Course syllabus

Course-PM

SSY345 Sensor fusion and nonlinear filtering lp4 VT20 (7.5 hp)

The course is offered by the Department of Electrical Engineering

Contact details

Examiner
Docent Lars Hammarstrand, lars.hammarstrand@chalmers.se

Teaching assistants
Yuxuan Xia, yuxuan.xia@chalmers.se
Ahmet Oguz Kislal, kislal@chalmers.se

Student representatives
Lucas Billstein (MPSYS), lucasbi@student.chalmers.se
Calle Hellberg (MPSYS), Calle.Hellberg.9710@student.uu.se
Johan Högstrand (MPSYS),  johhogs@student.chalmers.se
Frans Erik Isaksson (MPCOM), fraisa@student.chalmers.se
Fredrik Kerstis (MPSYS), kerstis@student.chalmers.se

Changes due to coronavirus

The syllabus below describes how we would like to give the course under normal circumstances. However, we acknowledge that we will need to make changes and adapt in order to protect the health and wellbeing of both students and teachers. We will follow Chalmers' recommendations regarding this. In this Section, we will continuously list the changes that we are making to the course. 

Update 200317: For the time being, all practice sessions will be given online using zoom instead of on-campus. We are currently looking into how we can do this as good as possible but this is a new experience for us but we hope to make the best of the situation. We will update you as soon as we have figured out more on how we will do this. Note that these practice sessions are mandatory and that you are expected to have watched to corresponding video lecture before the class.

Update 200317: On-campus version of Lecture 1 is cancelled.  The content of the lecture is instead given on scalable-learning and by carefully reading the information here on Canvas. 

Update 200313: We are currently looking into alternative ways of giving the course where most/all activities can be performed remotely and to limit the need to be at campus. For example, we are looking at giving the course completely online or, alternatively, offering the possibility to follow the in-class activities remotely. We will update the information on the course home page as things develop.

Note:  There are many mandatory activities in this course which you need to attend. However, it is important that you do not come to class if you are feeling sick. If this is the case, please let us know and we will sort something out. In this respect, we will be flexible with the number of late days that you can have (see below).

Course purpose

The purpose of this course is to give a thorough introduction to sensor fusion in time-varying settings (also known as filtering or smoothing), i.e., how to perform state estimation using a variety of sensors. Such methods continue to receive considerable attention due to their high versatility; famous examples include that they enabled the landing on the moon and that they are currently important for the development of self-driving cars.

In the course, we emphasize on the positioning of vehicles, people, mobile phones, robots, etc, though the potential applications go way beyond that. The intention of the course is to provide a solid theoretical background and to give hands-on experience on how to apply the techniques to solve problems of practical importance.

Schedule

The schedule for the in-class activities in the course can be found in TimeEdit

Note 1: There are several mandatory activities, both in terms of mandatory in-class activities as well as assignment deadlines, that you need to keep track of. To help out we provide a summary overview of the different activities here. 

Note 2: Both practice sessions and solution discussions are half-class and if you are in Practice session group 1 you are expected to go to the first occasion and if you are in Practice session group 2 you are expected to go to the second.

Note 3: The solution discussions are only 45 min for each group so group 1 is expected to come the first hour and group 2 the second hour.

Course literature

We have aimed to make the course self-contained in the sense that you should be able to solve all tasks using the material provided in the course. However, if you would like to read a book covering the material, we recommend:

Simo Särkkä, Bayesian Filtering and Smoothing, Cambridge University Press, 2013.

You can find the book for free as a pdf on Simo's homepage here.

Course design

The course is given in a flipping-the-classroom style which in short means that you are expected to watch lecture videos before coming to class and that the time in class will be used to be more actively working with the material in the course. A more detailed explanation of what it is and a motivation for why we have chosen to use it can be found here.

