Fine-tuned Dream-1 world action model jumps 159% on real-robot benchmark at 1/100th the cost

paigeinsf · x · 2026-11-19

Fine-tuning Dream-1 (a world action model from Dream Labs) on 500 real-robot demos across 10 tasks lifted its RoboDojo-Real score from 8.5 to 22 (+159%), beating fine-tuned π0.5 (20.6) but still trailing LLM methods overall:

Takeaway: LLMs know what to do; WAMs know how to do it.

Original post →

More from Embodied

Embodied channel →