CS CS391R: Robot Learnin ing Perception, Decision Making, and - - PowerPoint PPT Presentation

▶
cs cs391r robot learnin ing
SMART_READER_LITE
LIVE PREVIEW

CS CS391R: Robot Learnin ing Perception, Decision Making, and - - PowerPoint PPT Presentation

Oct 01, 2023 865 likes •1.3k views

CS CS391R: Robot Learnin ing Perception, Decision Making, and General-Purpose Autonomy Prof. Yuke Zhu Fall 2020 CS391R: Robot Learning (Fall 2020) 1 Robotics and COVID-19 Photos from the Internet CS391R: Robot Learning (Fall 2020) 2

slide-1
SLIDE 1

CS391R: Robot Learning (Fall 2020)

CS CS391R: Robot Learnin ing

Perception, Decision Making, and General-Purpose Autonomy

1

  • Prof. Yuke Zhu

Fall 2020

slide-2
SLIDE 2

CS391R: Robot Learning (Fall 2020) 2

Photos from the Internet

Robotics and COVID-19

slide-3
SLIDE 3

CS391R: Robot Learning (Fall 2020) 3

Today’s Agenda

  • Overview of general-purpose robot autonomy
  • Why studying Robot Learning now?
  • Research topics of Robot Learning
  • Course content and logistics
  • Student introduction
slide-4
SLIDE 4

CS391R: Robot Learning (Fall 2020) 4

Special-Purpose Robot Automation

Structured Environments Fixed Set of Tasks Pre-programmed Procedures

slide-5
SLIDE 5

CS391R: Robot Learning (Fall 2020) 5

General-Purpose Robot Autonomy … in the Wild

Unstructured Environments Ever-changing Tasks Human Involvement

slide-6
SLIDE 6

CS391R: Robot Learning (Fall 2020) 6

Special-Purpose Robot Automation

custom-built robots human expert programming special-purpose behaviors

General-Purpose Robot Autonomy

general-purpose behaviors general-purpose robots

?

slide-7
SLIDE 7

CS391R: Robot Learning (Fall 2020) 7

Special-Purpose Robot Automation

custom-built robots human expert programming special-purpose behaviors

General-Purpose Robot Autonomy

general-purpose behaviors general-purpose robots

Robot Learning

slide-8
SLIDE 8

CS391R: Robot Learning (Fall 2020) 8

General-Purpose Robot Autonomy: Im Imagi ginati tions

Un Unim imate - The First st Indust strial Robot British TV (1968)

slide-9
SLIDE 9

CS391R: Robot Learning (Fall 2020) 9

General-Purpose Robot Autonomy: Ch Challe llenges

DARPA Robotics s Challenge (2015)

“The Moravec's paradox”

slide-10
SLIDE 10

CS391R: Robot Learning (Fall 2020) 10

General-Purpose Robot Autonomy: Progress ss

Grasp sping (DexNet 4.0; 2019) Locomot Locomotion

  • n (ANYmal; 2019)

Mani Manipul ulat ation

  • n (OpenAI; 2019)

We will learn the algorithms and techniques behind the latest progress.

slide-11
SLIDE 11

CS391R: Robot Learning (Fall 2020) 11

What Makes Ro Robot Learning Special?

Robots are physi ysically y embodied and envi vironmentally y si situated.

[Levine et al. JMLR 2016] [Bohg et al. ICRA 2018] [Sa et al. IROS 2014]

slide-12
SLIDE 12

CS391R: Robot Learning (Fall 2020) 12

What Makes Ro Robot Learning Special?

Robots are physi ysically y embodied and envi vironmentally y si situated.

[Levine et al. JMLR 2016] [Bohg et al. ICRA 2018] [Sa et al. IROS 2014] Perceive Act Perceive Act Act Perceive

slide-13
SLIDE 13

CS391R: Robot Learning (Fall 2020) 13

[Levine et al. JMLR 2016] [Bohg et al. ICRA 2018] [Sa et al. IROS 2014] Perceive Act Perceive Act Act Perceive

A key challenge in Ro Robo bot Learning is to close the pe percept ption-act action

  • n loop
  • op.
slide-14
SLIDE 14

CS391R: Robot Learning (Fall 2020) 14

Now is the best time to study and work on Robot Learning.

Recent breakthroughs in machine learning and computer vision, e.g., deep learning (Turing awards 2018)

Ar Artificial Intelligence Co Computin ing Powe wer

Your smartphone is millions of times more powerful than all of NASA’s combined computing in 1969.

