[nfb-talk] Guide Bots

Thu May 30 15:02:23 UTC 2013

I've struggled with the how of a guide bot quite a bit.  Ideally, I would
not have considered a device to follow.  Initially I thought about a tactile
map, let's say, on your back and you would get a picture of the environment
in front of you.  The problem is that such a tactile reproduction of visual
information would be two dimentional.  That is, it would only give you the
general position of an object but not its distance from you.
It would also be extremely expensive by any measure I've devised.
And then there's the fact that traveling involves a lot of defining by
texture.  We get lots of textural information through our canes and our
dogs.  Slight slope up, switching from grass to pavement, ramp down, ramp
up, pot-hole in the sidewalk.  All of these things are best identified
directly rather than through some tactile representation.  
So, part of my guide bot concept is that the bot would be in front of me and
I would be picking up textural information through the handle of the device.
I've even thought that the device might have two wheels going forward with a
series of perpendicular wheels that would allow you the ability to sweep as
we do with canes.
I'm not sure about that idea though because much of the distance measuring
that the two cameras would be doing would depend on a focusing ability that
would be very difficult if the device was moving left to right all the time.
You raise an interesting question about whether one could, or would, trust
such a device.  I think lots of testing would have to be done and, just as
with a cane or dog, you would have to believe that it was giving you mostly
accurate information.  Just as with a dog, you would have to be prepared to
over-ride at any moment.
If the device gave you bad information often enough you'd send it back, just
like sending a dog back to the school.
As a technical matter, what we would have to do is "teach" the device all of
the things I've described.  It isn't like we would have the device engaging
in a number of programmed responses, although that's part of it.  What AI
folks have discovered is that you are far better off to give the device some
general instructions and then let it learn from those general instructions.
That is why I'm trying to get this project to the attention of Dr. Kurzweil
because he's got the best grasp of whether such a thing can be done in a
practical world.
I'm really not sure about some of the ideas and how complex they would be.
For example.  You come up to a corner and are waiting to cross the street.
So, the device sees the walk sign turn and alerts you.  You then start
across the street.  If there is a car turning and its trajectory will
intersect with your path of travel, is there enough computing power to
ascertain such a thing.  What humans and dogs can do with their eyes,
gauging distance and trajectory and speed look simple, but they are very
complex tasks that may be beyond the computing power one would have in a
portable device.
But, if we could do such a thing it would help identify those pesky quiet
cars.
Mike
-----Original Message-----
From: nfb-talk [mailto:nfb-talk-bounces at nfbnet.org] On Behalf Of Steve
Jacobson
Sent: Wednesday, May 29, 2013 10:17 AM
To: NFB Talk Mailing List
Subject: Re: [nfb-talk] Guide Bots

Mike,

I do not disagree with the assertions in this note.  I would add that when I
was younger, my hearing allowed me to gather additional information beyond
the reach of my cane.  I still use my hearing to augment my cane, but it is
somewhat less effective as I get older.  

Where my view of the problem differs from yours is with the idea that we
need something to physically guide us.  I think that adds complexity that is
not really necessary.  What we need is a device that gives us more feedback
about the environment around us that we may not get with our cane or with
our hearing.  There are devices that already try to do that, and some are
somewhat successful.  I've seen some good ideas and some that had
limitations as part of our R&D Committee over the years, but it has just
felt to me that something better could be developed.  Therefore, I think
this is a topic worth exploring, and would be interested in learning more
why you feel that a solution would have to physically guide us rather than
just provide us with useful additional feedback.  For one thing, if
something is guiding us, how to we loearn what kind of feedback it is
reacting to and whether its reactions are even correct?

Best regards,

Steve Jacobson

On Tue, 28 May 2013 16:48:33 -0400, Michael Bullis wrote:

