I'd like to build a 3D map of our lab (a number of rooms and hallways and cubicles, etc) using a Kinect and/or Turtlebot.
I've looked into a number of methods (right now I'm looking at possibly using RGBDSLAM), but I'd like to know if there are any typical methods people use.
Is RGBDSLAM a good choice for this? Are there other standard methods people use? What are their pros and cons?
(Related side-note: I tried using a Turtlebot with gmapping to build a 2D map of one section of the office, and it came out horribly. It also refused to work properly with any sort of loop closures involving hallways, I think largely because of the Turtlebot's horrible odometry.)
Edit: I Heart Robotics provided some very useful information, but if anyone has more information to offer, I'd very much appreciate it! :)
answered 2012-03-23 11:15:37 -0500I Heart Robotics
This is an area of open research, with several research groups working in slightly different directions. So, I am not sure there is a consensus on what is the best approach.
http://openslam.org is a nice resource the provides code for some 2D and 3D SLAM systems.
Also, if you are looking for papers you may want to look at the archives for the RGB-D Workshop at RSS. http://www.cs.washington.edu/ai/Mobile_Robotics/rgbd-workshop-2011/home.html
The biggest problems I see with large scale mapping is managing the massive amount of data coming from the sensor and limiting the growth of the map over time without discarding critical information. One solution is to simply wait for Moore's law to provide faster processors.
You can use RGBDSLAM for a large scale map (in the terms you mentioned), but it's not out-of-the-box-one-shot-mapping, but carefully-moving-the-camera/robot-mapping, most possibly with several attempts and some parameter tuning.
Asked: 2012-03-22 11:28:26 -0500
Seen: 500 times
Last updated: Mar 26 '12