Real-World Robot Learning

Spring 2025. ESE 6800/CS7000. Tue / Thu 10:15-11:45. AGH 105.

Image

Announcements

Welcome!

Jan 16 · 0 min read

Hi! We hope that you are as excited as us to start this new course!

Course Overview

Decision-making has been a cornerstone of artificial intelligence since the field’s inception in the 1960s. While the techniques and algorithms have evolved dramatically over time, the fundamental challenge remains: how to make intelligent decisions in the presence of uncertainty. Over the past six decades, this research has led to the development of highly advanced systems, with some achieving superhuman performance in cognitively demanding tasks like Go, Atari, Gran Turismo, Chess, StarCraft, and SOTA.

However, despite these remarkable successes, most of these systems excel only in controlled, simulated, or game-based environments. Why haven’t the same methods translated seamlessly to real-world decision-making, such as controlling a physical robot to perform household tasks? What makes real-world environments so uniquely challenging? And what recent advances are pushing the boundaries of what’s possible in real-world applications?

This course offers a structured framework to explore these questions. We will study techniques for learning-based decision-making, such as imitation learning and reinforcement learning, focusing on their practical challenges when applied in real-world scenarios. Through a combination of lectures, student presentations, hands-on projects, and guest presentations from leading experts in the field, students will gain a deep understanding of the state-of-the-art decision-making systems and their challenges when applied to robotics.

Prerequisites

This is a graduate-level course. Students are expected to have prior knowledge in deep learning and robotics, such as the topics covered in Robot Learning (ESE 650), Principles of Deep Learning (ESE 546), Applied Machine Learning (CIS 5190), and Introduction to Robotics (MEAM-520).

Schedule (Tentative)

The general idea behind this course is the following. We will start by going off the beaten arxiv track and read oldy but goldy papers. Then, we will use such papers to understand the roots of more recent works. Note that this schedule will evolve during the course.

Overture

Jan. 16
Intro + Overview
Syllabus
Jan. 21
Supervised Learning (Review)
Jan. 23
How to use Pytorch (Review)
Jan. 28
Reinforcement Learning Part I (Review)
Jan. 30
Reinforcement Learning Part II (Review)

Foundations

Feb. 4
Robot Perception I
Paper Reading
Feb. 6
Robot Perception II
Paper Reading
Feb. 11
The Development Perspective on Robot Learning I
Paper Reading
Feb. 13
The Development Perspective on Robot Learning II
Paper Reading

Data

Feb. 18
Learning from real-world robot data I
Paper Reading
Feb. 20
Learning from real-world robot data II
Project Proposal Due Paper Reading
Feb. 25
Learning from others’ data I
Paper Reading
Feb. 27
Learning from others’ data II
Paper Reading
Mar. 4
Learning from Internet Data I
Paper Reading
Mar. 6
Learning from Internet Data II
Paper Reading
Mar. 11
No Class Spring Break 🏝️
Mar. 13
No Class Spring Break 🏝️
Mar. 18
Learning from Synthetic Data I
Paper Reading
Mar. 20
Learning from Synthetic Data II
Project Midterm Report Due Paper Reading
Mar. 25
Lifelong Learning I
Paper Reading
Mar. 27
Lifelong Learning II
Paper Reading

Frontiers

Apr. 1
Guest Lecture
Apr. 3
Challenges for the field I
Paper Reading
Apr. 8
Guest Lecture
Apr. 10
Challenges for the field II
Paper Reading
Apr. 15
Challenges for the field III
Paper Reading
Apr. 17
Grand Finale
Paper Reading

Project Presentations

Apr. 22
Project Presentations
Presenters TBD
Apr. 24
Project Presentations
Slides Due 11:59 pm ET, April 24 Presenters TBD
Apr. 29
Best Presentation Awards 🍨
Project Report Due May 1

Instructors

Avatar
Avatar

Teaching Assistants

Avatar

Robots that learn, UC Berkeley.

Visual Scene Understanding, UC Berkeley.

Embodied AI Safety, CMU. This course is not only very interesting, but also has an awesome webpage.