OpenAI updates OpenAI-updates

Generalizing from simulation Generalizing from simulation

Read paper (dynamics randomization)(opens in a new window)Read paper (image-based learning)(opens in a new window) Read paper (dynamics randomization)(opens in a new window)Read paper (image-based learning)(opens in a new window)

Sim-to-real transfer of robotic control with dynamics randomization Sim-to-real transfer of robotic control with dynamics randomization

Read paper(opens in a new window) Read paper(opens in a new window)

Asymmetric actor critic for image-based robot learning Asymmetric actor critic for image-based robot learning

(opens in a new window) (opens in a new window)

Domain randomization and generative models for robotic grasping Domain randomization and generative models for robotic grasping

Read paper(opens in a new window) Read paper(opens in a new window)

Competitive self-play Competitive self-play

View code(opens in a new window)Read paper(opens in a new window) View code(opens in a new window)Read paper(opens in a new window)

Meta-learning for wrestling Meta-learning for wrestling

View code(opens in a new window)Read paper(opens in a new window) View code(opens in a new window)Read paper(opens in a new window)

Nonlinear computation in deep linear networks Nonlinear computation in deep linear networks

Learning to model other minds Learning to model other minds

Read paper(opens in a new window)(opens in a new window) Read paper(opens in a new window)(opens in a new window)

Learning with opponent-learning awareness Learning with opponent-learning awareness

Read paper(opens in a new window) Read paper(opens in a new window)

OpenAI Baselines: ACKTR & A2C OpenAI Baselines: ACKTR & A2C

Read code(opens in a new window)Read paper(opens in a new window) Read code(opens in a new window)Read paper(opens in a new window)

More on Dota 2 More on Dota 2

Our Dota 2 result shows that self-play can catapult the performance of machine learning systems from far below human level to superhuman, given sufficient compute. In the span of a month, our system went from barely matching a high-ranked player to beating the top pros and has continued to improve since then. Supervised deep learning systems can only be as good as their training datasets, but in self-play systems, the available data improves automatically as the agent gets better. Our Dota 2 result shows that self-play can catapult the performance of machine learning systems from far below human level to superhuman, given sufficient compute. In the span of a month, our system went from barely matching a high-ranked player to beating the top pros and has continued to improve since then. Supervised deep learning systems can only be as good as their training datasets, but in self-play systems, the available data improves automatically as the agent gets better.

Dota 2 Dota 2

Rewatch live event Rewatch live event