### Announcements

2021.03.01: Apply for enrollment Here.

### General Information

#### Times & Places

TuTh 5:00PM - 6:20PM

#### Course Staff

Name | Office Hours | Location | ||
---|---|---|---|---|

Instructor | Hao Su | haosu@eng.ucsd.edu | Tu 4:00PM - 5:00PM | Remote |

Course Assistant | Minghua Liu | mil070@eng.ucsd.edu | TBD | Remote |

#### Overview

This is a course for senior undergrads and graduate students, covering core concepts and algorithms in classical robotics and the more modern learning-based methods for robotics. **We assume that the course takers have already taken certain deep learning for perception courses** (e.g., by taking the *Machine Learning meets Geometry (CSE291-I00)* in the winter quarter), and are interested in how to train a robot that can interact with the physical world by machine learning methods. The first half of this course covers basic concepts and algorithms of robotics, and the second half introduces the basic concepts, algorithms, and research trends of reinforcement learning.

One feature of this course is that, we will instruct the students to build an armed robot in a simulated virtual environment through programming assignments. For the final project, we ask students to compete in a table-top object organization challenge using the built robot.

#### Prerequisites

- Strong background in calculus and linear algebra.
- Project experience in 2D/3D object recognition using deep learning.
- Familiar with Newtonian mechanics.
- Proficient with Python.
- Experience in physical simulation is a plus.

#### Enrollment

Fill in the Google form to apply for enrollment.#### Grading (tentative)

- Class presence: 10%
- Homework: 60%
- Final project: 30%

#### Syllabus

The planned syllabus is as below. Certain contents may be added or removed based upon the interactions in class and other situations.

- Robotics
- SE(3) Geometry
- Robot kinematics
- Robot-Object Interaction
- Optimal Control
- Physical Simulation

- Reinforcement Learning
- Concepts of RL
- RL as Optimization
- Long-horizon RL
- Generalizable RL

#### Acknowledgements

Thank Sapien for support.