Saturday, June 15, 2013

Useless projects


My dream is to have people working on useless projects. These have the germ of new concepts. —Charles Eames

True, said she whose current research in artificial emotions was inspired by trying to make the
HRP-2 robot play "Au Clair de Lune" on the theremin. #ItWasPrettyBad

Friday, June 07, 2013

Could motherese be the key to autism?

I'm increasingly curious about the hypothesis that motheresespecifically, a lack of itis linked to autism.

I was recently surprised to learn that autism is not purely caused by genetics. A study at Stanford suggests that environmental factors play a larger role than we had ever thought. They noted that identical twins, where at least one was diagnosed as autistic, did not always share an autism diagnosisthis hints that nature is not purely to blame. Secondly, fraternal twins had a curiously high probability of both developing autism when one was autistic. Remember, these diagnosed pairs did not have the same DNA. This points a finger to something in their shared environment.

The link to maternal depression is also interesting. One study in California showed that mothers of children with autism were twice as likely to have taken antidepressants before giving birth. The researchers talk about the possible chemical effects of the drug, but I wonder if there's more to it. Maternal bipolar depression has also been linked to Asperger syndrome. Admitting a lack of proper reference, I can imagine that depressive mothers are unable to express a wide range of emotions when raising their childwhich brings me to motherese.

(from nature.com)

Motherese, also known as baby talk or parentese, is an exaggerated form of speech towards babies, with a highly varying pitch and slower rate. It is very emotionally expressive, happens across cultures, and presents uniquely in front of a baby -- according to Anne Fernald of Stanford University, mothers can't seem to produce it in front of a microphone. The parent places their baby about 30cm* from their face and coochy-coos, grabbing the baby's attention. Consider this account of a mother in Italy:

Paolo is a six-month-old baby who never looks at his mother, he doesn’t orient to her voice, sometimes he looks at his father when he uses a vigorous voice [...] she appears very anxious because Paolo doesn’t pay attention to her call: ‘Paolo, Paolo...look at me...my baby’. The spectrogram of the mother’s voice is flat without any prosodic pitch or long pauses.
After some minutes the video shows a totally different scenario where Paolo is in a rich verbal and visual dialogical interaction with his uncle. The analysis of uncle’s voice show all the characteristic of the motherese. (--Maestro and Muratori)
Basically, Paulo ignores the flat, yet pleading voice of his mother, but does pay attention to his uncle's melodic voice. Spectrograms can be seen here**. Paulo will later be diagnosed with autism.

Some child developmental psychologists (Cohen, Maestro) have suggested a link between motherese and autism, but statistical evidence is thin. It's difficult to track the amount and quality of interaction a baby gets, barring a Deb Roy-like Big Brother system in autistic children's homes. Plus, autism is not often noticed until after the first year, when it's perhaps too late.

In my work to make robots emotionally sensitive, I've suggested that motherese-like interaction could let us learn the link between emotions and parameters like speed, intensity, irregularity, and extent.
"Lim thinks there are certain identifiable qualities to emotional expression, and that when we detect those qualities in the behavior of an animal or the sound of a song, we ascribe the associated emotion to it. “I’m more interested in how we detect emotions in other things, like music or a little puppy jumping around,” she says. Why, for instance, should we ascribe sadness to a particular piece of music? “There’s nothing intrinsically sad about this music, so how do we extract sadness from that?” She uses four parameters: speed, intensity, irregularity, and extent—whether something is small or large, soft or loud. Angry speech might be rapid, loud, rough and broken. So might an angry piece of music. Someone who’s walking at a moderate pace using regular strides and not stomping around might be seen as content, whereas a person slowly shuffling, with small steps and an irregular stride, might be displaying that they’re sad. Lim’s hypothesis, as yet untested, is that mothers convey emotion to their babies through those qualities of speed, intensity, regularity, and extent in their speech and facial expressions—so humans learn to think of them as markers of emotion." —Nautilus
What if autistic babies are not getting this training data? Perhaps they never learn to associate rapid-loud-rough-broken with anger, or andante-flowing-large with happy. Perhaps the brain, which is still developing its specific sensory regions during the baby's first year, does not develop these necessary connections.

