Stereo Visual Inertial Pose Estimation Based on Feedforward-Feedback Loops
This work addresses pose estimation for UAVs and robotics, offering a novel control-based approach that is incremental in improving stability and performance.
The authors tackled visual-inertial pose estimation by modeling it as a control system with feedback and feedforward loops, achieving high accuracy and robustness compared to state-of-the-art visual SLAM approaches on the EuRoc MAV dataset.
In this paper, we present a novel stereo visual inertial pose estimation method. Compared to the widely used filter-based or optimization-based approaches, the pose estimation process is modeled as a control system. Designed feedback or feedforward loops are introduced to achieve the stable control of the system, which include a gradient decreased feedback loop, a roll-pitch feed forward loop and a bias estimation feedback loop. This system, named FLVIS (Feedforward-feedback Loop-based Visual Inertial System), is evaluated on the popular EuRoc MAV dataset. FLVIS achieves high accuracy and robustness with respect to other state-of-the-art visual SLAM approaches. The system has also been implemented and tested on a UAV platform. The source code of this research is public to the research community.