Skip to content
Benchmarking Canonical Evolution Strategies for Playing Atari
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
configurations
networks
src
submit Add model with Qbert bug Mar 1, 2018
.gitignore Initial commit Feb 22, 2018
QbertBug_action_names.txt
README.md
main.py Initial commit Feb 22, 2018
viz.py

README.md

Canonical ES for benchmarking Atari Arxiv Paper: https://arxiv.org/abs/1802.08842

Code based on: https://github.com/openai/evolution-strategies-starter

Atari library version 0.9.1 : https://gym.openai.com/envs/#atari

QBert bug: https://www.youtube.com/watch?v=meE5aaRJ0Zs&feature=em-comments

QBert bug (fails to exploit):

https://www.youtube.com/watch?v=XoRV2rxkFG8&feature=youtu.be

All games (Cherry picked solutions): https://www.youtube.com/watch?v=0wDzPBiURSI

Run viz.py for evaluation of the policy that finds a Qbert bug. Rerun it until the bug appears. For evaluation of 30 runs it appeared in 8.

You can’t perform that action at this time.