Third-person imitation learning

In three linesOpenAI releases a third-person imitation learning method enabling agents to learn from external observations without direct action access. The technique improves generalization and reduces reliance on first-person demonstrations.

Summary generated by Claude — human-verified