In order for the flipped-classroom model to be effective, it is important that all students keep up to date with the material in the course. To make sure of this, there are several mandatory activities and tasks that need to be completed before a certain deadline.  These mandatory elements are (click on the links for more information about each activity):

• Online video lectures
  Before each practice session, you are expected to watch a set of online video lectures covering the material of that specific practice session.
• Practice sessions (PS)
  In-class sessions where we will actively discuss and work with the material of the corresponding video lecture.
• Home assignments (HA)
  After getting familiar with the material through the video lectures and the practice sessions you will get to apply your knowledge in a set of home assignments that you will solve individually. Each home assignment typically covers the material of two practice sessions. 
• Solution discussions (SD)
  After each home assignment, we have active sessions where we discuss the result of the home assignment and what conclusions and insights that we have obtained.
• Project
  To conclude the course you get to apply your knowledge together with a fellow student to solve a small problem using real measurements.
• Oral examination
  There is no written exam in the course. The examination is done by the home assignments, the project and a final oral exam. 

Students are allowed 8 “late days” if they cannot make a deadline. These late days do not extend future deadlines. If a late day is used for an in-class practice session (which implies that they miss that session), then the students have to turn in written solutions to problems related to that class (within a week after they receive the tasks). Missing out on a class where we discuss solutions to the home assignments only costs one late day. 

Lecture content

The content in the different lectures can be briefly summarized as:

• L1: course introduction and a primer in statistics (conventional lecture), March 23, 13.15-15.00, SB-H1. 
• L2: Bayesian statistics and decision theory (flipped).
• L3: State-space models and optimal filters (flipped).
• L4: The Kalman filter and its properties (flipped).  
• L5: Motion and measurement models (flipped). 
• L6: Nonlinear Gaussian filters (flipped),
• L7: RTS and Gaussian smoothing (flipped).  
• L8: Particle filters (flipped)
• L9: Guest lecture by Daniel Svensson, Product Owner, Mult-Object Tracking, Zenuity (Zoom lecture), May  29, 13.15-15.00.

Please, see schedule on scalable learning for when you need to have watched and answered the self-assessment questions for the flipped lectures above. Failure to meet a deadline will result in the loss of 1 late day per day. 

Changes made since the last occasion

A summary of changes made since the last occasion.

Learning objectives and syllabus

After the course, students should be able to

• explain the fundamental principles in Bayesian estimation
• describe and model commonly used sensors' measurements
• summarize and compare the most typical motion models in positioning in order to know when to use them in practical problems
• derive the expression for an optimal filter
• describe the essential properties of the Kalman filter (KF) and apply it on linear state-space models
• implement the key nonlinear filters (above all the extended Kalman filter, unscented Kalman filter and the particle filter) in Matlab, in order to solve problems with nonlinear motion and/or sensor models
• select a suitable filter method by analyzing the properties and requirements in an application
• apply the linear and nonlinear Rauch-Tung-Striebel smoothers to general smoothing problems
• solve a variety of important real-world filtering and smoothing problems, by employing and adapting the above knowledge to a variety of applications.

Link to the syllabus on Studieportalen.

Study plan

Examination form

There is no written exam in the course. Instead, you will be examined through your performance on the home assignments, the project and the oral exam.

The possible grades in this course are 3, 4, 5 or not passed. In order to pass the course, to obtain a grade 3, 4, or 5, you need to fulfil mandatory parts of the course without exceeding the eight late days. Students that wish to obtain a higher grade can do so by collecting points of excellence (POE) throughout the course, according to the following system:

	POE (Max)
Home Assignments (total / HA)	16
Well illustrated and commented analysis	14
Well performed peer-review	2
Project (total)	16
Report	14
Well performed peer-review	2
Oral examination (project and course)	20
Total	100

If you submit an analysis or project report late the maximum number of POE is divided by 2. For revised solutions, the maximum number of POE is divided by 2 (if you also submit it late you divide by 4).

In total, it is possible to obtain 100 POE and students who manage to collect 50-69 POE  will obtain grade 4 whereas students who collect 70 POE, or more, will obtain grade 5. 

Note that the performance on quizzes in the lecture videos and in-class practices is not graded apart from mere participation. We do so intentionally to emphasize that we arrange these activities for learning and not as a test. By doing so, we also seek to encourage all participants to ask about all sorts of things that they find interesting or are struggling with.