Lecture 9: MIT 6.832 Underactuated Robotics (Spring 2022) | "Computing Lyapunov Functions II"

This lecture is a rigorous and insightful exposition of sum-of-squares (SOS) optimization for computing Lyapunov functions, a cornerstone technique in nonlinear control and robotics. The instructor, a leading expert in the field, presents the material with clarity and depth, making it accessible to graduate students while maintaining technical precision.

The value of the information is high: the lecture bridges theory and practice, explaining both the mathematical foundations and the practical implementation using tools like Drake and PyTorch. The argumentation is solid, building from the relaxation of the Lyapunov equation to the formulation of SOS constraints as semidefinite programs. The instructor carefully highlights the gap between positive polynomials and SOS polynomials, using the Motzkin polynomial as a classic counterexample, which demonstrates a nuanced understanding of the limitations of the approach.

The scientific rigor is evident in the precise definitions and the attention to convexity and computational tractability. The instructor also addresses common misconceptions, such as the assumption that SOS is equivalent to global positivity, and clarifies that while the gap exists, it rarely poses practical issues for Lyapunov analysis.

The quality of sources is inherent to the lecture’s origin: it is part of MIT’s official curriculum, and the instructor references standard tools and literature in the field. The lecture does not cite specific papers, but the content is aligned with established research on SOS optimization and Lyapunov theory.

The adequacy between title and content is perfect: the lecture is entirely dedicated to computing Lyapunov functions using SOS techniques, as promised.

One minor limitation is the lack of explicit references to external sources, which could be beneficial for further study. However, this is typical for a lecture and does not detract from the overall quality.

In summary, this lecture is an excellent resource for anyone seeking a deep understanding of computational Lyapunov analysis. It is well-structured, technically accurate, and provides valuable insights into both theory and practice. The instructor’s teaching style is engaging, and the inclusion of code examples and practical tips enhances its utility.

The title accurately reflects the content: the lecture focuses on computational methods for Lyapunov functions, specifically using sum-of-squares optimization.

9/10

Lecture from MIT's Underactuated Robotics course, delivered by a leading expert in the field. The content is rigorous, mathematically sound, and based on well-established convex optimization and sum-of-squares techniques. The presentation includes practical examples and addresses common pitfalls, reflecting high reliability.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction and recap of previous lecture on Lyapunov functions and the relaxation from equality to inequality.
• 03:00Explanation of sum-of-squares (SOS) as a convex constraint and its relation to semidefinite programming.
• 06:00Discussion on the gap between positive polynomials and SOS polynomials, with the Motzkin polynomial as an example.
• 09:00Demonstration of using SOS to find the minimum of a polynomial, using the six-hump camel function.
• 12:00Q&A: applicability of SOS to non-polynomial functions, Taylor approximations, and piecewise functions.
• 15:00Discussion on the importance of polynomial bases and coefficient matching for practical implementation.
• 18:00Introduction to using SOS for verifying Lyapunov functions and regions of attraction.
• 21:00Mention of extensions to robustness and multibody systems, and practical considerations in implementation.