientation along the path). If it weren't for the full detail minimap near the end of the route, it would have been much harder to find my way back -- turns out downtown SD uses a grid system of streets A street and 20th Ave, etc. In any case, if quantity of paper isn't an issue, I think the best combination would be a set of LineDrive instructions and a full set of AAA maps along the route for unexpected detours.

[edit] Amanda Alvarez - Oct 24, 2007 06:34:10 pm

In their paper, Tversky and Lee suggest that depictions and descriptions have similar underlying structure; route maps and route descriptions were composed of similar elements, which fuelled the suggestion that there could be ways to automatically translate between the two. Yet the reason why handdrawn maps and LineDrive maps are so successful is because they use generalizations, distortions of length, angle and shape. These distortions, if translated from depictions into descriptions, would be useless, misleading, or unintelligible. In these cases, the depictions (maps) hardly "force concreteness"; we seem to find it much easier to accept that pictures can lie, whereas the same is hardly true for words. Not just the LineDrive maps, but also the carefully crafted furniture assembly instructions are distorted, yet easily matched up with a real world task. The assembly instructions are boiled down to the strict basics, and can clearly be followed, but you would never follow them in precisely the way depicted, eg. you might have to turn the tv stand upside down to put on the wheels. The task of furniture assembly seems to accommodate depicted or described distortions much better than following directions while driving does; the underlying cognitive structures change by task and are not absolute.

[edit] Mcd - Oct 28, 2007 02:26:45 pm

I tend to agree that a principal purpose of having larger more detailed maps is not necessarily for route-finding, but for navigation without precise directions, i.e. when you get lost. I think of hand-drawn, low detail maps as useful for short trips on local roads, but disorienting for longer drives. I looked up cross-country directions, and thought that the result would not necessarily have been helpful. On long drives, I check for landmarks and intersections frequently to mark my progress and confirm I'm on track, and suspect I would thus find the lack of detail frustrating--I would constantly be wondering what I didn't know and wasn't seeing. I wonder, too, if user tests would reveal a similar bias toward shorter trips, or if I am just an outlying map-lover.

[edit] N8agrin - Nov 03, 2007 09:51:15 pm

It's interesting to see Tversky and Lee's identification of pictorial and verbal tools applied in Agrawala's implementation of LineDrive. Reading these two papers together really drove home how powerful ethnographic and psychological research can be when used to inspire computer created visualizations.

LineDrive, by my experimentation, works incredibly well, and is one of those "Wow, why didn't I think of that?" ideas. While it does seem to suffer some shortcomings like not handling long distance routes very well, it does seem to map directions for short distances elegantly. As other have commented longer distances might benefit from some contextual cues, like major landmarks. While collecting this data is somewhat difficult, I wonder if it couldn't be outsourced to a community in some way. For example, the algorithm could generate the location of a lake based on its best guess, but a user might know that the lake is further north on the road, and moves it. Still other users might move it again to make it's position more accurate, each time providing the algorithm with a bit more data to learn from, until eventually the lake's position ends up in a configuration the users generally agree with.

I played around with LineDrive a bit on the mapblast.com site (now MSN) and was surprised that when I regenerated a map using the same two endpoints, the maps were identical. I was under the impression that the labeling algorithm and the road drawing algorithm had random components to them which helped the overall map reach a sort of "lowest energy state". It was my assumption that this randomness would be reflected, but you can see by the two maps below that the same maps are generated. This may be a one-off case and to be fair I didn't try the same problem with any other sets of directions. Perhaps my directions were too short and the lowest energy state was easily achieved by the algorithm, requiring few perturbations.

Image:LineDrive1.jpg

Image:LineDrive2.jpg

Also, here is a funny example of LineDrive's interpretation of very long distance directions:

(Note how insignificant the NJ Turnpike looks and the relative short piece that represents I80) Image:LongDistanceLineDrive.jpg

[edit] Hazel Onsrud- Nov 04, 2007 01:18:46 pm

I think David makes a really good point about the importance of other contextual data. I know I generally find my way by comparing my relative path with its surroundings and deducing a direction from each point, thus this map would be generally helpful, especially in northern Maine where there aren't too many roads, but a more detailed map would be better for trickier, more populated areas. That said, I know other folks who get confused by the overabundance of informaiton and just want to know where they are relative to the compass points. This would be perfect for them.



[add comment]
Personal tools