Continuous-Time Spline Visual-Inertial Odometry

Updates

• [June 2022]: The position and acceleration formulation is suboptimal. Please checkout the new branch for improved formulation.

Related Publications

• Direct Sparse Odometry, J. Engel, V. Koltun, D. Cremers, In IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2018
• Continuous-Time Spline Visual-Inertial Odometry, J. Mo and J. Sattar, In IEEE International Conference on Robotics and Automation, 2022, arXiv.

Dependencies

• ROS

• DSO dependencies

Install

cd catkin_ws/src
git clone https://github.com/IRVLab/spline_vio.git
cd ..
catkin_make

Usage

• Use Kalibr toolbox to calibrate camera and IMU.

• Convert camera parameters to DSO format.

• Create a launch file following the example of tum.launch.

roslaunch spline_vio [YOUR_LAUNCH_FILE]

• Ctrl-C to terminate the program, the final trajectory (results.txt) will be written to ~/Desktop folder by default.

Output file

• results.txt: poses of all frames, using the TUM RGB-D / TUM monoVO format ([timestamp x y z qx qy qz qw] of the cameraToWorld transformation).

Modifications to DSO for this project

• ROS interface: main.cpp
• Predict pose using spine for front-end tracking: FullSystem::trackNewCoarse()
• IMU/Spline state: HessianBlocks.h
• IMU/Spline Jacobians (check Jacobians.pdf): HessianBlocks::getImuHi()
• Constraints Jacobians (check Jacobians.pdf): EnergyFunctional::getImuHessianCurrentFrame()
• Solve the constraint nonlinear optimization problem: EnergyFunctional::solveSystemF()