6.8210 Spring 2024 Lecture 23: Feedback motion planning

The lecture provides a comprehensive and rigorous introduction to feedback motion planning, a key concept in robotics. The instructor, Russ Tedrake, is a leading expert in the field, and his presentation is both technically deep and pedagogically clear. The content is well-structured, starting with motivation, then building on classical AI planning, and finally detailing the mathematical framework of funnels and Lyapunov functions. The use of the juggling example is effective in illustrating the concepts, and the derivation of the dynamics and the funnel composition is thorough. The lecture also touches on modern developments, such as the use of LLMs for task planning, which adds relevance. The sources cited, including the seminal work by Burridge, Rizzi, and Koditschek, are appropriate and credible. The lecture does not include any commercial bias or advertising. The title accurately reflects the content. Overall, this is a high-quality lecture that would be valuable for students and practitioners in robotics and control.

The title accurately reflects the lecture's focus on feedback motion planning, a core topic in robotics.

8/10

Lecture from MIT's Underactuated Robotics course, presented by a leading expert. Content is technically rigorous, builds on established theory (Lyapunov, funnels), and includes references to classical and recent work. No commercial bias detected.

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

• 00:00Introduction and logistics for final project
• 02:00Motivation: high-level planning in robotics, dishwasher example
• 05:00Action primitives and software abstraction
• 08:00Historical context: STRIPS and PDDL
• 10:00Using LLMs for task planning (code as policies)
• 12:00Core idea: feedback motion planning, funnels and Lyapunov functions
• 15:00Juggling robot example: dynamics and control
• 20:00Sequential composition of funnels
• 30:00Verification of funnels using sums-of-squares
• 40:00Extensions to stochastic systems and robustness

• Sequential Composition of Dynamically Dexterous Robot Behaviors — Referenced as the foundational paper on sequential composition of funnels.
• PDDL - The Planning Domain Definition Language — Mentioned as a standard language for task planning.
• Code as Policies: Language Model Programs for Embodied Control — Referenced as a recent approach using LLMs for robot control.

• Underactuated Robotics — Course website with lecture notes and additional materials.