Custom model slam mapping issues

edit

Thanks @tryan.

I did not realise slam does not work well in an environment of long straight corridors. Infact, I had assumed the opposite. Those lines are actually the kerbs of sidewalks of a simulated city as detected by the laser(s) or lidar on my robot model. As the environment is a city, the lidar does detect buildings and other infrastructure. So, I wonder, is it possible to combine multiple sensor data to create better slam data ? My thinking is perhaps combining odometry, laser and lidar would provide better data for slam.

Yes, the main problem with such an environment arises when using only a short-range lidar. In a hallway, the scans from the lidar at different points will look very similar (two parallel walls), making it appear as though the robot has not moved.

As you suggest, if you are able to collect more data from the environment (longer range, more sensors, etc.), the SLAM algorithm will be better able to determine how far the robot has moved. In general, the more data, the better. You do have to make sure it's quality data, though. If your odometry is way off for example, it could hurt more than help.

Another option is to make the scans look different by adding objects to the environment, so the SLAM algorithm has more discernible features to use.

I'm going to try finding and adding some kind of unique patterns to the kerbs. I'm thinking qr codes to be read by camera on my model. BUT, how can I integrate such camera data, or lidar etc, into the SLAM algorithm ? This is the part I can't figure out yet. Gmapping needs a minimum of laser data. How can I introduce extra data from other sensors ?

Before adding more complicated sensors, I recommend you confirm what the problem is. If you increase the range of your current lidar(s) or add some small objects along the road and the mapping works, then you'll know that you need more feature-rich data to be successful. If it doesn't help, then you may still have a different problem with your current setup.

I'm not sure about using landmarks in gmapping; I've only used it with odometry and a lidar. A different package may be easier to integrate, but that's beyond the scope of this question.

Hi @tryan. Thank you for your input. I followed your advice and placed objects all over the streets of the virtual city world and I found using Hector mapping the robot model did not loose orientation so much. I managed to complete a far better map than all my previous attempts. It still lost orientation at some point but I had saved the map when it was almost complete. I then manually editted the saved map to remove the the objects etc. However when I did the same with gmapping, the map was far inferior.

It's good to hear there was improvement! Would you mind adding a picture of your newer map for reference?