Meta-learning for wrestling

In three linesOpenAI demonstrates that a meta-learning agent can quickly learn to defeat a stronger non-meta-learning opponent in simulated robot wrestling and adapt to physical malfunctions.

Summary generated by Claude — human-verified