Safe Autonomy, Spring 2025CSCI 4963/6963
Personnel
- Instructor:
Rado Ivanov
(ivanor@rpi.edu), Lally 309
- Office hours: M 1-2pm, W 2-3pm, Th 1-2pm (Lally 309)
Class Time and Location
- Class: MTh 10am-noon (Greene 120)
Course Description
This course will explore the challenges with ensuring the safety of modern autonomous systems. In the first half of the course, we will discuss some of the safety vulnerabilities of machine learning models, such as neural network robustness issues, and go over methods to alleviate these issues. In the second half, we will cover concepts from dynamical system control and analysis. We will also discuss how to analyze dynamical systems with neural network controllers in the loop. Students will get exposure to adversarial learning and control techniques, as well as to a number of neural network verification tools such as Reluplex, Verisig and others.
Textbook
There is no required book for the course. All of the necessary material will be included in the lecture slides. I will suggest additional reading material before each lecture. We will follow some of the material in the following two books, which are available for free online:
- Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep learning. MIT press, 2016. (available online)
- Lee, Edward Ashford, and Sanjit A. Seshia. Introduction to embedded systems: A cyber-physical systems approach. MIT Press, 2017. (available online)
Additionally, I will cover individual chapters from the following books. All necessary material will be provided in the lecture slides, though you are welcome to obtain your personal copies:
- Baier, Christel, and Joost-Pieter Katoen. Principles of model checking. MIT press, 2008.
- Thrun, S., Burgard, W. and Fox, D. Probabilistic robotics. Kybernetes, 2006.
- Alur, Rajeev. Principles of cyber-physical systems. MIT press, 2015.
- Scharf, Louis L., and Cédric Demeure. Statistical signal processing: detection, estimation, and time series analysis. Prentice Hall, 1991.
- Bertsekas, Dimitri. Dynamic programming and optimal control: Volume I. Vol. 1. Athena scientific, 2012.
Grading
Students will be graded on 5 homework assignments and 2 presentations. The final grade for the course will be determined as follows:
- Class participation (10%)
- Homework (60%)
- In-class presentations (30%)
Late submission rule: Each homework will be worth 100 points. Late programs will be penalized 5 points per day, midnight to midnight. Assignments which are late by more than 7 days will receive a score of 0.
Useful Resources
Announcements
It is the student's responsibility to be aware of and understand all announcements made in the lectures.
Assignments
Please check the tentative schedule for a tentative homework deadline schedule. Unless otherwise noted, each assignment will be due at midnight Thursday. I will post each assignment after the previous one is completed. Email announcements will be sent accordingly.
All assignments will be posted on and must be submitted through LMS. You are expected to work alone on all assignments, unless specifically noted otherwise -- please check the syllabus for a clarification of what constitutes academic dishonesty.
Lectures
All lecture material will be posted on LMS.