Timeline for How do SLAM algorithms handle a changing environment?
Current License: CC BY-SA 3.0
10 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 12, 2013 at 16:26 | comment | added | Josh Vander Hook | let us continue this discussion in chat | |
| Aug 12, 2013 at 14:50 | comment | added | anaximander | Yeah, I have an application in mind, but a lot of the stuff I'm reading only covers it tangentially because it's not strictly a core SLAM concern. Thanks for the help, anyway. I'm sure I'll be back with more questions as I dig through the reading list! | |
| Aug 12, 2013 at 14:39 | comment | added | Josh Vander Hook | SM is one way to handle this problem. As I said, good luck. SLAM is enormous. When reading lit, it's sometimes better to start with an application in mind, rather than trying to tackle the whole thing (and all tradeoffs) at once. I.e., I want to build a robot that does this, which, oh, btw, requires precise localization without GPS / semantic markers. | |
| Aug 12, 2013 at 14:32 | comment | added | anaximander | That's what I'm doing; the more I read, the longer my list of things to read gets! This was the first of the many questions my reading has raised. Apologies for extending the question somewhat in the comments; I've not been getting much response, and SLAM is such a huge topic... you seemed to know what you're talking about so I was hoping I could get a few pointers. I have semantic mapping papers on my list, but there's a few others I need to read as a foundation for that... Anyways, I guess it's back to the reading for now. | |
| Aug 12, 2013 at 14:28 | comment | added | Josh Vander Hook | Review. Literature. Such a problem arises from A) Stuff moving. B) robot moving. C) Robot getting lost. D) incorrect maps. Each of those 4 things has 4 different correct solutions. If you don't handle all 4, you will build completely incorrect maps. This is why SLAM is not solved, and still hard. Read Up and come back with specific questions related to what you have learned. | |
| Aug 12, 2013 at 14:25 | comment | added | anaximander | Yeah, the idea here is to sidestep the problem of semantics by simply highlighting "this is an area where I keep seeing things that weren't there before, or ceasing to see things that were there". Chances are, this area is some kind of transition - door, box, window, occluded corner. Which of those we're dealing with is less important; this is purely for collision avoidance, so we want to avoid all of these areas. Likewise, we don't particularly need to know which object is which; only that there's a significant change in the overall amount of space being occupied by Stuff. | |
| Aug 12, 2013 at 14:23 | comment | added | Josh Vander Hook | Way harder. The only way to do this correctly is to uniquely identify objects. Like, put a barcode on them. Otherwise object A may have moved to location B, or maybe A and B swapped, etc. Read up on semantic mapping. You need an algorithm which can "recognize" that moving things are actually "doors" which should be attached to "walls" but only when I'm "inside" but what does "inside" mean to a robot anyway? I think you should read more and report back. | |
| Aug 12, 2013 at 14:20 | comment | added | anaximander | Thanks! Do you know of any techniques that track where objects were "acquired" and "lost"? I'm looking at applications of a subset of SLAM-type algorithms and one area of interest is the identification of "transition" areas like doors and occluded corners where objects might appear from. This application turns the usual fading probability metric on its head - instead of having objects "fade" when out of sight, unobserved areas slowly increment their value to denote that we don't know what's here because we haven't looked recently, so we should be cautious when moving into this space. | |
| Aug 12, 2013 at 14:13 | history | edited | Josh Vander Hook | CC BY-SA 3.0 |
added 387 characters in body
|
| Aug 12, 2013 at 14:08 | history | answered | Josh Vander Hook | CC BY-SA 3.0 |