Lecture 22 | MIT 6.881 (Robotic Manipulation), Fall 2020 | Planning Under Uncertainty, Course Summary

Item: Lecture 22 | MIT 6.881 (Robotic Manipulation), Fall 2020 | Planning Under Uncertainty, Course Summary
Rating: 5
Author: PostSingularity Institute

by Russ Tedrake

Applied Sciences & Engineering Control & Robotics TJFMAutomatic control engineering TJFM1Robotics

🎙 Russ Tedrake 👥 17K 📅 December 4, 2020 ⏱ 89 min 👁 2K 📄 lecture 🧭 2026-08-05

Available in: English (current) Français

Keywords

planning under uncertaintystochastic optimizationrobustnesstrajectory optimizationrobotic manipulation

Summary

This is the final content lecture of MIT 6.881 on robotic manipulation, taught by Russ Tedrake. The lecture focuses on planning under uncertainty, building on previous discussions of contact-rich manipulation. Tedrake explains how stochastic optimization can smooth discontinuities in gradients when contact is made or broken, by considering expected values over distributions of trajectories. He discusses the limitations of using expected value as a robustness metric, citing John Doyle’s famous paper on LQG margins. He then illustrates these concepts with a book-pushing example, where uncertainty in friction and initial conditions requires robust planning. The lecture emphasizes the importance of optimizing over multiple initial conditions and using risk-sensitive objectives. Tedrake also provides a course summary, highlighting key topics covered and future research directions in manipulation, such as machine learning and contact-rich planning. He mentions the textbook and slides available online. The lecture concludes with logistics for the final presentations.

148 words

Critical Evaluation

This lecture provides a high-quality, in-depth exploration of planning under uncertainty in robotic manipulation. Russ Tedrake, a renowned expert in the field, delivers a clear and rigorous presentation that builds on previous lectures. The content is scientifically sound, with references to relevant literature, such as John Doyle’s paper on LQG margins, and practical examples like the book-pushing experiment. The argumentation is solid, explaining the benefits and limitations of stochastic optimization for robustness. The lecture effectively bridges theory and practice, discussing both mathematical formulations and implementation details. The sources cited are credible and directly relevant. The title accurately reflects the content, as the lecture covers planning under uncertainty and provides a course summary. The presentation is well-structured, with a logical flow from concepts to examples. The technical level is appropriate for an advanced undergraduate or graduate course, but the explanations are accessible to those with a background in robotics and optimization. Overall, this is an excellent lecture that offers valuable insights into current research and future directions in manipulation.

168 words

Title / Content Match

The title accurately reflects the content: a lecture on planning under uncertainty and a summary of the course.

Quality & Reliability

9/10

Lecture by a leading MIT professor, based on established research and textbook, with clear explanations and references to academic work.

Key Moments

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

Cited Sources

Robotic Manipulation Textbook — Course textbook website with additional resources
Lecture Slides — Live slides used during the lecture

Concurring Sources

Guaranteed Margins for LQG Regulators — John Doyle's paper on LQG margins, cited in the lecture

Contribution & Novelties

This lecture provides a comprehensive overview of planning under uncertainty in robotic manipulation, emphasizing stochastic optimization and robustness. It offers a clear explanation of how expected value can smooth contact discontinuities and discusses the limitations of this approach, citing John Doyle’s work. The lecture also highlights the importance of optimizing over distributions of initial conditions and using risk-sensitive objectives. For further exploration, consider the following:

Risk-Sensitive Reinforcement Learning — Discusses objectives beyond expected value.
Conditional Value at Risk — A risk metric often used in optimization.
Trajectory Optimization — Core technique for planning in robotics.

95 words

Radar Profile

The radar profile shows high scores across all dimensions, indicating a well-rounded lecture with strong information content, technical depth, and reliability.

Reliability 9/10