Mobile Robots: Localization, Mapping and SLAM
Where am I and what is around me: odometry and its calibration, map representations, Monte Carlo localization in known maps, occupancy grid mapping, the SLAM problem, EKF-SLAM, particle-filter SLAM, graph-based SLAM with robust back ends, lidar and visual-inertial SLAM, navigation stacks, map management and ATE/RPE evaluation, ending with a warehouse SLAM design.
ROB-340 Year 3 3 credits 12 lessons · 5 h 53 min
Offered by HSM University · Robotics
This is the American version of this course. See the Canadian version.
What you'll learn
- Compute and calibrate differential-drive odometry, explain how systematic and random errors make it drift, and model its uncertainty for use as a motion model
- Compare occupancy grids, point clouds, octrees, signed distance fields, feature, topological and semantic maps, and choose map layers for a given robot and task
- Configure and diagnose Monte Carlo localization in a known map, including adaptive particle counts, kidnapping recovery, ambiguous beliefs and scan matching
- Build occupancy grid maps from known poses with log-odds updates, inverse sensor models and ray tracing, and choose resolutions and thresholds
- Formulate online and full SLAM, explain why correlated errors make SLAM solvable, and analyze data association, validation gates and loop closure
- Explain EKF-SLAM's state, prediction, landmark initialization and update, trace how uncertainty grows and collapses, and identify its scaling and consistency limits
- Explain Rao-Blackwellized particle-filter SLAM, compute importance weights and effective sample sizes, and analyze particle depletion
- Formulate and solve pose-graph SLAM with Gauss-Newton on sparse normal equations, anchor the gauge freedom, and apply robust kernels and incremental solvers
- Register lidar scans with ICP, NDT and correlative matching, detect degenerate geometry, and describe the structure of practical 2-D and 3-D lidar SLAM systems
- Describe feature-based, direct and visual-inertial SLAM, including keyframes, bag-of-words place recognition, scale drift and IMU preintegration
- Explain how a navigation stack uses maps and localization, manage maps over time, and evaluate SLAM with ATE and RPE
- Design, budget and plan the evaluation of a complete mapping and localization system for a warehouse robot, including failure modes
Course content
12 lessons · 5 h 53 min total · final exam
- 1 Mobile robot platforms and odometry 32 min · quiz
- 2 Map representations 30 min · quiz
- 3 Localization in a known map 28 min · quiz
- 4 Occupancy grid mapping 28 min · quiz
- 5 The SLAM problem 28 min · quiz
- 6 EKF-SLAM 28 min · quiz
- 7 Particle-filter SLAM: FastSLAM and grid-based mapping 31 min · quiz
- 8 Graph-based SLAM 28 min · quiz
- 9 Lidar SLAM in practice 30 min · quiz
- 10 Visual and visual-inertial SLAM 29 min · quiz
- 11 Navigation stacks, map management and evaluation 29 min · quiz
- 12 Capstone: a SLAM system for a warehouse robot 32 min · quiz
- E Final exam 20 questions
Requirements
Recommended background. You can start any course at any time.
Description
Where am I and what is around me: odometry and its calibration, map representations, Monte Carlo localization in known maps, occupancy grid mapping, the SLAM problem, EKF-SLAM, particle-filter SLAM, graph-based SLAM with robust back ends, lidar and visual-inertial SLAM, navigation stacks, map management and ATE/RPE evaluation, ending with a warehouse SLAM design.
Part of the Science stream, typically taken in year 3.