Machine Learning for Robot Perception (Advanced Topics in CS)

Longin Jan Latecki

Computer and Information Sciences Dept.

 

Graduate Seminar CIS 9590 in Fall 2008, Wednesday, 4:40-7:10 PM in EA519

 

    Schedule         References

 

We will study powerful machine learning algorithms that have successfully contributed to the recent progress in robot perception. The progress is documented in the spectacular way by the outcomes of DARPA Grand Challenge Competitions in 2005 and 2007.

The students will also be introduced to the underlying geometric, statistical and computational concepts of robot perception. The input data from which robot perception is inferred is coming from two main types of spatial sensors: LIDARs (also called laser range finders) and visible light cameras. The recent progress in robot perception is to large extend due to the usage and progress in LIDARs, which provide robots with precise depth perception. However, visible light cameras still remain an important sensor for robot spatial perception.

In addition to learning the machine learning algorithms and the underlying concepts of robot perception, the students will be able to estimate the perceptive abilities of autonomous robots.

The course topics listed below are sorted by types of machine learning algorithms that will be covered. These algorithms are widely used not only in robotics but also in many other machine learning applications, in particular in data mining and web mining.

Course topics:

Prerequisites:

CIS 8511 Programming Techniques, and at least one of the following five courses: CIS 8526 Machine Learning, CIS 8525 Neural Computation, CIS 8527 Data Warehousing, Filtering and Mining, CIS 9664: Knowledge Discovery and Data Mining, or CIS 9601 Computer Graphics and Image Processing, or permission of the instructor.

Elementary knowledge of statistics is required. We will review all required statistical concepts during the course.

 

Instructor

Exams, Project and Grading

Final: The final exam will be given to those students who did not complete their projects and did not give their presentations, in which case it will account for 70% of the grade.

Homework: Homework due dates will be announced. Late homework will result in a penalty. 

Class attendance: Class attendance is expected, and may be recorded from time to time. Absences for legitimate professional activities and illnesses are acceptable only if prior notice is given to the instructor by e-mail. Scheduling conflicts with your work, extra-curricular activities, or any other such activities is not a valid excuse. Although attendance is not a specific part of the course evaluation it has a direct effect on class participation. If you are not in class you cannot participate. Class participation means that you attend class regularly and have completed your assigned readings. It means that you ask relevant questions and make informed comments in class. Class participation will contribute to the final grade.

Honor Code

Disabilities

I encourage students with disabilities, including "invisible" disabilities such as chronic diseases and learning disabilities, to discuss with us any appropriate accommodations that we might make on their behalf. Student must provide me with a note from the office of Disability Resources and Services at in 100 Ritter Annex, 215-204-1280, regarding their disability.

Text

The course will be based on research articles and on parts of the following text books:

 

Christopher M. Bishop. Pattern Recognition and Machine Learning. Springer 2006. book website

David A. Forsyth and Jean Ponce . Computer Vision: A Modern Approach. Prentice Hall 2003.  website

Richard O. Duda, Peter E. Hart, and David G. Stork. Pattern Classification. (2nd ed.), Wiley 2000. website

 

Questions, email: latecki@temple.edu