QVPR/aarapsi_robot_pack

aarapsi_robot_pack

ROS package for the AARAPSI project (github) (website)

Installation

This assumes you have a valid ROS noetic (ROS1 for Ubuntu 20.04 with python 3.8.10) environment. If not, please visit the ROS wiki for noetic installation and follow the tutorials to install and configure ROS. Once ROS is configured, install our catkin workspace or build one from scratch with the equivalent packages. For more information, visit the wiki page for ROS catkin workspaces.

• In the root of the catkin workspace, execute rosdep install --from-paths src --ignore-src -r -y
• Install missing python libraries: pip3 install numexpr fastdist bokeh gdown nvidia-ml-py3 piexif

Package Contents

• cfg: Configuration files including yaml files, reference documentation, SVMs, urdf files
• data: Generated or utilised data sets, maps, images
• exe: Executables
• launch: roslaunch files for nodes in this package as well as custom configurations for other packages such as HDL_graph_slam
• media: Images for using in GUIs or plots
• msg: rosmsg files
• nodes: rosnode python files
• rviz: RViz configuration files
• scripts: Helper nodes, testing scripts, and experiments
• servers: Bokeh server files (Broken)
• src: Where pyaarapsi lives
• srv rosservice files

Weird things to note for aarch64 (AgileX Scout Mini)

• Xavier AGX Error: libgomp-d22c30c5.so.1.0.0: cannot allocate memory in static TLS block
  • FIX: export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libgomp.so.1
• Xavier AGX Torch:
  • https://forums.developer.nvidia.com/t/pytorch-for-jetson/72048
    • https://nvidia.box.com/shared/static/i8pukc49h3lhak4kkn67tg9j4goqm0m7.whl
    • pip install ./torch-2....
  • I learnt the hard way: do not try to custom build or install CUDA/CUDNN/anything. Sort out all of these things first; you must use the CUDA and CUDNN version installed with jetpack otherwise tears will be shed and days will be wasted. There are prebuilt PyTorch and torchvision wheels you can install if you need them.