PDA

View Full Version : Achieving human-like android movement



Dwight Meglan
07-21-1996, 10:48 PM
Hi Caroline,

>Caroline Nachman wrote:
>part. but have been educated to believe that e..g the human wrist, hand,
>knee, etc. and the human balance sense were so complex that in practice, we
>would never achieve a robot that moved like a human. is this still correct?

In regards to your inquiry about when or if we will achieve human-like
movement using machines, I think that in some regards soon, in others maybe
never. I know this your typical political waffle answer, but allow me to
explain.

We should be able to achieve very reasonable locomotion approximation
within the next 10 years. Now, this means that someone will build machine
that can walk and run about with sufficient detail for an observer at a
distance not able to see what is under a covering of the machine to think
that they are seeing a person. This is sort of like the china room test for
artificial intelligence to be truly "intelligent" -- no flames here ;-) , I
am not a computer science/AI person so my terminology is undoubtly naive.
Note also that I am not saying that the device will be a standalone entity.
I doubt that power units will be compact enough for such a device to move
unlinked to a power source. It can probably contain all the control
computers however.

I am resonably sure of this because of my experience with computer
synthesis of locomotion. There are a number of people who have done a good
deal of fundamental work on the mathematics and implementation of
locomotion simulation (my thesis was on multi-body dynamic simulations of
locomotion so I used to study this a lot-- I'm a bit out of date now
though. I had move on and get a job a few years back since locomoting
machines doesn't exactly pay bills ;-)

Also, I am not referring to different levels of scale in the above. What I
mean is that the above device would not have fulll articulated wrists or
fingers or all the joints of the foot for that matter. It would do good
gait period. Once you start adding all the other joints it starts to get
real complicated and begs the question of why do this at all. Depending
upon the design goal, there may be more optimal solutions than just
reproducing the human form in high detail. To produce such an accurate
rendition of a human, it will take a while-- say 50 years as a wild guess.
It could be a lot less but there has to be someone throwing alot money at
the problem. Sort of an Apollo program for android development.

I for one don't see that coming any time soon. Realistically, I think we
will see human-like walking machines in 10 years, though I can't tell why
they will exist. Probably, they will come about through academic inquiry
more than any practical application. There any number of walking machines
in existence now, especially in Japan, where for a while there it seemed
that any self-respecting, university-based controls research group had to
build a walking machine (keeping up with the Jones' ;-). But, these will
not be confused with a human-- they are quite crude in their movements,
even though some have wonderfully ingenious mechanisms and controls
algorithms embodied in them.

Sorry for the not so hopeful predictions, but unless androids becomes the
next means for some nation to achieve the image of world leadership, we'll
just have to bide our time.

--dwight



~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~
Dwight Meglan, PhD | Developers of complete surgery simulation
Engineering Director | training systems and surgery simulation
High Techsplanations, Inc. | creation software tools
6001 Montrose Rd., Suite 902 |
Rockville, MD 20852-4874 | 301 984 3706 x238 301 984 2104 : FAX
dwight@ht.com | http://www.ht.com