drawing

ECE 720B01 Robot Learning: Principles and Advances

Instructors:
Professor: Jun Jin
TAs: TBD

Lectures: Mondays, 14:00 - 15:20, ECERF W6-087
Wednesdays, 9:30~10:50 (NOTE this new change), ECERF W6-087

Office Hours: Wednesdays, 11:00 - 11:50, 11-365 Donadeo Innovation Centre For Engineering

Communication:
eClass is our primary digital channel for communication about the course. To sign up, go to Bear Tracks and enroll with “ECE 720 B01 Advanced Topics in Software Engineering and Intelligent Systems”

Announcements

Prerequisites

  • Familiarity with mathematical proofs, probability, algorithms, and linear algebra; ability to implement algorithmic ideas in code.
  • Key Machine Learning courses that can build your background: ECE 447, ECE 449, COMPUT 466/566. If your foundations are rusty, we recommend you refresh them through those courses.
  • All graduate students are welcome to register.

Class Goals

  • Learn the math and algorithms in modern machine learning based robotic systems.
  • Grasp an overall picture of state-of-the-art literature in robot learning, with a special focus on robotic reinforcement learning (RL), robot learning from human preference, multi-task learning and continual learning (CL) and foundation models (LLMs, decision transformers) for robotics.
  • An intended side-effect of the course is to generally strengthen your expertise in these areas.

  • Be able to understand research papers in the field of robotics:

    • Main conferences: ICRA, IROS, RSS, CoRL, NeurIPS, ICML, ISER, ISRR.
    • Main journals: IJRR, T-RO, Autonomous Robots, RA-L.
  • Try out some ideas/extensions of your own.
  • Note: the focus of this course is on math and algorithms. We will not study mechanical or electrical design of robots.

Grading

  • Assignments (45%)

    • There are 4 assignments.
    • Each assignment will either include 1~2 short-answer questions or 1 coding task (<10 lines).
    • We'll drop your lowest scoring assignment.
    • Your remaining 3 (highest scoring) assignments will each equally contribute 45/3 of the points.

  • Open-ended final project (55%)

    • 1~2 students per project.
    • Implement a learning algorithm (e.g., imitation learning, RL) of your choice to solve the robotic grasping problem.
    • We’ll provide a code repo with (1) a robotic simulator ready for training, (2) a standard openai-gym environment ready for plug-and-play using existing opensource ML algorithm codes (e.g., the StableBaselines)
    • Extra points will be awarded to solutions that explore more advanced topics, e.g., inverse reinforcement learning (IRL), generalization, meta-learning, motion primitives, LLMs, etc.
    • 1/5 quality of presentation, 1/5 quality of writing of the final paper, 3/5 quality of the results themselves.

Syllabus and materials

All slides will be released on the website after each lecture. All slides are open to the public and free to download.