tysik/obstacle_detector

Offset in detected obstacle position

huckl3b3rry87 opened this issue · 3 comments

huckl3b3rry87 commented

@tysik first of all, I wanted to thank you for making this package, and mention that I am using it in a project that I am working on called MAVs. Your package is documented some here for our project.

We are running into a couple things that we are having trouble explaining and I am wondering if maybe you know what may be causing them or how to explain them.

First of all, here are some results that we are getting when we run demoA

test

Questions:

  1. There is a dark red circle that is offset from the green circle(filled with light red). Can you please explain what these different circles are and why they are offset? I thought that it might be radius_enlargment but it is offset, so I am not sure.

  2. The light blue points are the point cloud that gets fed into the algorithm. For the top right obstacle, it can be seen that the predicted obstacle does not encompass all of the blue points. Is this expected? I would think that it would predict that an obstacle would be there.

  3. It looks like the predicted obstacles are always towards the LiDAR , is this because the obstacle_extractor node fits a line to the visible data and then a circle over this line?

Thank you very much!

huckl3b3rry87 commented

@tysik I am still struggling to figure this out, I have simplified the problem a bit where the vehicle is just stationary and I see a similar issue. Are there parameters that I can tune so that it is more sensitive to lidar points?

issue
Any help would be greatly appreciated!!

Thanks!

tysik commented

Hmm, I cannot reproduce this bug because I do not work on the project anymore so it's hard to tell what may be the cause. Are you sure your points are ordered in an angular manner (clockwise or c-clokwise)?

huckl3b3rry87 commented

@tysik sorry for the delay, unfortunately, I am still running into this issue. I am not sure if the points are ordered in an angular manner. I am using a LiDAR model in gazebo that built in this demo
http://gazebosim.org/tutorials?tut=guided_i1

My launch file for your package is here
https://github.com/JuliaMPC/MAVs/blob/develop/ros/src/computing/perception/detection/obstacle_detection/launch/obstacle_detector.launch

I have 3D data, but I convert to PointCloud1 with this package
http://wiki.ros.org/point_cloud_converter

I know that you said your package does not work for 3D, but I am just using a 3D sensor and squishing the data (I think..) to 2D.

I know that you are no longer working on the project, but any help would be hugely appreciated:)