Probabilistic Robotics and State Estimation
Reasoning under uncertainty: probability and Gaussians, the Bayes filter and histogram localization, motion and sensor models, Kalman, extended and unscented Kalman filters, particle filters and Monte Carlo localization, practical sensor fusion with gating, information filters, smoothing and factor graphs, pose estimation, tracking and data association, and a capstone localization design evaluated for accuracy and consistency.
ROB-310 Year 3 3 credits 12 lessons · 6 h 5 min
Offered by HSM University · Robotics
What you'll learn
- Explain the sources of uncertainty in robotics and use random variables, Gaussians, covariance and Bayes' rule to represent and update beliefs
- Derive the Bayes filter from the Markov assumption and implement it exactly with a histogram for low-dimensional localization
- Apply odometry and velocity motion models, sample from them, and propagate Gaussian uncertainty through them with Jacobians
- Build beam, likelihood field and range-bearing measurement models that handle noise, outliers and correlated beams
- Implement the Kalman filter, interpret its gain and innovation, and tune process and measurement noise using NIS and NEES
- Apply the extended Kalman filter to landmark-based localization and recognize when linearization makes it biased or inconsistent
- Use the unscented transform and unscented Kalman filter, and choose between EKF, UKF and particle filters for a given problem
- Implement particle filters and Monte Carlo localization with systematic resampling, effective sample size and recovery from kidnapping
- Design practical multi-sensor fusion systems with bias states, time synchronization, coordinate frames and Mahalanobis gating
- Use the information form, batch least squares and factor graphs to smooth trajectories, and explain sliding-window and incremental smoothing
- Estimate object poses by point alignment, track multiple targets, and solve data association with gating and global assignment
- Specify, simulate, evaluate and validate a localization system against requirements for accuracy, consistency, rate, robustness and recovery
Course content
12 lessons · 6 h 5 min total · final exam
- 1 Uncertainty in robotics and probability review 31 min · quiz
- 2 The Bayes filter and histogram localization 29 min · quiz
- 3 Motion models 30 min · quiz
- 4 Sensor models 28 min · quiz
- 5 The Kalman filter 29 min · quiz
- 6 The extended Kalman filter 29 min · quiz
- 7 The unscented Kalman filter and comparing Gaussian filters 29 min · quiz
- 8 Particle filters and Monte Carlo localization 29 min · quiz
- 9 Sensor fusion in practice 33 min · quiz
- 10 Information filters, smoothing and factor graphs 32 min · quiz
- 11 Estimation for manipulation and perception: pose, tracking and data association 31 min · quiz
- 12 Capstone: designing and evaluating a localization filter for a mobile robot 35 min · quiz
- E Final exam 20 questions
Requirements
Recommended background. You can start any course at any time.
Description
Reasoning under uncertainty: probability and Gaussians, the Bayes filter and histogram localization, motion and sensor models, Kalman, extended and unscented Kalman filters, particle filters and Monte Carlo localization, practical sensor fusion with gating, information filters, smoothing and factor graphs, pose estimation, tracking and data association, and a capstone localization design evaluated for accuracy and consistency.
Part of the Science stream, typically taken in year 3.