search

lardemua / atom

Calibration tools for multi-sensor, multi-modal robotic systems
GNU General Public License v3.0
245 stars 28 forks source link

Implement Intel RealSense L515 to prepare for data collection at INESC 2024 (zaubot) #903

Closed manuelgitgomes closed 2 months ago

manuelgitgomes commented 3 months ago

This is done! I implemented a launch file to launch this camera and instructions on how to use are on the README.md. There was only one caveat with the USB rules for this sensor, but the instructions to solve it are on the README.md as well.

image

miguelriemoliveira commented 3 months ago

Hi @manuelgitgomes ,

how about controlling the publication frequency and camera's resolution? Is it possible? What are the current values?

manuelgitgomes commented 3 months ago

Hello @miguelriemoliveira.

Answering your question, yes, it is changeable but only to predefined resolutions/fps combinations. The bad news is that I cannot find a list of these combinations. I tried searching around in librealsense but I could not find anything, but I can still explore more.

I reopened this issue because when trying to answer you questions I found a critical error when using the L515 camera. In my experience, when launching the camera various times in somewhat quick succession, my OS freezes completely.

I found a lot people with the same issue [1][2]. The solution for all involved not installing the binaries of the realsense-ros, but installing by source. This causes a problem, as it involves patching the kernel in order to support the cameras. The kernel I am currently running on my host is not supported by the librealsense. My Ubuntu 20.04 docker image shares kernel versions with the host, so it does not work.

I could try and create a docker image running this kernel version. I can have problems with this as I am uncertain if I need any library on host device to detect the camera, as I needed to change my udev rules on host as well.

We can also try in a computer that has a compatible kernel version.

miguelriemoliveira commented 3 months ago

Hum, these are not good new. The solution for a new docker with a compatible kernel is not good, because this is just one sensor from a large set of sensors. How about disregarding this sensor and using the asus camera instead as the rgbd sensor?

brunofavs commented 3 months ago

Might be a long shot but VM's don't share the kernel with the host. Given that you @manuelgitgomes have KVM sorted in your personal desktop it may be a option to experiment there.

That or we must not launch the camera repeatedly, which is annoying indeed :c

brunofavs commented 3 months ago

Also I noticed the issues you linked belong to different realsense sensors, I hope those problems don't carry over to the my T265 as I'm pretty certain we are using the same kernel.

manuelgitgomes commented 3 months ago

Might be a long shot but VM's don't share the kernel with the host. Given that you @manuelgitgomes have KVM sorted in your personal desktop it may be a option to experiment there.

Yes, but that is a long-shot, honestly. I don't know if it is a feasible solution.

How about disregarding this sensor and using the asus camera instead as the rgbd sensor?

That could be an option, but:

I hope those problems don't carry over to the my T265 as I'm pretty certain we are using the same kernel.

I am fairly certain these problems will carry out to the T265. I will continue trying and if I figure something out I'll say it.

brunofavs commented 3 months ago

Hey. I tested thoroughly the L515 on my computer and I couldn't replicate the crash. Run for long periods of time, restarted the device more than 10 times, all seemed fine.

I'm currently on this kernel 6.6.23-1-lts

miguelriemoliveira commented 3 months ago

That's good news.

miguelriemoliveira commented 3 months ago

I think your laptop is becoming the prime candidate to be the system which will be used to collect all data.

What are your specs?

brunofavs commented 3 months ago

My pc is the Asus ROG Strix G531GT and the main specs are: i7-9750h GTX 1650 16Gb Ram 3xUSB 3.2 Gen 1.

My main worry would be the battery life. I doubt it lasts even 1 hour with a full ROS system running full blast. I also have a RJ45 port to connect the LiDAR, so no extra adaptor is necessary.

I don't mind using mine but I reckon your pc @miguelriemoliveira is superior to mine.

miguelriemoliveira commented 3 months ago

I don't mind using mine but I reckon your pc @miguelriemoliveira is superior to mine.

right, but I will not be with you three most of the times. I am not even sure I will going to INESC in all the days of the tests.

So best to aim to use your laptop. Battery is not a problem as we can have the charger always connected, even with the agv moving around.

Tagging @Kazadhum for visibility...

brunofavs commented 3 months ago

Ok that settles it then. Do you have perhaps a USB hub we can borrow?

I only have 3 USB ports and we will certainly need more. I talked with @manuelgitgomes but his hub has USB-c input which I do not have in my laptop.

miguelriemoliveira commented 3 months ago

I have a hub, but generally those will cause problems. Lets test it.

@manuelgitgomes you still have my office's key? Can you go there and pick up a black usb hub that I have near the monitor?

Kazadhum commented 3 months ago

Hello @miguelriemoliveira @brunofavs and @manuelgitgomes !

I agree that @brunofavs's PC is most likely the best candidate, especially since it appears to run everything with no problems. I think I have a USB Hub somewhere if needed, but I never used it so I don't even know if it works...

manuelgitgomes commented 2 months ago

Data collection done, closing this.