- The robot hand can now solve a Rubik’s cube in simulation:
https://drive.google.com/a/openai.com/file/d/0B60rCy4P2FOIenlLdzN2LXdiOTQ/view?usp=sharing (needs OpenAI login)
Physical robot will do same in September
- 1v1 bot is no longer exploitable
It can no longer be beaten using “unconventional” strategies
On track to beat all humans in 1 month
- Athletic competitive robots:
https://drive.google.com/a/openai.com/file/d/0B60rCy4P2FOIZE4wNVdlbkx6U2M/view?usp=sharing (needs OpenAI login)
- Released an adversarial example that fools a camera from all angles simultaneously:
- DeepMind’s directly used one of our algorithms to produce their parkour results:
DeepMind’s results: https://deepmind.com/blog/producing-flexible-behaviourssimulated-environments/
DeepMind’s technical papers explicitly state they directly used our algorithms
Our blogpost about our algorithm: https://blog.openai.com/openai-baselines-ppo/ (DeepMind used an older version).
- Coming up:
Designing the for-profit structure
Negotiate merger terms with Cerebras
More due diligence with Cerebras