Fun_People Archive
7 Mar
Translated seminar announcement


Date: Thu,  7 Mar 91 11:34:05 GMT-0800
To: Fun_People
Subject: Translated seminar announcement


Here's a translation of a recent seminar announcement:

               WHAT HE SAYS                   |        WHAT HE MEANS


        Autonomous Navigation Using           |      How to Get Around
      Evidential Reasoning Techniques         |    the Office by Yourself



Purposeful autonomous behavior requires  hav- | You must know where you are
ing  reliable  locative  information in order | in order to get around.
to help assess the desirability  of  pursuing |
particular  actions to  sense,  perceive,  or |
alter  one's  position  in  the  environment. |

Having  the  ability  to  ascertain  location | You'd like to be able to get
based  on  imperfect  perceptual   cues  that | around with your glasses off
are  extracted  from real-world contexts is a | or with your eyes closed.
critical requirement  for  truly   autonomous |
closed-loop    behavior.

                          While  some  of the |
well known non-perceptual locative techniques | You can walk around with your
such   as  dead  reckoning,  registration  of | eyes closed, but it's hard.
inertial  or  geo-stationary-satellite  loca- |
tion  data  with  an  environmental  map, and | Some people have tried
maintaining    a    global      cartesian-map | navigating with satellite 
representation  throughout  movement  activi- | uplinks, but this didn't
ties remain topics  of  investigations,  they | work well in our office.
require near perfect measurements  of  direc- | 
tion and   traversed  distances,  significant |
computation resources, and are limited by the |
range of  their  sensors.  

                           To  extend current |
autonomous  navigation  capabilities that are | We have found that we can get
limited by these requirements, increasing at- | around by feeling our way along
tention  has   focused   on developing robust | the wall of our office.
perceptual-based  inference  techniques  that |
derive location hypotheses  through  the  in- |
terpretation   of  perceptual   cues   within |
the   context   of environmental map descrip- |
tions.

Some recent work on perceptual-based  naviga- | You can feel your way around
tion schemes has  emphasized  the integration | better if you know the room
of distinct sensor modality data  within  the | (that's why we're using our
context  of an environmental model to  assess | office).
location.

          In this talk,  we  discuss  an  ap- |
proach to autonomous perceptual-based naviga- | Sometimes you have to flip
tion that uniquely  exploits evidential  rea- | a coin.
soning  (ER) techniques which embody interval |
probabilistic  methods  that  are  based   on |
work   of   Dempster   and  Shafer.

                                      We will |
describe how such ER techniques can  be  used | We describe how to get around
to fuse and interpret multisensor data within | our office on a skateboard
the context of a sparse spatial map of an of- | that we built ourselves.
fice  environment to infer location, and show |
the results of preliminary  experiments  with |
an autonomous mobile platform.

Translated by Paul Heckbert and Seth Teller






prev [=] prev © 1991 Peter Langston []