A Tutorial on Graph-Based SLAM
Cyrill Stachniss Wolfram Burgard
-
A graph
- Node: The poses of the robot at different points
- Edges: Constraints between the poses. Obtained from observations or movement actions
- a probability distribution over a relative transformations between two pose nodes
- eg odometry measurement, aligning observations from two robot locations.
- Include data association problem. included in front-end.
- tasks of pose graph slam
- Graph construction (Front-end)
- Graph optimization (Back-end)
-
Solutions for SLAM
Estimating the robot trajectory and the map of environment.
- Filtering: on-line state estimation of current robot position and the map
- Smoothing: estimate the full trajectory from full set of measurements.
-
Solving a large error minimization problem to find a configuration of the graph (trajectory) that is maximally consistent with the measurement.
min x ∑ < i , j > ∈ C e i j T Ω i j e i j \min_{x}{\sum _{<i,j>\in C } e^T_{ij}\Omega _{ij} e_{ij}}\\ xmin