Position determination: Difference between revisions

From OLPC
Jump to navigation Jump to search
(tweaks, typos)
m (Reverted edits by 198.83.124.250 (Talk) to last version by Leejc)
 
(9 intermediate revisions by 7 users not shown)
Line 2: Line 2:


Approaches:
Approaches:
*optical tracking
*Optical tracking
*sound-delay distance measurement
*Sound-delay distance measurement
*''what else...''
*''what else...''
** How about [[Photogrammetric Astronavigation]] (using photos of the sky to calculate geographical location)?
Different approaches are very much better at different things, so several seem likely to be implemented.
Different approaches are very much better at different things, so several seem likely to be implemented.
If we architect it right, the estimates from multiple approaches might be combined to yield more accurate estimates.
If we architect it right, the estimates from multiple approaches might be combined to yield more accurate estimates.
Line 10: Line 11:
===Getting position from sound-delay distance measurements===
===Getting position from sound-delay distance measurements===


One approach is to derive localization from pairwise distance measurements made using sound delay (trilateration). An XO initiates or observes an audible speaker event and announces it on wireless. Other XO's hear and recognize the event, and derive distance from speed-of-sound delay.
One approach is to derive localization from pairwise distance measurements made using sound delay (trilateration). An XO initiates or observes an audible speaker or mic event and announces it on wireless. Other XO's hear and recognize the event, and derive distance from speed-of-sound delay.


Robustly assembling such distance measurements into location reports has been a research problem for almost a decade. Dealing with noisy measurements and positional uncertainty are the big issues. The sound measurements are multipath and thus their distributions are multimodal. If XO's are moving rapidly with respect to the measurement rate, then predictive motion correction is needed.
Robustly assembling such distance measurements into location reports has been a research problem for almost a decade. Dealing with noisy measurements and positional uncertainty are the big issues. The sound measurements are multipath and thus their distributions are multimodal. If XO's are moving rapidly with respect to the measurement rate, then predictive motion correction is needed.
Line 17: Line 18:


Resources:
Resources:
*[[Acoustic Tape Measure]]
*[http://cricket.csail.mit.edu/ Cricket] is perhaps the most heavily exercised such system. [http://cricket.csail.mit.edu/V1Exp.pdf This paper](PDF) describes some of the lessons learned. We might transliterate some algorithmic code from the cricket v2 java source.
*[http://cricket.csail.mit.edu/ Cricket] is perhaps the most heavily exercised such system. [http://cricket.csail.mit.edu/V1Exp.pdf This paper](PDF) describes some of the lessons learned. We might transliterate some algorithmic code from the cricket v2 java source.
*[http://citeseer.ist.psu.edu/moore04robust.html Robust Distributed Network Localization with Noisy Range Measurements (2004)] is addressing the same(?) problem. But note the unfortunate "degree 10" requirement, though perhaps in practice it might not be a problem. Algorithm pseudocode included.
*[http://citeseer.ist.psu.edu/moore04robust.html Robust Distributed Network Localization with Noisy Range Measurements (2004)] is addressing the same(?) problem. But note the unfortunate "degree 10" requirement, though perhaps in practice it might not be a problem. Algorithm pseudocode included.
*[http://citeseer.ist.psu.edu/capkun01gpsfree.html GPS-free positioning in mobile ad hoc networks (2001)]
*[http://citeseer.ist.psu.edu/capkun01gpsfree.html GPS-free positioning in mobile ad hoc networks (2001)]
*[http://www.ee.usyd.edu.au/~guoqiang/papers/Mao06Wireless.pdf Wireless Sensor Network Localization Techniques] - a recent review article. Limited usefulness.
*[http://www.ee.usyd.edu.au/~guoqiang/papers/Mao06Wireless.pdf Wireless Sensor Network Localization Techniques] - a recent review article. Limited usefulness.

===Getting position from WiFi signal strength===
Getting position from WiFi signal strength does not seem promising.
:Quozl, olpc_wad, et al, report (#olpc, 2007-09-20) tests showing only lightly correlation between distance and wireless signal strength, especially not at close range. The only good correlations were on a flat plain with no obstructions with distances over 50m and up to 1.6km. At WiFi frequencies, interference and reflections tend to make signal strength pretty unusable. I saw a presentation by Ember a couple of years ago where a researcher (Rob Poor) had measured the signal strength in a medium sized room at 1 in. intervals, and it was a wild patchwork with no discernable rhyme or reason. yes, that's what i've observed.


===Data fusion===
===Data fusion===

Latest revision as of 06:56, 17 December 2008

A mesh of nearby XO's should be able to roughly calculate their 2D and 3D relative positions.

Approaches:

  • Optical tracking
  • Sound-delay distance measurement
  • what else...

Different approaches are very much better at different things, so several seem likely to be implemented. If we architect it right, the estimates from multiple approaches might be combined to yield more accurate estimates.

Getting position from sound-delay distance measurements

One approach is to derive localization from pairwise distance measurements made using sound delay (trilateration). An XO initiates or observes an audible speaker or mic event and announces it on wireless. Other XO's hear and recognize the event, and derive distance from speed-of-sound delay.

Robustly assembling such distance measurements into location reports has been a research problem for almost a decade. Dealing with noisy measurements and positional uncertainty are the big issues. The sound measurements are multipath and thus their distributions are multimodal. If XO's are moving rapidly with respect to the measurement rate, then predictive motion correction is needed.

Apparently there is existing code for XO's to measure mutual distance using sound delay. (cjb, #olpc, 2007-04-17) <10% error in the room it was tried.

Resources:

Getting position from WiFi signal strength

Getting position from WiFi signal strength does not seem promising.

Quozl, olpc_wad, et al, report (#olpc, 2007-09-20) tests showing only lightly correlation between distance and wireless signal strength, especially not at close range. The only good correlations were on a flat plain with no obstructions with distances over 50m and up to 1.6km. At WiFi frequencies, interference and reflections tend to make signal strength pretty unusable. I saw a presentation by Ember a couple of years ago where a researcher (Rob Poor) had measured the signal strength in a medium sized room at 1 in. intervals, and it was a wild patchwork with no discernable rhyme or reason. yes, that's what i've observed.

Data fusion

If multiple approaches are used to estimate position, it would be nice to combine them in a way which increased accuracy. There is also the architectural challenge of sharing raw data and processed estimates among laptops.

Possible uses

  • Walking around as input.
  • Exposing social dynamics.