SEAR: Self-Evolving Agentic Robotics
Frontier language-model agents can now improve robots on their own: they write control code, test it, and keep what works as skills and memories for the next task. We introduce SEAR (Self-Evolving Agentic Robotics), a framework and benchmark built around two questions. Is the agent really learning, or just trying more often? And does what it learns still help on new tasks, in a new simulator, and…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.