>Let me see if I can bring the discussion of a guide bot to its essence 
>without making it too controversial.
>Sometimes, a pair of eyes is very useful in travel and more efficient 
>than a cane.
>I'm a lifelong cane user and would probably be described as an 
>excellent traveler.  But, having said that, there are some things that 
>a pair of eyes, whether provided by a dog or provided by a human, can 
>do that I cannot do as quickly with a cane.  One example would be 
>working through a crowd.  Another example would be moving through a 
>room that has many obstacles in it that need to be circumvented.
>All of this can be done with a cane but I have to slow down to some 
>extent to examine the objects and work my way through them.
>If a guide-bot were to be practical the question would be could it give 
>me enough information in advance to move me through these situations 
>more efficiently.
>I must hasten to add that I would still keep a cane with me just as I 
>did when working with a dog because there are times when I need to 
>orient myself in the environment.  In other words, neither a dog nor a 
>guide bot can replace my own awareness of where I am and where I'm going.
>This morning, while on the way to work, I hit my head on three very 
>annoyingly wet tree branches that were hanging down at about face height.
>Interestingly, there was a clear path of travel which, if I had a pair 
>of eyes, I would have seen, but the cane couldn't tell me where that 
>path was because the branches were at head height, something the cane
doesn't cover.
>A well trained dog could have seen these branches and would have 
>perhaps maneuvered me past them.  And this guide bot could hopefully do 
>the same thing.
>I walk fifteen or twenty miles a week with my cane, often in new and 
>therefore unfamiliar neighborhoods.  Because I walk in large part for 
>the exercise, it is annoying to have to slow down to analyze a new 
>situation--fire hydrant in the middle of the sidewalk, Etc.  If a guide 
>bot could provide me with a quick way of maneuvering in these new 
>environments it would be worth having, assuming it didn't break my limited
bank account.
>So, in my view, much of the benefit of the bot will depend upon how 
>many things it can do that I now find slower to do.  Nothing will stop 
>me from traveling but if the bot worked, it would make it more 
>efficient in some situations.
>My reason for putting this out there is to tear it apart and ask the 
>questions that need to be asked.  Thanks for all the feedback.
>Mike
>Mike Bullis

>-----Original Message-----
>From: nfb-talk [mailto:nfb-talk-bounces at nfbnet.org] On Behalf Of Misty 
>Dawn Bradley
>Sent: Tuesday, May 28, 2013 1:42 PM
>To: NFB Talk Mailing List
>Subject: Re: [nfb-talk] Guide Bots

>I am the same way. I have been using my cane since I was in 
>kindergarten or first grade, which is about 20 years ago, so I don't 
>feel right when I don't

>have my cane with me. I also like the challenge of figuring out things 
>on my

>own, although it is nice to have a GPS sometimes, but it is not a
necessity.
>Misty

>----- Original Message -----
>From: "Gloria Whipple" <glowhi at centurylink.net>
>To: "'NFB Talk Mailing List'" <nfb-talk at nfbnet.org>
>Sent: Tuesday, May 28, 2013 1:33 PM
>Subject: Re: [nfb-talk] Guide Bots

>>I am not sure about this. I am so use to using my cane and like it.
>>
>> Gloria Whipple
>>
>> -----Original Message-----
>> From: nfb-talk [mailto:nfb-talk-bounces at nfbnet.org] On Behalf Of Steve
>> Jacobson
>> Sent: Tuesday, May 28, 2013 10:28
>> To: NFB Talk Mailing List
>> Subject: Re: [nfb-talk] Guide Bots
>>
>> Mike,
>>
>> To me, the idea of a device that is responsible for guiding, even if made
>> simpler as described here, adds so much complexity and
>> size that really has limited advantage to my way of thinking.  A small
>> device that could be mounted on a cane or connected to a
>> dog's leash to detect objects and provide feedback would provide most of
>> what a guiding device would without the size and other
>> complicating issues.  Why does it have to aim itself at all if the blind
>> person were to receive enough information to do the
>> aiming from a smaller travel aid?
>>
>> Best regards,
>>
>> Steve Jacobson
>>
>>
>> On Tue, 28 May 2013 11:24:59 -0400, Michael Bullis wrote:
>>
>>>I've been thinking about a new travel device lately and would be 
>>>interested
>>>in any feedback members might have.
>>
>>>I've been wondering lately if there is enough reasonably priced portable
>>>processing power available to make a guide bot.
>>
>>>From looking on the web there are a few projects, mostly not very 
>>>efficient
>>>ones.  There is a company called nst that is working on it but expects to
>>>bring a product to market in the 2020's.  The reason most projects are so
>>>difficult is that they are spending inordinate amounts of time and money
>>>teaching the guide bots to climb stairs and fully "lead" the blind 
>>>person.
>>>We know this is unnecessary.
>>
>>>If you remove the necessity to "lead" the blind person the project
becomes
>>>much simpler.
>>
>>>The device doesn't need to climb stairs.  The blind person can simply
lift
>>>it up or down.  This assumes of course that it is a lightweight  device.
>>
>>>For example, suppose you have a two wheeled device with two forward
facing
>>>cameras  and a handle on it that you direct.  You push the device down
the
>>>sidewalk let's say.  When it detects an object that you should go around,
>> to
>>>the left, the left wheel has a brake on it.  When the left brake engages
>> and
>>>slows the left wheel, the device naturally turns that direction.  The
same
>>>thing happens if you want to go right.  If there's a curb coming, both
>>>wheels slowly engage their brakes.
>>
>>>At any time, if you as the person in charge wish to override the braking,
>>>you simply give a quick shove to the device.
>>
>>>The blind person provides the forward motion, eliminating any need for
>>>motive power.  The only thing the device needs power for is processing
and
>>>braking.
>>
>>>Whether verbally, or with a small keypad, you tell the device what you 
>>>want
>>>it to do.  If the device is going to talk to you at all it needs to be
>>>through an open-ear earbud.  But this may be unnecessary.
>>
>>>
>>
>>>I believe it would require two cameras for measuring distance and for
>>>scanning both in front and to the sides.
>>
>>>In order for the device to be practical it would have to solve problems
>> that
>>>the cane and most dogs don't solve.  That is, it would have to detect
>>>objects at head height.  And, it would have to be fast.  The thing that
>>>limits a cane user in new territory is the necessity of slowing down
while
>>>detecting an object and moving around it.
>>
>>>
>>
>>>I place a list of things below that are within reach of today's
technology
>>>but I'm not sure if they are within the practical financial boundaries of

