Lecture 7 | MIT 6.832 (Underactuated Robotics), Spring 2018

This lecture is a masterclass in bridging theoretical control theory with computational practice. Russ Tedrake’s presentation is exceptionally clear, building from the discrete case to continuous state spaces with function approximation, always highlighting the computational implications. The mathematical derivations are rigorous and well-explained, making the content accessible to graduate students with a background in linear algebra and optimization.

The core value lies in the explicit comparison between dynamic programming and Lyapunov analysis, emphasizing the trade-off between optimality and computational tractability. Tedrake correctly points out that while dynamic programming yields an optimal controller, it suffers from the curse of dimensionality and, with function approximation, loses all guarantees. In contrast, Lyapunov analysis, even with approximation, can provide certificates of stability and robustness if the inequalities are satisfied.

The lecture’s strength is its pedagogical approach: starting with the simplest discrete case, deriving the linear equations, and then showing how the same principles extend to more complex scenarios. The use of least squares and the mention of linear programming as a tool for finding Lyapunov functions are particularly valuable, as they connect the theory to practical algorithms.

One minor limitation is that the lecture does not delve into the specifics of how to formulate the linear program for Lyapunov functions, but it explicitly sets the stage for the next lecture. The focus on conceptual understanding over exhaustive technical detail is appropriate for a lecture format.

The sources cited are minimal (only the course website), but the content is based on established literature in optimal control and Lyapunov theory, which is implicitly referenced. The lecture’s quality is high, and it successfully conveys the importance of convex optimization in modern robotics.

Overall, this is an excellent lecture that provides deep insights into the computational aspects of control synthesis, making it a valuable resource for students and researchers in robotics and control.

The title accurately describes the content: a lecture from MIT's Underactuated Robotics course.

9/10

Lecture by a renowned MIT professor, part of a well-established course. Content is rigorous, mathematically grounded, and presented with clear derivations. The lecture is part of a series and references the course website for further materials.

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

• 00:00Introduction and recap of previous lecture on swing-up control.
• 03:00Comparison of dynamic programming and Lyapunov analysis.
• 10:00Derivation of the discrete-time Lyapunov inequality.
• 20:00Policy evaluation as a linear equation and its solution via least squares.
• 30:00Extension to function approximation and the issue of error accumulation.
• 40:00Discussion on how to ensure errors have the right sign for guarantees.
• 50:00Introduction to using linear programming for Lyapunov functions.
• 1:00:00Conclusion and preview of next lecture.

• Underactuated Robotics Course Website — Course materials and further resources.

• Underactuated Robotics Course Website — Course materials and further resources.