Prabhat Nagarajan

icon indicating a website link prabhatnagarajan.com

icon company The University of Alberta icon location Edmonton, Canada icon location Reinforcement Learning PhD student at UAlberta

Results 25 comments of Prabhat Nagarajan

Specify successful configurations for examples

Shouldn't the default configuration work well?

[WIP] Implements Hindsight Experience Replay

![herFetchReach-v1](https://user-images.githubusercontent.com/10005453/59177781-124f7d80-8b98-11e9-88c2-a70d34b1fd8a.png) ![herFetchPush-v1](https://user-images.githubusercontent.com/10005453/59177783-1380aa80-8b98-11e9-8119-4b4c0ce597d7.png) ![herFetchPickAndPlace-v1](https://user-images.githubusercontent.com/10005453/59177787-154a6e00-8b98-11e9-85ab-fbbb1a16e37d.png) ![herFetchSlide-v1](https://user-images.githubusercontent.com/10005453/59177789-167b9b00-8b98-11e9-898b-a984403f863e.png)

Hindsight Experience Replay Buffer

/test

Hindsight Experience Replay Buffer

/test

Hindsight Experience Replay Buffer

/test

Hindsight Experience Replay Buffer

/test

Hindsight Experience Replay Buffer

@ummavi also mentioned this in person. The main point of this PR is to add the HER feature for users, not to reproduce the Hindsight Replay paper, which has a...

Hindsight Experience Replay Buffer

I'm not sure what you mean by `just "works"'. I've shown that it works on the bit-flip domain used in the paper (though they don't provide code or hyperparameters for...

Hindsight Experience Replay Buffer

/test

Hindsight Experience Replay Buffer

/test