Revision history [back]

I'm not sure if I fully grasped your question, but the point of not setting motion model noise to zero is that the simulated robots mimics faithfully enough the robot so it can serves as test-bench for your algorithms.

I'm Ah, ok. Yes, it's an interesting information what you are providing here. I will try to contribute with this answerm as my previous comment was not sure if I fully grasped your question, but the point of not setting motion very precise, admittedly. There're two different things here:

• Gmapping odometry model noise to (srr, srt, str, stt) reflects the confidence the slam algorithm can have on odometry readings, and so cannot be zero unless the odometry is that perfect, what only can happen on simulation.
• Simulated (Gazebo) odometry model noise should not be zero, (except, maybe, for particular experiments) so the simulated robots mimics faithfully enough the robot robot, with their inaccuracies, so it can serves as test-bench for your algorithms.
  algorithms. But as explained in this issue, Kobuki (and so TB2) is perfect on Gazebo, what should be fixed.

Ah, ok. Yes, it's an interesting information what you are providing here. I will try to contribute with this answerm as my previous comment was not very precise, admittedly. There're two different things here:

• Gmapping odometry model noise (srr, srt, str, stt) reflects the confidence the slam algorithm can have on odometry readings, and so cannot be zero unless the odometry is perfect, what only can happen on simulation.simulation. I supposed that, when gmapping corrects current pose based on scan matching, it should have some kind of "inertia" to keep in the same position, the bigger the more accurate is the odometry, and don't move at all if the odometry is perfect. But I noticed that that's not the case, that even with zero srr, srt, str and stt the robot can jump, and thanks to your question I see now that the reason can be that minimumScore parameter is 0, by default.
• Simulated (Gazebo) odometry model noise should not be zero, (except, maybe, for particular experiments) so the simulated robots mimics faithfully enough the robot, with their inaccuracies, so it can serves as test-bench for your algorithms. But as explained in this issue, Kobuki (and so TB2) is perfect on Gazebo, what should be fixed.

Ah, ok. Yes, it's an interesting information what you are providing here. I will try to contribute with this answerm as my previous comment was not very precise, admittedly. There're two different things here:

• Gmapping odometry model noise (srr, srt, str, stt) reflects the confidence the slam algorithm can have on odometry readings, and so cannot be zero unless the odometry is perfect, what only can happen on simulation. I supposed that, when gmapping corrects current pose based on scan matching, it should have some kind of "inertia" to keep in the same position, the bigger the more accurate is the odometry, and don't move at all if the odometry is perfect. But I noticed that that's not the case, that even with zero srr, srt, str and stt the robot can jump, and thanks to your question I see now that the reason can be that minimumScore parameter is 0, by default.
• Simulated (Gazebo) odometry model noise should not be zero, (except, maybe, for particular experiments) so the simulated robots mimics faithfully enough the robot, with their inaccuracies, so it can serves as test-bench for your algorithms. But as explained in this issue, Kobuki (and so TB2) is perfect on Gazebo, what should be fixed.

Ah, ok. Yes, it's an interesting information what you are providing here. I will try to contribute with this answerm as my previous comment was not very precise, admittedly. There're two different things here:

• Gmapping odometry model noise (srr, srt, str, stt) reflects the confidence the slam algorithm can have on odometry readings, and so cannot be zero unless the odometry is perfect, what only can happen on simulation. I supposed that, when gmapping corrects current pose based on scan matching, it should have some kind of "inertia" to keep in the same position, the bigger the more accurate is the odometry, and don't move at all if the odometry is perfect. But I noticed that that's not the case, that even with zero srr, srt, str and stt the robot can jump, and thanks to your question I see now that the reason can be that minimumScore parameter is 0, by default.
• Simulated (Gazebo) odometry model noise should not be zero, (except, maybe, for particular experiments) so the simulated robots mimics faithfully enough the robot, with their inaccuracies, so it can serves as test-bench for your algorithms. But as explained in this issue, Kobuki (and so TB2) is perfect on Gazebo, what should be fixed.

mmm... I pretended to write only some lines! ^_^ Sorry for been so verbose.