>>>a
>>>marketable device.
>>
>>>I thought those of you who think about matters like this might weigh in
on
>>>this and let me know.
>>
>>>Here is my list of things the device could do.
>>
>>>1.  avoiding objects-people and other interferences to the path of travel
>>
>>>2.  Detecting objects as high as one's head.
>>
>>>3.  identifying curbs and ramps
>>
>>>4.  recognizing intersecting sidewalks or other paths of travel
>>
>>>5.  finding doors
>>
>>>6.  locating up or down stairs
>>
>>>7.  locating elevators
>>
>>>8.  locating elevator buttons
>>
>>>9.  finding restrooms
>>
>>>10.  finding empty urinals
>>
>>>11.  finding empty seats on busses or in conference rooms
>>
>>>12.  returning to an already found seat
>>
>>>13.  finding a bus stop
>>
>>>14.  finding a specific address
>>
>>>15.  following an indicated person
>>
>>>
>>
>>>I'd be interested in feedback on this idea.
>>
>>>If you wish to write me off-line, do so at
>>
>>>mbullis at imagemd.org
>>
>>>Thanks for your thinking.
>>
>>>Mike
>>
>>>
>>
>>>
>>
>>>
>>
>>>
>>
>>>_______________________________________________
>>>nfb-talk mailing list
>>>nfb-talk at nfbnet.org
>>>http://nfbnet.org/mailman/listinfo/nfb-talk_nfbnet.org
>>>To unsubscribe, change your list options or get your account info for
>> nfb-talk:
>>>http://nfbnet.org/mailman/options/nfb-talk_nfbnet.org/steve.jacobson%40vi
s
>i
>> .com
>>
>>
>>
>>
>>
>> _______________________________________________
>> nfb-talk mailing list
>> nfb-talk at nfbnet.org
>> http://nfbnet.org/mailman/listinfo/nfb-talk_nfbnet.org
>> To unsubscribe, change your list options or get your account info for
>> nfb-talk:
>>
>http://nfbnet.org/mailman/options/nfb-talk_nfbnet.org/glowhi%40centurylink.
n
>> et
>>
>>
>> _______________________________________________
>> nfb-talk mailing list
>> nfb-talk at nfbnet.org
>> http://nfbnet.org/mailman/listinfo/nfb-talk_nfbnet.org
>> To unsubscribe, change your list options or get your account info for 
>> nfb-talk:
>>
>http://nfbnet.org/mailman/options/nfb-talk_nfbnet.org/mistydbradley%40gmail
.
>com 

>_______________________________________________
>nfb-talk mailing list
>nfb-talk at nfbnet.org
>http://nfbnet.org/mailman/listinfo/nfb-talk_nfbnet.org
>To unsubscribe, change your list options or get your account info for
>nfb-talk:
>http://nfbnet.org/mailman/options/nfb-talk_nfbnet.org/bullis.michael%40gmai
l
>.com

>_______________________________________________
>nfb-talk mailing list
>nfb-talk at nfbnet.org
>http://nfbnet.org/mailman/listinfo/nfb-talk_nfbnet.org
>To unsubscribe, change your list options or get your account info for
nfb-talk:
>http://nfbnet.org/mailman/options/nfb-talk_nfbnet.org/steve.jacobson%40visi
.com

_______________________________________________
nfb-talk mailing list
nfb-talk at nfbnet.org
http://nfbnet.org/mailman/listinfo/nfb-talk_nfbnet.org
To unsubscribe, change your list options or get your account info for
nfb-talk:
http://nfbnet.org/mailman/options/nfb-talk_nfbnet.org/bullis.michael%40gmail
.com