Pose graph optimization datasets:

 

These input files are created to test graph optimization algorithms as LAGO, g2o, and TORO.

The datasets describe planar SLAM problem instances. The files will be provided in TORO format (which is currently the same used to feed LAGO) and in g2o format. Regardless the format, these input files contain two "sections":

  • VERTICES (lines starting with "VERTEX2" in TORO, or "VERTEX_SE2" in g2o): description of the vertices of the pose graph. Each line follows the format:   " ID   x   y   th ", where “ID” is a unique index assigned to the corresponding pose, and “x”, “y”, “th” describe the initial guess (usually obtained from odometry) of the pose. “x” and “y” are the initial guess of node's planar position, while “th” is the initial guess for node's orientation (in radians);

  • EDGES (lines starting with "EDGE2" in TORO, or "EDGE_SE2" in g2o): description of the edges of the pose graph. Each line follows the format:   " IDout   IDin   dx   dy   dth   I11   I12  I22  I33  I13  I23 ", and describes an edge going from node with index “IDout” to node with index “IDin”. The edge describes a measurement of the pose of node “IDin” in the reference frame of node “IDout”. The relative (planar) pose is coded in the elements “dx”, “dy”, “dth”. The remaining elements describe the uncertainty on the relative pose measurement: the six elements completely define the (symmetric) 3x3 Information matrix I = [ I11  I12  I13;  I12  I22  I23;  I13  I23  I33  ] . In g2o the ordering of the elements in the EDGES description is slightly different from TORO and LAGO, and becomes: " IDout   IDin   dx   dy   dth   I11   I12  I13  I22  I23  I33 ". Moreover, for letting g2o optimize the same cost function as TORO, the corresponding information matrices are chosen as described in this document.

 

INTEL:

This dataset describes the pose graph obtained by processing the raw measurements from wheel odometry and laser range finder, acquired at the Intel Research Lab in Seattle (raw data provided by Dirk Hähnel and available online at [http://ais.informatik.uni-freiburg.de/slamevaluation])

 intel lago_map

 

 

 

                         Download input file in TORO format

 

 

 

 

 

FR079:

This dataset is built from raw data acquired at the Freiburg Building (the relative pose measurements are also available online at [http://ais.informatik.uni-freiburg.de/slamevaluation])

fr079

                        

                       Download input file in TORO format           

 

 

 

 

CSAIL:

This dataset is built from raw data acquired at the MIT CSAIL building (the relative pose measurements are also available online at [http://ais.informatik.uni-freiburg.de/slamevaluation])

csail

                        

                     Download input file in TORO format 

 

 

 

 

 

 

FRH:

This dataset is built from raw data acquired at the Freiburg University Hospital (the relative pose measurements are also available online at [http://ais.informatik.uni-freiburg.de/slamevaluation])

frClinic

                        

 

                     Download input file in TORO format            

 

 

 

 

 

MITb:

This dataset describes the pose graph obtained by processing the raw measurements from wheel odometry and laser range finder, acquired at the MIT Killian Court (raw data provided is available online at [http://ais.informatik.uni-freiburg.de/slamevaluation])

 MIT eg2o

                        

 

                         Download input file in TORO format

 

                         Download input file in g2o format

 

 

 

 

M3500:

Manhattan world with 3500 nodes, created by Olson et al. [E. Olson, J.J. Leonard, S.J. Teller, "Fast Iterative Alignment of Pose Graphs with Poor Initial Estimates", 2006]

 M3500 eg2o

                        

 

                         Download input file in TORO format

 

                         Download input file in g2o format

 

 

 

 

M3500a:

Variant of the M3500 dataset. Extra Gaussian noise with standard deviation 0.1rad is added to the relative orientation measurements

 M3500a eg2o

                        

 

                          Download input file in TORO format

 

                          Download input file in g2o format

 

 

 

M3500b:

Variant of the M3500 dataset. Extra Gaussian noise with standard deviation 0.2rad is added to the relative orientation measurements

 M3500b eg2o

                        

 

                         Download input file in TORO format

 

                         Download input file in g2o format

                                             

   

                        

     

M3500c:

Variant of the M3500 dataset. Extra Gaussian noise with standard deviation 0.3rad is added to the relative orientation measurements

 M3500c eg2o

                        

 

                         Download input file in TORO format

 

                         Download input file in g2o format

  

   

 

 

M10000:

Manhattan world with 10000 nodes

 M10000

                        

 

                         Download input file in TORO format