SERVO
Original Implementation: ORB_SLAM2
Authors: Raul Mur-Artal, Juan D. Tardos, J. M. M. Montiel and Dorian Galvez-Lopez (DBoW2)
Modified: Ueli Graf, 07.03.2017
Current version: 1.0.0
This is SERVO, a robust SLAM extension to ORB-SLAM2. SERVO relies on additional odometry estimates provided in a ROS environment. At the time of writing, only stereo SLAM mode is supported.
#1. Quick start
##1.1 Install dependencies
Pangolin including Required Dependencies glew, cmake
sudo apt-get install libglew-dev
sudo apt-get install cmake
git clone https://github.com/stevenlovegrove/Pangolin.git
cd Pangolin
mkdir build
cd build
cmake ..
make -j
kindr:
sudo add-apt-repository ppa:ethz-asl/common
sudo apt-get update
sudo apt-get install ros-indigo-kindr-*
OpenCV Required at leat 2.4.3. Tested with OpenCV 2.4.11
Eigen3 Required at least 3.1.0
BLAS and LAPACK:
sudo apt-get install libblas-dev
sudo apt-get install liblapack-dev
##1.2 Install additional odometry software This software was tested with rovio which is included as a submodule. If rovio is to be used, this step can be skipped. If a different visual-inertial odometry framework should be used, install it seperately. Reference body velocity (IMU body velocity in case of rovio) should be published to /rovio/odometry and the transformation from reference body to left camera should be published to /rovio/extrinsics0.
##1.3 Clone and build underlying frameworks
Create a catkin workspace, clone the repository and submodules and execute the build script found in the root directory at SERVO/build.sh. Make sure your terminal is setup with the appropriate ROS environment variables.
mdir src
catkin init
cd src
git clone --recursive https://github.com/jonasbareiss/SERVO.git
cd SERVO
chmod +x build.sh
./build.sh
##1.4 Build catkin packages
catkin build rovio orb_slam2 -DCMAKE_BUILD_TYPE=Release
##1.5 Unpack rosbag
Join the rosbag provided for testing as a split archive and decompress it for better runtime performance.
Open a terminal at SERVO/Examples/ROS/
cat SERVO_bag_* > SERVO.bag
rosbag decompress SERVO.bag
##1.6 Launch Use the exemplary launch files found at SERVO/Examples/ROS/launch/
Open a terminal at SERVO/
source ../../devel/setup.bash
roslaunch Examples/ROS/launch/orb_slam2.launch
launches ORB_SLAM2 using the provided rosbag in original configuration (Stereo/ORB_SLAM2.yaml). Wait for a few seconds until the vocabulary has been loaded, then press space bar in the terminal to start playback.
Use
roslaunch Examples/ROS/launch/servo.launch
to launch with the proposed changes (Stereo/SERVO.yaml) and verify increased robustness and performance. Wait for a few seconds until the vocabulary has been loaded, then press space bar in the terminal to start playback.
Use the parameter named ORBextractor.nFeatures in the *.yaml files for tuning computational cost.
The current version assumes that rovio is used in parallel (that is, rovio must be running - it is launched automatically when using the provided launch files) and publishes odometry estimates to /rovio/odometry and the transformation from camera frame to odometry reference frame to /rovio/extrinsics0. rosbag can be selected within the launch file. If the resulting estimated trajectory is implausible, try decrease the rate (-r parameter for the rosbag node section in the launch files) or tune the compational complexity by changing the ORBextractor.nFeatures parameter in the *.yaml settings files found at Examples/ROS/Stereo/ Topics can be changed by changing the remap command in launch files or by changing the source code at ros_stereo.cc.
#2 Options
Most configuration parameters can bechanged by a .yaml-file found in Examples/Stereo. Use the launch file to specitfy the .yaml file that should be loaded. The parameter file contains short descriptions of the influence of each variable.
#3 Available Topics Additional to all topics of the respective base implementation:
/orb_slam2/odometry: odometry with respect to the left camera (cam0)
#4 Changes/Implementation
The extension implements a supported tracking procedure for ORB_SLAM2 that incorporates visual-inertial information in the SLAM feature matching and map building process. To this end, VI-odometry is used to support the matching process by providing a pose initial guess at each timestep. Furthermore, the external pose estimate serves as a pose prior in the pose optimization and helps performs a dead reckoning method for contiunous tracking even when no features are matched successfully.
Implementation details will be published soon.
#5 Original Documentation For extensive and additional documentation on the base implementations of the frameworks used, see the original repositories of ORB_SLAM2 and rovio.