These parameters also seem to be important for deaf infants, who require "visual motherese":
"Are you smiling, and letting your signs flow? Are you frowning and signing sharp, emphatic signs as you run to cover the electric outlet? Are you pretending to cry as you see a sad character in a story?" Boys Town National Research Hospital
We could test whether at-risk babies are learning these basic dynamic traits properly. As early as 7 months old, normally-developing babies can match happy and angry voices and faces—in other words, the babies look longer when the happy face matches with the happy voice. At that age, they even tell the difference between happy and angry moving lights. And at 9 months, they can distinguish between happy and sad music. Have we tried testing babies at-risk for autism with this battery of simple, yet multimodal emotional tests?

Motherese is important for a couple other very important skills associated with autism spectrum disorder (ASD). For instance, motherese, with its accentuated, drawn-out speech is necessary for children to learn languageTV doesn't cut it. I wonder if a lack of motherese would impair both a child's emotion recognition ability and cause delayed language development. Who knows? Perhaps the rich eye contact between mother and baby also helps develop this other social skill found lacking in those with ASD. As a non-expert in autism but a specialist in artificial emotions, this is speculation, so I welcome critiques and feedback.

So does that mean that we should speak in a sing-song voice to our babies to prevent autism? Well, it could be more specific than that. It's possible that motherese, which also involves a feedback loop of empathy and mirroring, helps a baby connect what he hears and sees with his own internal emotional state ("My diaper needs changing and I'm sad!", says the mother making a sad face at her daughter). In other words, maybe it's the empathy part of motherese that's important. I don't know.

I don't want to suggest that there is any "fault" to lay to parents of children with ASD. It's clear that all parents love their children and do everything in their power to teach them the skills they need for success. But what if we are simply unaware of the correct way to do it? Where is the guidebook for motherese? What about people who say, "Why are you talking so much to your 3-month-old? She doesn't understand, anyway." How do we teach them about emotions? Should we all take acting classes?

As a lowly roboticist, I can't answer these questions. But I hope these ideas will get to the psychologists and researchers who can. Please repost.

* Newborns focus best on objects around 25cm-30cm from their eyes. Adult-like vision is only developed at around 6 months.
** In French. See under the name of Pedro.

Thursday, June 06, 2013

Healthcare robots

Yesterday I attended a talk by our new post-doc, Osamu Sugiyama. He's worked with the tiny Robovie MR-2 at ATR:



The robot could be used as a healthcare robot, he said. For example, when patients visit the doctor, explanations can get long and complicated. A bit embarrassed, patients just nod and pretend to understand instead of asking questions. So he proposed a healthcare robot that patients would be comfortable asking many questions to.

Also, even if the patient understood instructions at the doctor's office, they could forget once at home. A healthcare robot could remind them to make sure that they continue to follow doctor's orders.

Wednesday, October 17, 2012

Ada Lovelace Day - Céline Boudier, Robot Programmer

On this Ada Lovelace day, I'd like to express my admiration for an inspiring robotics programmer, who I think has the coolest job in the world!



Céline Boudier has been a NAO robot Behavior Architect at Aldebaran Robotics for almost 4 years. She has worked on countless projects to bring the humanoid robot NAO to life. To name a few, she's made a robot enigma adventure game called Shana Spirit Journey, a NAO choose-your-own adventure storyteller app, a robot Akinator, Doctor Who behaviors, and more. Her projects have been downloaded hundreds of times. On top of these fun and creative applications, she also is in charge of developing behaviors to help treat children with autism.

Her tutorials are useful and have made an positive impact on many other robot developers.

She is artistic, funny, quirky and cool.

So hats off to Céline Boudier, robot programmer extraordinaire!

