nvidia's new WAM is impressive:
It needs very few demonstrations ("Zero-shot generalization and few-shot adaptation"), and self improvement with RL on the real robot increases task success rate.
Finally, a new architecture emerges: a (cloud) frontier model in agentic mode orchestrates the full robot. This was already demonstrated in [1] and may be the reason why openai closes its robotics department [2].
nvidia's new WAM is impressive: It needs very few demonstrations ("Zero-shot generalization and few-shot adaptation"), and self improvement with RL on the real robot increases task success rate.
Finally, a new architecture emerges: a (cloud) frontier model in agentic mode orchestrates the full robot. This was already demonstrated in [1] and may be the reason why openai closes its robotics department [2].
[1] https://research.nvidia.com/labs/srl/projects/astranomy/
[2] https://fortune.com/2026/11/13/openai-disbands-robotics-team...