I would like to build 2D EKF-SLAM in openGL. I've implemented the entire virtual environment in which there is a robot that moves in 2D and there are some landmarks(feature-based map). I have the motion and observation models. Also, I've implemented the sensors with Gaussian noise. Now, I would like to use MRPT to build SLAM. At this point, I don't want to use data association that is the robot moves and detects its pose and landmarks and discard the previous data which means I only concern with the current state vector. My question is Is it possible to build SLAM without data association? Please suggest me some articles so that I can enrich my background about only this issue.
$\begingroup$
$\endgroup$
2
-
2$\begingroup$ Data association is merely a technique to cope with landmarks that look identical. If every feature or landmark which the robot can see is unique, you don't need to concern yourself with this problem. $\endgroup$– Daniel EbertsCommented May 3, 2014 at 10:10
-
$\begingroup$ @DanielEberts, deserve to be an answer. Thank you. $\endgroup$– CroCoCommented May 4, 2014 at 1:45
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
1
If you build a robot in simulator only, then the data association is not a problem. The true landmarks would the robot see at time $t$ can be given by hand. But if you equip a small vehicle with a laser scanner or depth camera and willing to run a real SLAM job, then data association is a really big problem you have to solve, especially for a grid-based map. By the way, it seems that you can implement a landmark based SLAM by SURF feature, but in fact, it requires a lot of computing resources - especially for EKF SLAM and it can make your job fail.
-
$\begingroup$ Hi Steve, and welcome to Robotics! Could you elaborate more on why data association is such a big problem? Also, how to deal with it? $\endgroup$– mactroCommented Feb 6, 2017 at 8:14