[GSoC 2024] RMF Obstacle Detection Pipeline (#445)

Posted by @Yadunund:

Creating this discussion board to discuss the GSoC 2024 project. Interested candidates may any questions they have about the project.

Please do not use this discussion to indicate your interest to work on this project.

Useful references:

Posted by @ErykHalicki:

Hey @Yadunund, I see that most of the proposed work is updates to the overall pipeline, making it more general and useable. rmf_human_detector has piqued my interest, but it seems it’s not in the current scope. I saw in Realsense D400 series Cameras support? · Issue #7 · open-rmf/rmf_obstacle · GitHub mention of updating the detection system to be able to handle more than just humans, as well as obtain 3d pose data from a depth stream. I think the idea is super interesting and was wondering if that might also be explored?

Posted by @Yadunund:

Hi @ErykHalicki
Thanks for your interest in the project! That’s a good suggestion. I’ve added a new line item Generalize rmf_human_detector to detect other classes of objects.
Other suggestions are also welcome. Once we move things to become plugin-able, I’m hoping we can start supporting a variety of detectors.

Posted by @suchetanrs:

Hey @Yadunund
When I ran the lane blocker node, it initially said Message Filter dropping message: frame '' at time 0.000 for reason 'the frame id of the message is empty'
I later realised that the obstacles message header was not filled before publishing in LaserscanDetector.cpp. Is this a bug or was I doing something wrong? Can this issue be raised with a PR?

Posted by @suchetanrs:

Hi, I have detailed this further in this issue.

Posted by @Yadunund:

In addition to what I mentioned in the ticket, feel free to include whatever else needs to be fixed in your proposal.