Real-Robot Experiments
This section compares our synchronous and asynchronous implementations of ME-PPO, ME-TRPO, and MB-MPO and each of their respective performance in completing three tasks: Reaching, Shape Matching, and Lego Stacking. Further details can be found in the paper.
Reaching
Reaching
![](https://www.google.com/images/icons/product/drive-32.png)
ME-PPO
ME-PPO
![](https://www.google.com/images/icons/product/drive-32.png)
ME-TRPO
ME-TRPO
![](https://www.google.com/images/icons/product/drive-32.png)
MB-MPO
MB-MPO
Shape Matching
Shape Matching
![](https://www.google.com/images/icons/product/drive-32.png)
ME-PPO
ME-PPO
![](https://www.google.com/images/icons/product/drive-32.png)
ME-TRPO
ME-TRPO
![](https://www.google.com/images/icons/product/drive-32.png)
MB-MPO
MB-MPO
Lego Stacking
Lego Stacking
![](https://www.google.com/images/icons/product/drive-32.png)
ME-PPO
ME-PPO
![](https://www.google.com/images/icons/product/drive-32.png)
ME-TRPO
ME-TRPO
![](https://www.google.com/images/icons/product/drive-32.png)
MB-MPO
MB-MPO