Robots Achieve Few-Shot Generalization
tonyzzhao · x · 2026-07-17
The shared post emphasizes that "the era of robots is arriving," quoting an observation about robot learning: by scaling up pre-training and then using a small amount of internal data for hill-climb fine-tuning, you can get a universal recipe.
The quote provides a more specific result: For the first time, "a single fine-tuning sample can teach a new behavior, and it generalizes." The folding task serves as an example—the model learned using just one sample per folding strategy and was tested on unseen settings, successfully learning 4 folding strategies.
Related event: Robot General Intelligence Accelerates(2 posts)→
More from Embodied
- Berkeley and NVIDIA's AW-0 hits 80.8% on RoboTwin and 71.1 rubric score on real robots — trevordarrell · 2026-10-27
- AW-0 hits 80.8% on RoboTwin tasks, beating strongest baseline by 16 points — trevordarrell · 2026-10-27
- AW-0's code stream fixes world model prediction: picks mango where baseline grabs apple — trevordarrell · 2026-10-27
- AW-0 hybrid mode lifts RoboTwin success from 20.6% to 29.2% under visual randomization — trevordarrell · 2026-10-27
- Inside AW-0: code, video and action experts jointly denoise in one Mixture-of-Transformers — trevordarrell · 2026-10-27
- AW-0 runs at 230ms on a single RTX 5090, 2.4x faster than autoregressive code planning — trevordarrell · 2026-10-27