Lecture 16 | MIT 6.832 (Underactuated Robotics), Spring 2018

The lecture provides a rigorous and insightful introduction to global motion planning, focusing on the use of mixed-integer optimization to overcome the limitations of local trajectory optimization. The instructor, Russ Tedrake, is a leading expert in the field, and the content reflects a deep understanding of both theoretical foundations and practical implementation. The explanation of the Big-M method is clear and well-illustrated, demonstrating how logical disjunctions can be encoded in a mixed-integer convex framework. The discussion of branch-and-bound provides a solid intuition for how solvers achieve global optimality, balancing search and convex relaxation. The lecture is technically dense, assuming familiarity with convex optimization and trajectory optimization, but it is well-structured and accessible to advanced students. The main strength is the clarity of the exposition, with concrete examples and a logical progression from problem formulation to solution strategies. The lecture also appropriately acknowledges the computational complexity (NP-hardness) and the practical limitations of mixed-integer programming, while highlighting its effectiveness for moderate-sized problems. The content is accurate and aligns with established literature in robotics and optimization. However, the lecture does not delve into the details of sampling-based methods, which are mentioned as an alternative but not covered in depth. Additionally, the lack of citations to specific papers or resources beyond the course website is a minor weakness, though the material is standard and well-known. Overall, this is a high-quality lecture that effectively conveys key concepts in global motion planning, suitable for graduate-level study.

The title accurately reflects the content: a lecture on underactuated robotics, specifically focusing on global motion planning via mixed-integer optimization.

8/10

Lecture by MIT professor Russ Tedrake, part of a well-established course on underactuated robotics. Content is technically rigorous, based on established optimization methods (mixed-integer programming, sampling-based planning). No external sources cited beyond course website, but the material is standard and presented accurately.

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

• 00:00Introduction: recap of previous lectures, motivation for global motion planning, and outline of the lecture.
• 03:00Discussion of limitations of local trajectory optimization: local minima and false infeasibility.
• 07:00Introduction to two approaches: explicit decomposition of nonconvexities and sampling-based methods.
• 10:00Explanation of convex trajectory optimization for linear systems and convex constraints.
• 15:00Nonconvexity introduced by obstacle avoidance: polygons and logical OR constraints.
• 20:00Introduction to mixed-integer optimization: binary variables and Big-M method.
• 25:00Detailed example of encoding logical OR constraints using Big-M.
• 30:00Discussion of the nonconvex nature of binary constraints and the role of commercial solvers.
• 35:00Intuition behind branch-and-bound: combining search with convex optimization.
• 40:00Conclusion and transition to next lecture topics.

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

• Underactuated Robotics Course Website — The course website provides lecture notes and materials that align with the content presented in this lecture.