This is the code release for ICRA Submission "Obstacle avoidance using raycasting and Riemannian Motion Policies at kHz rates for MAVs"
- ROS Noetic
- CUDA (10.2 - 11.5 are tested by nvblox developers, others may or may not work)
System Dependencies for nvblox (nvblox itself is cloned as a submodule of this repository which is done automatically):
sudo apt-get install -y libgoogle-glog-dev libgtest-dev libgflags-dev python3-dev
cd /usr/src/googletest && sudo cmake . && sudo cmake --build . --target install
The following commands creates a new catkin environment, clones the required repositories and builds.
mkdir rmp_planning
cd rmp_planning
mkdir src
catkin init
cd src
git clone git@github.com:ethz-asl/eigen_catkin.git
wstool update
git clone https://github.com/ethz-asl/reactive_avoidance --recurse-submodules
catkin build
Gflags dependency conflict (build fails when compiling nvblox with errors about gflags). Go to the src directory and run:
rm -rf rmpcpp/rmpcpp_planner/nvblox
git clone -b gflags_namespace_fix git@github.com:Isarm/nvblox.git rmpcpp/rmpcpp_planner/nvblox
catkin build
All planner types are compiled into the same executable and can be selected via command line arguments.
See the parser class for the full list of options. I’ll briefly list the most important classes for the RMP planner.
The policies are implemented in
- NVBlox Raycasting obstacle avoidance policy:
rmpcpp_planner/src/policies/raycasting_CUDA.cc - Lidar ray obstacle avoidance policy:
rmpcpp_planner/src/policies/lidarray_CUDA.cc - ESDF policy:
rmpcpp_planner/src/policies/simple_ESDF.cc
The tester class sets up testing runs, using the worldgen class to generate random worlds or load custom worlds.
There are a lot of options you can pass to the executable, which are all defined in the parser class.
Check out rmpcpp_planner/src/testing/parser.cc for the full list of options.
Some of our code heavily depends on nvblox - check it out here: https://github.com/nvidia-isaac/nvblox, its a great open-source package for 3D mapping on robots.