# GraphSLAM data accumulation problem

I try to implement GraphSLAM from this tutorial, The GraphSLAM Algorithm with Applications to Large-Scale Mapping of Urban Structures. I have some doubt while studying this paper. I hope people who research on same field may solve my query.

As I don't know how to input notation in StackExchange editor I have uploaded a picture with my queries: I have some questions, which are written in the images, at line numbers 11, 14, 19:

• What is the dimension of this noise covariance matrix? If it is 3*3 then the dimension of the matrix is mismatched at line numbers 18 and 19
• How to calculate $µ(j,x)$ and $µ(j,y)$?
• What is the value of $z_t^i$?

I have some other doubts:

1. What is the need to use two for loop one line number 10 and another one at line number 12?
2. Is that use to change the value of $Q(t)$ in each iteration? How can all measurements and observed features be different from each other?

I am using this dataset, UTIAS Multi-Robot Cooperative Localization and Mapping Dataset, where there is a sensor measurement data set and odometry measurement dataset. They are two separate files. Their timestamp is also different. Their dataset is like that after 100 odometer data there is a data which observed a features. Then how could I relate these datasets with this algorithm?

## 1 Answer

No:11 -> it is 3x3

No:19 -> $i_{th}$ measurement of $z_t$

What is the need to use two forloop one lineNo:10 and another one LineNo:1 -> Each feature $z_t$ have multiple observations $z_t^i$. First loop is for each feature and the other one for each observation of that feature.

• The OP also asks for the value of z_i^t – Greenonline Jul 14 '18 at 9:45
• How to calculate µ(j,x) and µ(j,y)? – Encipher Jul 14 '18 at 14:03
• To be honest, if your purpose is to learn graph based slam by implementing it yourself, you have got the wrong paper. Have a look at the Wolfram Burgard's classes on mobile robotics where they provide the example matlab source code of a graph slam. Running and analyzing the few lines of their code is much better than looking at the paper and trying to implement from that. – C.O Park Jul 14 '18 at 20:38