NavCAM-Contest

Work developed by ROS-Force for ActiveSpace's NavCAM contest.

Overview

Enviroment

This package was developed and tested with the following software:

• Ubuntu 20.04
• ROS Noetic
• Python 3
• CUDA v11.2
• cuDNN v8.1.1
• OpenCV v4.5.1
• Tensorflow v2.4.1

To compile this package for the Nvidia Jetson Xavier NX please go to this fork.

Camera

This package was tested using a Intel RealSense D435i camera. The following table was made to evaluate the accuracy of the camera in good lighting conditions against an opaque material.

Real Value (m)	Mean (m)	Standart Deviation (m)
0.2	0.181	0.00048
1.0	1.038	0.0068
3.0	3.153	0.078
5.0	5.322	0.231
7.0	7.707	0.723
9.0	10.018	0.732
10.0	11.453	1.22

Setting up the packages

The first step is to clone the repository:

cd ~
git clone --recursive [email protected]:ROS-Force/NavCAM-Contest.git

However, before we build and run our packages, we will need to install some dependencies first.

---

Dependencies

ROS Noetic

The best way to install ROS is to follow this tutorial. Follow the Desktop-Full installation to install packages like RVIZ to help visualize the results.

We need a few more ROS related packages, install them with the following command:

sudo apt install ros-noetic-octomap ros-noetic-ddynamic-reconfigure

Intel® RealSense™ D435i

Follow this tutorial to install Intel® RealSense™ SDK 2.0.

Other Python packages

To install the packages, run:

pip3 install tensorflow catkin_tools colorutils pathfinding

CUDA

The best way to install CUDA is to follow this tutorial. This package was tested with version 11.2 of CUDA.

cuDNN

In order to setup cuDNN in your system, there are several methods. Here, we will follow the package manager installation, fOr more information or a more detailed tutorial follow the NVIDIA cuDNN Documentation guide.

First we add the apt repositories and install the needed packages:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin 

sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub
sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /"
sudo apt update

Copy this commands to install the latest cuDNN version:

# for the latest cuDNN version (ensure this matches your CUDA version)
sudo apt install libcudnn8
sudo apt install libcudnn8-dev

OR use these commands to install a specific version of cuDNN:

# if you wish to install for a specific cuDNN version & CUDA version, then replace the values below
export cudnn_version=8.1.1.33-1
export cuda_version=11.2
sudo apt install libcudnn8=${cudnn_version}-1+${cuda_version}
sudo apt install libcudnn8-dev=${cudnn_version}-1+${cuda_version}

Compile OpenCV from source

To make use of GPU acceleration, we need to compile OpenCV from source. First step is to create a build folder:

cd NavCAM-Contest/depends/opencv
mkdir build
cd build

Then copy the following command and change the CUDA_ARCH_BIN value to the compute capability of your machine. If you don't know your compute capability look for your Graphic Card here:

cmake -D CMAKE_BUILD_TYPE=RELEASE \
-D CMAKE_INSTALL_PREFIX=/usr/local \
-D BUILD_EXAMPLES=ON \
-D INSTALL_PYTHON_EXAMPLES=ON \
-D INSTALL_C_EXAMPLES=ON \
-D OPENCV_ENABLE_NONFREE=ON \
-D WITH_CUDA=ON \
-D WITH_CUDNN=ON \
-D WITH_EIGEN=ON \
-D OPENCV_DNN_CUDA=ON \
-D ENABLE_FAST_MATH=1 \
-D CUDA_FAST_MATH=1 \
-D CUDA_ARCH_BIN=6.1 \
-D WITH_CUBLAS=1 \
-D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules/ \
-D HAVE_opencv_python2=ON \
-D HAVE_opencv_python3=ON \
-D BUILD_opencv_python2=ON \
-D BUILD_opencv_python3=ON \
-D BUILD_EXAMPLES=ON ..

Now compile:

# you can change the amount of make jobs with the -j argument to speed up the compilation process
make -j`nproc`
sudo make install

and you should be done!

RTAB-Map

RTAB-Map (Real-Time Appearance-Based Mapping) is a RGB-D SLAM approach based on a global loop closure detector with real-time constraints. It can be used to generate a 3D point clouds of the environment and/or to create a 2D occupancy grid map for navigation.

In order to maximize performance on RTAB-Map, we need to take advantage of the GPU acceleration features provided by certain OpenCV packages used by RTAB-Map, as well as faster mapping by mixing with another mapping algorithm called Octomap. Which means, we will need to compile RTAB-Map from source to link to the our CUDA-enabled version and to the Octomap libraries. Before we compile RTAB-Map however, we will also need to compile some additional packages:

# building g2o for graph optimization
mkdir /root/NavCAM-Contest/depends/g2o
cd /root/NavCAM-Contest/depends/g2o/build
cmake ..
make -j`nproc`
sudo make install

# building libnabo for libpointmatcher
mkdir /root/NavCAM-Contest/depends/libnabo/build
cd /root/NavCAM-Contest/depends/libnabo/build
cmake -DLIBNABO_BUILD_PYTHON=OFF ..
make -j`nproc`
make install

# building libpointmatcher for better ICP algorithm
mkdir NavCAM-Contest/depends/libpointmatcher/build
cd /root/NavCAM-Contest/depends/libpointmatcher/build
cmake ..
make -j`nproc`
sudo make install

# building GTSAM for graph optmization (if you installed g2o, you don't have to install this one)
mkdir NavCAM-Contest/depends/gtsam/build
cd /root/NavCAM-Contest/depends/gtsam/build
cmake -DGTSAM_BUILD_WITH_MARCH_NATIVE=OFF -DGTSAM_USE_SYSTEM_EIGEN=ON ..
make -j`nproc`
sudo make install

# building Octomap for better & faster mapping
mkdir NavCAM-Contest/depends/octomap/build
cd /root/NavCAM-Contest/depends/octomap/build
cmake ..
make -j`nproc`
sudo make install

Finally, we build and install RTAB-Map:

mkdir NavCAM-Contest/depends/rtabmap/build
cd /root/NavCAM-Contest/depends/rtabmap/build
cmake -DWITH_PYTHON=ON -DBUILD_EXAMPLES=OFF ..
make -j`nproc` 
sudo make install

---

After installing all the required dependencies, configure and compile the catkin workspace:

cd ~/NavCAM-Contest/catkin_ws

catkin init
catkin config --profile Release -x _release --cmake-args -DCMAKE_BUILD_TYPE=Release
catkin profile set Release
catkin config --install --merge-devel --extend /opt/ros/$ROS_DISTRO
catkin build -j`nproc` 
source ~/NavCAM-Contest/catkin_ws/devel_release/setup.bash

After the installation, edit your ~/.bashrc file to avoid having to source all the time:

echo "source ~/NavCAM-Contest/catkin_ws/devel_release/setup.bash --extend" >> ~/.bashrc

Sub-Packages

Object Detection

Inclination

Wall_follower

Mapping

Path Finding