About

This project implemented a Extended Kalman Filter to solve the Simultaneous Localization and Mapping (SLAM) problem on a robot moving in an initially unknown environment. Specifically we implemented a Visual Inertial Extended Kalman Filter that uses a gaussian distribution to estimate the robot pose and landmark positions at any given time. The map consists of the collection of these landmark feature points.

Project Structure

├── README.md
├── code
│   ├── main.py
│   ├── myutils.py
│   ├── pr3_utils.py
│   └── slam_scratch_code.ipynb
├── data
│   ├── 03.npz
│   ├── 03_video_every10frames.avi
│   ├── 10.npz
│   └── 10_video_every10frames.avi
├── plots
├── problem_statement.pdf
├── report
│   └── report.pdf
└── requirements.txt

Technical Report

• Saqib Azim. "Visual-Inertial SLAM" March 2023

To run the Visual-Inertial SLAM code

cd code/

python3 main.py --ds=3 --mapping --featSkip=6 --initPoseCov=0.01 --initLMCov=1.0 --vcov=10 --wcov=1e-2 --distThresh=200

• --ds (either 3 or 10)
• --mapping if specified, then performs visual mapping otherwise directly performs visual-inertial SLAM
• --initPoseCov: initial pose covariance diagonal values
• --initLMCov: initial LM covariance diagonal values
• --vcov: observation model noise covariance values
• For all parts (IMU localization, visual mapping and visual-inertial SLAM) - The main code is main.py while most utility functions are implemented in myutils.py and in pr3_utils.py

Results

Independent IMU predict and Landmark Updates

Data 03.npz

Data 10.npz