Ro Robot Hardware

More reliable and affordable cobot hardware that costs around annual salary of American workers

slide-15
SLIDE 15

CS391R: Robot Learning (Fall 2020) 15

Now is the best time to study and work on Robot Learning.

https://www.therobotreport.com/tag/coronavirus/ Positive and negative so societal impacts of robot learning research is an important part of our in-class discussions.

slide-16
SLIDE 16

CS391R: Robot Learning (Fall 2020) 16

Robot Learning as a Growing Research Community

10 20 30 40 50 60 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Number of Papers (k) Year

Growth of “Ro Robot Learni Learning ng” Publications

[Source: Google Scholar]

Conference on Robot Learning is 3 years old. 6x

slide-17
SLIDE 17

CS391R: Robot Learning (Fall 2020) 17

Harnessing the priors and structures of a problem goes a long way… Learning is most effective when used in conjunction with modeling.

When NOT to Make Robots Learn?

Learning is not for every problem in robotics.

slide-18
SLIDE 18

CS391R: Robot Learning (Fall 2020) 18

When to Make Robots Learn?

  • bject variation

environment uncertainty adaptation Learning is critical for getting robots to work in the real world.

slide-19
SLIDE 19

CS391R: Robot Learning (Fall 2020) 19

You learn CS391R: Robot Learning so that Robots learn faster and better.

slide-20
SLIDE 20

CS391R: Robot Learning (Fall 2020) 20

Key Ingredients of General-Purpose Robot Autonomy

Perception

seeing and understanding 3D environments

Decision Making

planning and control for long-term interactions

Intelligence

cognitive reasoning & fast adaptation to new situations physical embodiment & hardware constraints

Real-World Systems

slide-21
SLIDE 21

CS391R: Robot Learning (Fall 2020) 21

We review the Robot Learning literature in these topics.

Perception

seeing and understanding 3D environments

Part I Decision Making

planning and control for long-term interactions

Part II Intelligence

cognitive reasoning & fast adaptation to new situations

Part III Real-World Systems

physical embodiment & hardware constraints

Part IV

Course Content

Prerequisi site: coursework / experience in AI and Machine Learning

slide-22
SLIDE 22

CS391R: Robot Learning (Fall 2020) 22

Perception

seeing and understanding 3D environments

Part I Decision Making

planning and control for long-term interactions

Part II Intelligence

cognitive reasoning & fast adaptation to new situations

Part III

physical embodiment & hardware constraints

Part IV

Course Content

Real-World Systems

slide-23
SLIDE 23

CS391R: Robot Learning (Fall 2020) 23

Course Content: Pe Percepti ption

  • bject detection

3d point cloud unsupervised visual learning multimodal understanding recursive state estimation pose estimation visual tracking interactive perception

slide-24
SLIDE 24

CS391R: Robot Learning (Fall 2020) 24

Perception

seeing and understanding 3D environments

Part I Decision Making

planning and control for long-term interactions

Part II Intelligence

cognitive reasoning & fast adaptation to new situations

Part III Real-World Systems

physical embodiment & hardware constraints

Part IV

Course Content

slide-25
SLIDE 25

CS391R: Robot Learning (Fall 2020) 25

Course Content: Decisi sion Maki king

imitation as supervised learning adversarial imitation learning inverse reinforcement learning model-free reinforcement learning model-based reinforcement learning

slide-26
SLIDE 26

CS391R: Robot Learning (Fall 2020) 26

Perception

seeing and understanding 3D environments

Part I Decision Making

planning and control for long-term interactions

Part II Intelligence

cognitive reasoning & fast adaptation to new situations

Part III Real-World System

physical embodiment & hardware constraints

Part IV

Course Content

slide-27
SLIDE 27

CS391R: Robot Learning (Fall 2020) 27

Course Content: In Inte tellige gence

learning to learn: meta-learning learning to learn: lifelong learning compositionality: hierarchy compositionality: task and motion causal reasoning

slide-28
SLIDE 28

CS391R: Robot Learning (Fall 2020) 28

Perception

seeing and understanding 3D environments

Part I Decision Making

planning and control for long-term interactions

Part II Intelligence

cognitive reasoning & fast adapting to new situations

Part III Real-World Systems

physical embodiment & hardware constraints

Part IV

Course Content

slide-29
SLIDE 29

CS391R: Robot Learning (Fall 2020) 29

Course Content: Syst ystems

simulation-reality gap data-driven robotic grasping building robotic systems

slide-30
SLIDE 30

CS391R: Robot Learning (Fall 2020) 30

Learning Objectives

  • understand the potential and societal impact of general-purpose robot autonomy in the

real world, the technical challenges arising from building it, and the role of machine learning and AI in addressing these challenges;

