System requirements

This is the supported system i.e. development and testing is done on these specs. Other systems may also work but are not tested.

Simulation host

NVIDIA GPU and CUDA drivers Strongly recommended
Memory: 16GB Recommended, 8GB should also work but not tested
Storage: 50GB free
Storage
Storage mostly needed for storing Docker images that may include deep learning model weights and map rasters. Current suite of images is quite unoptimized from a storage perspective and improvement could be done here.

Ubuntu 22.04 LTS (Jammy Jellyfish) Recommended
Debian
Most Debian derivatives may work with some modifications, but this guide assumes you are running Ubuntu Jammy.

Raspberry Pi 5
- Memory: 8GB
- Storage: 128GB SD card Recommended, 64GB should also work
- OS: Debian and its derivatives Recommended
GPU recommended
The goal is to make GISNav work without a GPU so it is developed and tested on a Raspberry Pi. However, you will likely get better results using e.g. an Nvidia Jetson Orin Nano board. There should be no hard dependency on an NVIDIA GPU, but PyTorch will most likely by default upload the model into main (CPU) memory if CUDA is not available.
Warning: 4GB of memory not enough
The deep learning model may be too large for the 4GB model. Attempts to make GISNav work on 4GB of memory have been made and may be made in the future.

Pixhawk FMUv4
Tested on NXP RDDRONE-FMUK66 (FMUv4)
See the HIL instructions here.
GPS 2 serial port
You may want to look for Pixhawk boards with a dedicated serial port for a secondary GPS (GPS 2).