Wednesday, May 02, 2012

How to batch convert 1-up photos to 2-up in Mac OS X

I've recently started printing photos again and wanted to get more bang for my buck. Instead of printing 200 photos of a normal size, I wanted to print 100 photos normal size, plus 100 of them 2-up, halfsize. The benefit? Only paying for 150 prints, and having a variety of photo sizes to go into my scrapbook.

In other words,

These:




Become:


Here's a script I wrote to convert those 100 photos into 50 2-up photos.


#!/bin/python


import os


# Create output directory
outputDir = 'output'
directory = os.getcwd()
try:
    os.makedirs(outputDir)
except OSError:
    pass


# Generate 2-up photos
counter = 0
prevFile = ''
curFile = ''
for root, dirs, files in os.walk(directory):
for f in files:
filename, extension = os.path.splitext(str(f))
if (extension.lower() == '.jpg'): 
counter = counter + 1
if counter % 2 == 0:
curFile=filename
cmd = 'montage '+ prevFile + '.jpg ' + curFile + '.jpg -geometry +2+2 -tile 1x2 ' + outputDir+'/'+ prevFile+curFile + '.jpg'
print cmd
os.system(cmd)
else:
prevFile = filename

How to use it:

1. Install ImageMagick, making sure to do all the export commands in Terminal.
2. Using Terminal, change into the directory containing the photos you want to convert.
3. Create a file called script.py in that directory and paste the above code into it.
4. Run the script using python script.py and the 2-up photos will be placed into a folder called output.


Note: This works only on .jpgs for now but could easily work with any other photo format by replacing the .jpg's in the code above to whatever format you use. 


Thanks to Ben for the 2-up idea!

Wednesday, April 25, 2012

Robot ideas

I've been making a list. I'll let you figure out what the theme is :)


  • A robot to scare away birds
  • A robot to feed the fish
  • A robot to water the plants
  • A robot to accept deliveries
  • A robot to read stories
  • A robot to talk about a shrine's history
  • A robot to answer questions about a location
  • A robot to give directions
  • A robot to give opera performances
  • A robot to introduce a band or orchestra
  • A robot to MC a wedding
  • A robot to tell jokes
  • A robot to explain cultural differences
  • A robot to play with kids
  • A robot to ask questions to kids
  • A robot to answer kids' questions
  • A robot to wash dishes
  • A robot to sort laundry
  • A robot to take the laundry out when the wash is done
  • A robot to hang laundry
  • A robot to iron clothes
  • A robot to stir sauce
  • A robot to tell when the cooking is done
  • A robot to take pictures of the family
  • A robot to take panoramas
  • A robot to fix broken chairs
  • A robot to sympathize
  • A robot to give encouragements
  • A robot to give candy
  • A robot to sell stuff
  • A robot to be your telepresence avatar
  • A robot to surprise people
  • A robot to keep someone company on a walk
  • A robot to fly
  • A robot to bang dirty carpets
  • A robot to dry towels
  • A robot to take out the trash
  • A robot to sweep the floor
  • A robot to sweep leaves
  • A robot to cut the grass
  • A robot to paint what it sees/dreams
  • A robot to distribute flyers
  • A robot powered by solar panels
  • A robot to guard a door
  • A robot to explain how to play a game
  • A robot to juggle
  • A robot to sing
  • A robot to do 'marionnette'
  • A robot to cry
  • A robot to change the TV channel
  • A robot to play the radio
  • A robot to play pretend with kids
  • A robot to scrub the tub
  • A robot to cut onions
  • A robot to peel
  • A robot to crush garlic
  • A robot to make bentos
  • A robot to guard the home
  • A robot to jack (steal) your stuff
  • A robot to be anti-sympathetic
  • A robot to be Arnold Schwartzanegger
  • A robot to be the Simpsons
  • A robot to say D'oh
  • A robot to be Morgan Freeman
  • A robot to echo everything
  • A robot to be your chorus
  • A robot to be Siri
  • A robot to be Smeegol
  • A robot to say "That's what she said"
  • A robot to make fart sounds
  • A robot to fetch the tennis ball
  • A robot to check the temperature of chocolate and stir
  • A robot to harmonize with you
  • A robot to transform your voice into Darth Vader or anonymize it
  • A robot to watch you sleep (creepy)
  • A robot to watch your sleep patterns to wake you up at the right time
  • A robot to make baguettes
  • A robot to make chocolate mousse (en fait on utilise déjà un robot pour en faire...)
  • A robot to make a sandwich
  • A robot to sort black socks
  • A robot to monitor a baby
  • A robot to pour beer with the right amount of foam
  • A robot to do your presentation for you
  • A robot to tie shoe laces
  • A robot to start up with a theme ("Go Texas!")
  • A robot to take a picture without getting your thumb in the photo
  • A robot to serve soft-serve ice cream
  • A robot to knock over all the cups it sees
  • A robot to collect all red things in the room and put it in the corner
  • A robot to turn off all the lights
  • A robot to give out tissue paper to passerby
  • A robot to give tissue paper when you cry
  • A robot to give you a recipe idea based on the day's weather and seasonal ingredients
  • A robot to sing romantic songs based on a girlfriend's request ("When I see your FACE...")
  • A robot to question your decisions
  • A robot to make prophecies
  • A robot to say really insightful things (bonus points for rhyming)
  • A robot to make up surprising haiku's
  • A robot to translate for you with terrible accuracy ("She says you stink")
  • A robot to play Scrabble
  • A robot to play Truth or Dare

