Lots of work was done but not a lot of progress this week involving the Jetsons(s). Mae and I worked with Professor Mukherjee for several days to get the Jetson Xavier flashed with Jetpack 5.0, which required us to set up the Xavier from anew, including registering it as a device to access internet through Ethernet in the Hamerschlag lab. Once updated to Jetpack 5.0, I installed pytorch and discovered that CUDA was indeed available, which was great. However, once I went on to install the other packages needed for our system I soon discovered that pyrealsense2 was not available as a pip download on python3.8, which was what was included on the Jetpack 5.0. So, Mae and I then spent the next day trying to downgrade the Xavier to python3.7 which would have worked with the pip install but did not allow us CUDA availability once we went through the steps of installing pytorch once more on python3.7. Finally, we tried building the pyrealsense2 library using cmake but were unable to due to how we messed with the python3 bindings. We then attempted to reflash the SD card with Jetpack 4 to attempt to roll back to the version that would have python3.6 natively and thus work with the pip install for pyrealsense2 but discovered that in the process of updating the Xavier to Jetpack 5 we overwrote on board memory with a new image as well (QSPI), and so once we inserted the new SD with Jetpack 4 on it to the Xavier, it wouldn’t boot since the QSPI version differed from the Jetpack version. We then started working with the Nano instead in hopes of it having Jetpack 4 and CUDA availability, but ran into issues again with building the pyrealsense2 library, especially with the memory constraints of the Nano causing a high latency and forcing installs to stall. As we’ve hit a dead end here, I will be reflashing the SD card with Jetpack 5 so we can return to the Xavier, then attempt to build the pyrealsense2 library once more without messed up python3 routings.
I also worked with Meghana this week to test out a 18V to 12V step down converter that I ordered with our Arduino and a new motor driver. This worked beautifully, and we will work on fine tuning the speed to drive our motor at so that the bottles are not so forcefully propelled into our intake storage.
This coming week I will be attempting to start over with Jetpack 5 as aforementioned and attempt to build the pyrealsense2 library properly. This week was a significant roadblock but at least we were able to ensure that CUDA would be available with pytorch on Jetpack 5 with our Xavier.
0 Comments