Archive for January 27, 2010

Incorporeal Punishment?

My previous post was about robot dog trainers. Today I’m following up with some observations on an interactive canine art exhibit that might make me re-consider my opinion about robot dog trainers.

SNIFF is an interactive projection created by artist Karolina Sobecka and software developer Jim George.  This high-tech virtual dog interacts with passers-by in real time. From the artists’ website:

As you walk down the street you are approached by a dog. He is on his guard trying to discern your intentions. He will follow you and interpret your gestures as friendly or aggressive. He will try to engage you in a relationship and get you to pay attention to him.

Sniff is an interactive projection in a storefront window. As the viewer walks by the projection, her movements and gestures are tracked by a computer vision system. A CG dog dynamically responds to these gestures and changes his behavior based on the state of engagement with the viewer.

Video tracking data collected from infrared sensors allows SNIFF to interact with observers in real time. The positions of moving objects on the sidewalk outside the installation are tracked and a simple gesture recognition algorithm interprets them.  Fast, big movements are interpreted as threatening and slow, approaching actions are interpreted as friendly. SNIFF’s software stores a history of its interactions with viewers to form “relationships” with them over time.

SNIFF’s behavior strikes me as unnaturally awkward and stereotypic. Like a severely under-stimulated zoo animal. Note the eerie similarity between SNIFF’s movements and those of the caged Thylacine in the video below:

What makes SNIFF tick? According to the developers:

SNIFF is composed of two main components, a video tracking system and a game engine for real time graphics. The video tracking system is built in openFrameworks; for the game engine we chose is Unity3d.

People on the sidewalk are monitored by an IR camera in openFrameworks. In oF each individual person is isolated and assigned a unique id for the duration of their interaction. Each persons’ position and gesture information is continually sent to Unity3d via OSC networking protocol. In Unity, an artificial intelligence system representing the dog forms relationships with the individuals. He chooses which person to pay attention to, is able to move towards them or back away, responds to their gestures and initiates gestures of his own. Based on the interaction he gets excited or bored, friendly or aggressive, which is reflected in his behavior.

SNIFF’s algorithm includes a mood module that is constructed based on how each observer’s friendliness and enthusiasm changes over time. Like many real dogs, SNIFF reacts in a wild and unpredictable way when he gets over-stimulated. And when he gets bored with you he lays down or wanders off to investigate something else. SNIFF’s behavioral repertoire is currently very basic, but his developers plan to use data on how he reacts to his human audience to program him to engage in more complex interactions in the future.

So – if we can create a simulation that behaves like a dog – is the next step a virtual dog trainer? Perhaps, but I think that success with both projects still lies some time in the future. SNIFF only acts enough like a dog to provide us with a transient bit of entertainment. A simulation that reacts in a stilted and highly repetitive way to a few broad types of physical motions will need a major upgrade to evolve into a system that can consistently read canine body language, interpret it correctly, determine how the animal’s behavior should best be modified to achieve a specific goal and then respond to the dog in a way that correctly elicits the desired response. And, of course, the incorporeal nature of a simulation could also present problems in applying punishments and rewards.

Still – SNIFF’s ability to read and react to basic gestural behavior in real time could lead to some interesting developments in a lot of areas.

I think that even if we can build a canine training simulator with super-human observational skills and perfect timing – dogs will still prefer to work with flawed flesh and blood dog trainers because the dog is, and always will be, our first friend.

When the Man waked up he said,
“What is Wild Dog doing here?”
And the Woman said,
“His name is not Wild Dog any more,
but the First Friend,
because he will be our friend
for always and always and always.”
— Rudyard Kipling

January 27, 2010 at 2:47 am 9 comments


Because A Dog’s Mind is a Terrible Thing to Waste

RSS

Copyright notice

All original content on this Web site is copyright © on the date of publication by this author. All rights reserved except, of course, that others may quote from original content under the 'Fair Use' provisions of US copyright law.

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 284 other followers

Recent Comments

lynellex on House of Horrors
Mary on I Hate Jon Katz Too
Susie on I Hate Jon Katz Too
Susan Jaffe on See no evil. Read no evil. Cit…
Top Dog Blog
Featured in Alltop
ResearchBlogging.org

sitemeter

Add to Technorati Favorites
Dog Blogs - BlogCatalog Blog Directory
blogarama - the blog directory
Blog Directory
Blog Directory & Search engine
January 2010
M T W T F S S
« Dec   Feb »
 123
45678910
11121314151617
18192021222324
25262728293031