ProbRob-Project: Planar Monocular SLAM

This project implements a full SLAM pipeline using only a monocular camera and odometry in a planar environment (SE(2) motion embedded in SE(3)). It includes:

• Landmark triangulation from 2D image correspondences
• Pose graph and projection-based Bundle Adjustment (Total Least Squares)
• Evaluation against ground truth for both trajectory and map

Requirements

• Octave (tested on version ≥ 6)
• Standard packages (no toolboxes required)

Running the Code

Run the main entry point:

octave main.m

Note

This is an attempt at implementing Planar Monocular SLAM from scratch which is currently not successful:

Current problems:

• Triangulated landmarks suffer from high reprojection error.
• Bundle Adjustment (BA) diverges and unstable.
• The trajectory estimate does not meaningfully improve over raw odometry.