This project provides the simulation and experiment results for the ICRA 2022 submission "BLock online EM algorithm for visual-inertial SLAM backend". This project focuses on the backend of SLAM systems. In particular, we investigate both optimization-based and EM-based algorithms. For experiments, we use EuRoC dataset.

You will need to install the following dependencies,

• OpenCV
• CMake
• Eigen3
• ceres

To build this project, just follow the standard CMake procedure.

mkdir build
cd build
cmake ..
cmake --build .

The files in test folder ensure that the functionality in the implemented SLAM backend in corrected.

There are 3 main simulations that we present in the paper:

• sim_easy_run.sh: Fig. 1. The 3D trajectory plot
• sim_fixed_run.sh: Fig. 2. The estimation error plot
• sim_exp_window_run.sh: Fig. 3. The estimation accuracy and the processing time plot

The setup parameters for the simulation are in config/config_sim.yaml. The parameters for running the simulation, for example the duration and the number of trials, are in each bash script.

The experiment is conducted by exp_run.sh, with parameters provided by the datasets in config/config_fpga_p2_euroc.yaml. The experiments in this project use the raw data from the images and the IMU sensors, as well as from the estimation trajectory from the frontend. Both kinds of data are preprocessed, and stored in data.