Sunday, January 01, 2012

Hiroshi Ishiguro tweets

Dr. Hiroshi Ishiguro tweets rather profound things. I've started translating his book here, but in the meantime, here are a couple of his tweets from today that I've translated into English.


hiroshiishiguro 石黒浩
価値 高校生に”自分に価値があるかどうかわからないんです”と聞かれて,”僕もまだ探している途中だよ”と答える.それ以外に答えがない.
Worth: I was asked by a highschooler, "I don't know if I'm worth anything." I replied, "I'm still looking, you know." Couldn't say much more than that.

hiroshiishiguro 石黒浩
自分 本当に難しい問題にチャレンジしているか?結果が見えるようなチェレンジはチャレンジではない.今の自分のままで到達できるゴールはたかがしれている.本当のチャレンジとは自分を変革を伴うもの.自分を変えながら全く新しい道を歩む.だからゴールも結果も見えていない.
Self: Are you really challenging yourself with hard problems? If you can see the outcome, it isn't a challenge. Any goal that I can attain as I am now doesn't amount to much. A real challenge is accompanied by reforms to oneself. As you change yourself, you'll walk along a completely new path. Thus, both the goal and end result cannot be seen.

hiroshiishiguro 石黒浩
自分 人間というのは常に進化してるはず.だから,もし自分も進化しているなら,自分が何になるかはなってみないとわからない.今から自分の将来がわかっているというのは,それは進化をやめたということではないか.チャレンジをやめたということではないか.
Self: Humans should always progress. Indeed, if humans progress, this means they don't know what they want to become, because knowing one's future from now means giving up on progress, doesn't it? Isn't that giving up on challenges?

hiroshiishiguro 石黒浩
自分  自分が何者かがわかった時,ああ,自分はこんなやつ何だとわかった時というのは,それ以上自分が進化しないと認めて,あきらめた時なんだろう.
Self: When you understand what kind of person you are, when you understand that 'oh, I'm this kind of chap', if you realize that you're not progressing to someone/thing higher than that, haven't you given up?

hiroshiishiguro 石黒浩
無いものねだり 幼少の頃の無いものねだりはただのわがままかもしれないが、大人の無いものねだりは、夢の実現や発明につながる。
Asking too much. Asking for frivolous things during infanthood is simple indulgence, but asking for frivolous things during adulthood tends to be related to realizing or inventing your dreams.