Главная
Study mode:
on
1
Intro
2
Comparison with Dactyl system
3
High-level overview
4
Tasks Rubik's cube and block reorientation
5
Physical system overview
6
Reading angles from the cube electronics
7
Realistic modeling of the system in simulation
8
Automatic Domain Randomization ADR
9
Cube size randomization during training blog
10
Entropy and rand param probability distribution
11
ADR pseudocode
12
Rapid
13
Randomizations
14
PPO
15
Actions and rewards
16
Policy network, embed and add
17
Behavioural cloning
18
Vision pipeline
19
Focal loss
20
Results
21
Perturbation robustness
22
Meta-learning
23
Predicting environment variables from LSTM hidden state
Description:
Explore OpenAI's groundbreaking research on solving the Rubik's Cube with a robotic hand in this comprehensive 55-minute video explanation. Delve into the intricacies of the system, from its physical components to the advanced machine learning techniques employed. Learn about Automatic Domain Randomization (ADR), the vision pipeline, and how the robot achieves perturbation robustness. Gain insights into the policy network architecture, behavioral cloning, and meta-learning aspects of this innovative project. Understand the challenges and solutions in training a robotic hand purely through simulation to manipulate real-world objects with precision.

OpenAI - Solving Rubik's Cube with a Robot Hand - RL Paper Explained

Aleksa Gordić - The AI Epiphany
Add to list