  • get familiar with a variety of model-driven and data-driven principles and algorithms on

robot perception and decision making;

  • be able to evaluate, communicate, and apply advanced AI-based techniques to

robotics problems. … through literature reviews, research presentations, and course projects

slide-31
SLIDE 31

CS391R: Robot Learning (Fall 2020) 31

Learning Objectives

Get a taste of Robot Learning research in the full circle

slide-32
SLIDE 32

CS391R: Robot Learning (Fall 2020) 32

Logistics

Lectures

Time: 2:00-3:30pm CT, Tuesdays and Thursdays Location: Online (Zoom links on Canvas)

Office Hours

Instructor: TBA next week TA: TBA next week Fill in the time zone survey on Canvas!

slide-33
SLIDE 33

CS391R: Robot Learning (Fall 2020) 33

Logistics

Instructor Lectures

  • verview of research topics

Student Presentations presentation of research papers Final Project Spotlights spotlight talks of course projects https://www.cs.utexas.edu/~yukez/cs391r_fall2020/

slide-34
SLIDE 34

CS391R: Robot Learning (Fall 2020) 34

Logistics

Required Readings Optional Readings key papers that will be discussed in class recommended papers for in-depth reviews Required Readings (No Review)

  • verview or survey papers with lectures
slide-35
SLIDE 35

CS391R: Robot Learning (Fall 2020) 35

Logistics

Grading Policy Student presentation (20%) Paper reviews (30%) Course project (40%) In-class participation (10%) 20% each

  • At least one presentation for each student (chances to do more)
  • Length: 25min (± 1min) + 5min Q&A
  • Format: problem formulation, technical approach, results, … (see

slide template for more details)

  • Followed by 5-10min in-class discussions
  • Email the slides to the TA and the instructor seven days (EOD)

prior to the presentation date

  • Presentation recordings posted in Canvas (protected under

FERPA)

  • Breakout rooms and in-class discussions will NOT be recorded.
slide-36
SLIDE 36

CS391R: Robot Learning (Fall 2020) 36

Logistics

Grading Policy Student presentation (20%) Paper reviews (30%) Course project (40%) In-class participation (10%) 1.5% each x 20 reviews

  • Due by 9:59pm the previous night of each student presentation
  • Write a review for one paper from the required readings (2 choices

for each class)

  • Online review form in R:SS format
  • No late date - but more than 20 presentation classes (feel free to

skip some)

  • Have energy to do more? Top-scored 20 for grading
  • Class attendance and participation is required for review grades
slide-37
SLIDE 37

CS391R: Robot Learning (Fall 2020) 37

Logistics

1.5% each x 20 reviews

  • Due by 9:59pm the previous night of each student presentation
  • Write a review for one paper from the required readings (2 choices

for each class)

  • Online review form in R:SS format
  • No late date - but more than 20 presentation classes (feel free to

skip some)

  • Have energy to do more? Top-scored 20 for grading
  • Class attendance and participation is required for review grades

Online Learning Considerations

For students studying online in a different time zone, they may submit their written responses to the discussion questions on Canvas within 48hrs after the class as an alternative to in-class attendance.

slide-38
SLIDE 38

CS391R: Robot Learning (Fall 2020) 38

Logistics

Grading Policy Student presentation (20%) Paper reviews (30%) Course project (40%) In-class participation (10%) 40%

  • Project Proposal (5%). Due Thu Sept 17.
  • Project Milestone (5%). Due Thu Oct 22.
  • Final Report (25%). Due Fri Dec 11.
  • Spotlight Talk (5%). Week 15.

+

Hands-on experience of robot learning research

slide-39
SLIDE 39

CS391R: Robot Learning (Fall 2020) 39

Logistics

Grading Policy Student presentation (20%) Paper reviews (30%) Course project (40%) In-class participation (10%)

More details on the website! Tutorials, computing resources, potential project ideas, …

slide-40
SLIDE 40

CS391R: Robot Learning (Fall 2020) 40

Logistics

Grading Policy Student presentation (20%) Paper reviews (30%) Course project (40%) In-class participation (10%)

Tutorials, computing resources, potential project ideas, …

Do Robot Learning with Simulated Environments

slide-41
SLIDE 41

CS391R: Robot Learning (Fall 2020) 41

Logistics

Grading Policy Student presentation (20%) Paper reviews (30%) Course project (40%) In-class participation (10%)

slide-42
SLIDE 42

CS391R: Robot Learning (Fall 2020) 42

Tell Us About Yourself