S2025 Perception Merge

[lukasdumasius]

Updated perception stack with cones, pedestrians, sensor fusion bug fixes.

Some of our commit history reset merging with s2025. Please check #223 if you're interested in that.

[krishauser]

Some minor changes requested.

[krishauser]

The structure of the sensor fusion code is not conformant to GEMstack. You cannot be launching random ROS nodes. PR is not approved, sorry.

[krishauser]

Why is there a ROS node in GEMstack? The architecture doesnt make sense.

[LucasEby]

PointPillars requires specific package versions which have dependency issues with GEMstacks packages so it cannot be directly integrated into GEMstack. This is why we needed to create a docker container in order to use it. It was written as a ROS node so that it could communicate with GEMstack. I can't think of any other way to integrate PointPillars with GEMstack? Is there something I am unaware of?

[LucasEby]

I got rid of the YOLO node to make the code conformant to GEMstack. I cannot remove the PointPillars node because it needs to be run in a docker container due to the package dependency issues.

[LucasEby]

Before creating the docker container, I spoke to Rachel about the issues I was having with integrating the AI models directly into GEMstack due to the package dependency issues I kept running into. I mentioned that the models required specific package versions that conflicted with GEMstack and prevented direct integration (unless I could downgrade some of the vehicle’s packages). I asked her if it would be possible to downgrade package versions on the vehicle to directly integrate the model(s) I had been trying to directly integrate. I also asked if I could set up a docker container and run a node inside of it as a worst case scenario to integrate another model for the sensor fusion work. She directed me to Jiaming to get my software questions answered. I messaged Jiaming on slack and he said that I could not downgrade the torch version (something which I needed to do for all of the models we looked into) but did say that he didn’t see any issue with setting up a docker container on the vehicle and additionally suggested that I could maybe try use a virtual environment if I couldn’t directly integrate the model(s) into GEMstack. I did not speak to you about whether or not a docker container would be okay because I had already spoken to them about it and they didn’t have any issues with it. I can provide the Jiaming slack messages if needed.

I was not able to get the virtual environment to work due to reasons that I have documented in the shared google document but I was able to create a docker container and set up PointPillars inside of it. I had to set up PointPillars inside of its own ROS node so that it could communicate outside of the docker container with the running Combined Detector which was fully integrated into GEMstack with our launch file.

I would really appreciate it if you could accept our pull request.

[sonarqubecloud]

Quality Gate failed

Failed conditions
11 Security Hotspots
4.9% Duplication on New Code (required ≤ 3%)
C Reliability Rating on New Code (required ≥ A)

See analysis details on SonarQube Cloud

Catch issues before they fail your Quality Gate with our IDE extension SonarQube for IDE