This week I mostly took it easy after spending ~30hrs on capstone last week. I was able to collect a bunch of water bottles and soda cans for future testing, however.
While I did work on some code for using a trt model instead, a couple hours wasn’t long enough to get it working. I was able to run the model on dummy data however using trtexec and following the procedure NVIDIA used to generate their benchmarks here, and the trt model runs at nearly 450 fps when clocking the nano. Certainly above what we need and much more than the capabilities of our camera which runs at 90fps.
As completing our robot and code for the other subsystems are most at risk right now, and since integration cannot happen without them, next week I will be pivoting to help out with build and motion integration before worrying about integrating the ML subsystem (and more generally, the Jetson Orin Nano) with the path finding subsystem (and more generally, the Raspberry Pi).
I am on track as per our schedule.
0 Comments