Lecture 14 | MIT 6.832 (Underactuated Robotics), Spring 2018

The lecture provides a valuable overview of practical approaches to humanoid robot locomotion, drawing on real-world experience from the DARPA Robotics Challenge. The speakers, Robin Deits and Thon Kuhlman, are credible experts with substantial research contributions. The content is well-structured, starting with the problem’s complexity and then proposing a hierarchical decomposition: footstep planning, simplified dynamics, and execution. This approach is pragmatic and aligns with common practices in the field. The discussion of limitations of LQR and sum-of-squares for high-dimensional systems is accurate, and the suggestion to use trajectory optimization and MPC is reasonable, though the lecture does not delve into specific algorithmic details. The concept of reachability as an inner approximation is clearly explained, and the use of action sets for discrete search is a practical simplification. However, the lecture lacks depth in certain areas, such as the specifics of implementing MPC for humanoids or handling contact dynamics. The sources cited are minimal, primarily the course website, which limits the ability to verify claims independently. The title accurately reflects the content, and the lecture is suitable for an advanced undergraduate or graduate audience. Overall, the lecture is informative and provides a solid foundation for understanding humanoid locomotion, but it could benefit from more technical depth and references.

Title accurately reflects content: a lecture on underactuated robotics focusing on humanoid locomotion.

8/10

Lecture by experienced researchers from MIT's lab, presenting established methods (LQR, MPC, trajectory optimization) applied to humanoid robotics. Content is technically sound, but limited depth due to lecture format. No external sources cited beyond course website.

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

• 00:00Introduction by Russ Tedrake, introducing guest lecturers Robin Deits and Thon Kuhlman.
• 01:00Robin Deits begins discussing the DARPA Robotics Challenge and the challenges of humanoid robotics.
• 03:00Discussion of the high-dimensional, nonlinear, hybrid nature of humanoid robots.
• 05:00Evaluation of LQR and sum-of-squares for humanoid control, concluding they are insufficient.
• 07:00Introduction of trajectory optimization and model predictive control as potential solutions.
• 09:00Proposal of divide-and-conquer approach: footstep planning, simple dynamics, and execution.
• 11:00Explanation of footstep planning and the concept of reachability.
• 13:00Discussion of using action sets for discrete search in footstep planning.
• 15:00Mention of terrain mapping and the need for conservative approximations.
• 17:00Conclusion and transition to next part of lecture.

• Underactuated Robotics Course Website — Course website for MIT 6.832, providing lecture notes and additional resources.

• Underactuated Robotics Course Website — Course materials likely align with the lecture content.