AI 539: ST/ Safe and Reliable Autonomy

School of EECS, Oregon State University

Instructor: Sandhya Saisubramanian

Instructor Contact: sandhya.sai AT OregonState DOT edu

Course Credits: 4 units

Meeting Information: MW 10am-11:50am in STAG 261 - Strand Agriculture Hall 261

Instructor Office Hours: TBA

Course Overview: Recent years have seen dramatic rise in the deployment of AI systems across sectors. Despite the advances in AI, many of the deployed systems produce undesirable consequences which affects their safety and reliability. Academic researchers and industry practitioners have recognized the growing need for designing and deploying safe and reliable AI systems.

This course will highlight the challenges and recent trends in intelligent decision-making to improve the safety and reliability of autonomous systems from a computational perspective, and identify important open research questions. This is a seminar-style course in which each student will summarize the papers in the reading list, lead the discussions on analyzing the strengths and weaknesses of each paper, and apply the learned concepts in a final project. The reading list will contain a mix of papers that cover both theory and applications of safe and reliable autonomy, published at leading AI/ robotics conferences and journals (AAAI, ICAPS, IJCAI, AAMAS, NeurIPS, ICRA, IROS, RSS, JAIR, JMLR, etc). Application areas include robotics, autonomous cars, and healthcare, among others. The topics covered will broadly include techniques for safe sim-to-real transfer; using human assistance to improve system reliability, including human feedback and imitation learning; explainability to improve reliability; and safe exploration.

Key takeaways from this course for the students:

• Develop a deep understanding of the challenges and current tools available to design safe and reliable AI systems

• Identify important open research questions in this space

• Hands-on experience implementing approaches to improve AI safety and reliability

• Learn to analyze the strengths and weaknesses of a research paper, and communicate scientific content to a peer audience

Announcements and Discussion Board: We will be using Canvas for the course. All course-related announcements and readings will be posted on Canvas.

Course Evaluation: Paper summary: 20%, Paper presentation and discussion: 40%, Final project: 40%

Course Prerequisites: Familiarity with sequential decision making (Automated Planning/ RL), ideally AI 533 or ROB 534.

Email me for overrides if you:

• are a graduate student unable to register (Non-CS/AI/Robotics majors)

• have not satisfied the prerequisites but have taken a rigorous undergraduate course in AI

• are an undergrad interested in this course

Academic Honesty Policy: You are encouraged to discuss the readings with your classmates. You may also discuss your projects, but only to brainstorm ideas. All writing and coding must be done on your own. Sharing or copying solutions is unacceptable and could result in failure. If in doubt about a particular collaboration, ask for permission in advance.

Tentative Course Schedule: