Question about G2O file about distributed-mapper

[itzsid]

Correct. You can directly use the odometry measurements as EDGE_SE3 since they are already noisy. VERTEX_SE3 provides an initialization for the poses and you can initialize it using odometry too (similar to how you explained). This algorithm should be robust to bad initialization as well as long as you don't have outlier measurements.

Let me know if it's still not clear.

[itzsid]

Hi @Quest2GM,

The edges in the synthetic dataset are simulated to be noisy to reflect the realistic conditions where odometry or communication measurements have errors. If you have perfect dx (without any noise), then you don't need to use this optimizer to find the most likely estimate.

Thanks!

[Quest2GM]

Right, I understand that; maybe I should provide a little more context.

My robot provides odometry measurements in a ROS topic of type nav_msgs/Odometry. This topic provides the pose calculated through odometry, as well as the covariance matrix of the pose. These odometry measurements already have Gaussian noise added to them.

When adding a VERTEX_SE3 to the g2o file, I just use the pose outputted in the above topic. When adding an EDGE_SE3, I simply calculate dx (and the other values dy, dz, etc.) by subtracting the previous pose from the current pose. In that sense, my dx value is exactly equal to the subtraction of the two corresponding VERTEX nodes.

Is this the way you would do it, or would you need to compute the VERTEX or EDGE in a different way?

Thanks again,
Quest2GM

[Quest2GM]

Perfect thanks!