Episode 28: Sergey Levine, UC Berkeley, on the bottlenecks to generalization in reinforcement learning, why simulation is doomed to succeed, and how to pick good research problems

1

Introduction

00:00 • 2min

2

Computer Graphics Research

01:38 • 2min

3

Postdocs - What Happened After That?

03:12 • 2min

4

Is There a Future for Computer Animation?

04:58 • 2min

5

I Think Jason Pang's Robotics Demo Is the Work

07:25 • 3min

6

Using Convolutional Network Techniques to Control Robot Motion

10:47 • 4min

7

The Challenges of N10 Neural Nets in Robotics

14:30 • 3min

8

Deep Learning

17:02 • 2min

9

How to Scale Up Deep Robotic Learning

19:09 • 2min

10

The Scale of the Robotics Data Set

21:34 • 2min

11

The Importance of Large Data Sets in Robotics

23:08 • 2min

12

Robots That Are Like Humans?

25:24 • 2min

13

Goal Condition DRL for Robotic Control

27:06 • 4min

14

Is There a Future for Vision Language Models?

31:06 • 2min

15

Is There a Higher Level Goal?

32:49 • 3min

16

Picking Research Questions

35:27 • 2min

17

Machine Learning Systems That Emulate Humans Through Imitation Learning Through Supervised Learning

37:07 • 2min

18

Offline Reinforcement Learning - What's Next?

39:14 • 2min

19

Reinforcement Learning - The Basic Problem With Offline RL

41:08 • 2min

20

Is There a Principle to It?

42:47 • 2min

21

Offline Deep Learning

45:10 • 4min

22

The Intuition for Regression Learning (RL)

49:08 • 2min

23

Large Scale Regression Learning (RL) in Atari Games

50:40 • 2min

24

How to Improve Language Models Through Value-Based Reinforcement

52:30 • 2min

25

Is There a Middle Ground to Scaling Up Reaper Learning?

54:26 • 3min

26

The Importance of Robotics Versus Not?

56:56 • 2min

27

Is It Sufficient to Just Build Larger Language Models?

58:38 • 3min

28

Is There an Alternative to Offline RL?

01:01:24 • 3min

29

Using Language Models to Build Robotic Robotic Systems

01:03:54 • 2min

30

General and Reusable Robotic Control Models

01:06:02 • 3min

31

Is Lifelong Robotic Learning Really the New Normal?

01:08:42 • 2min

32

What's the Most Influential Research in Robotics?

01:10:42 • 3min

33

The Relationship Between Inference and Sequence Models and Control

01:14:11 • 2min

34

Metal Learning

01:16:22 • 3min

35

Predictive Modeling

01:19:19 • 2min

36

The Robotics Experiment at Google

01:21:46 • 2min

37

Is There a Difference Between Intrinsic Motivation and Reinforcement Learning?

01:23:23 • 2min

38

Is There a Future in Robotic Learning?

01:25:04 • 4min

39

Is Language a Prerequisite for Artificial Intelligence?

01:29:33 • 2min

40

Robotics Research - Language Models Hit Walls

01:31:32 • 2min

41

Generally Intelligent Podcast

01:33:03 • 2min