Intention Distillation lifts GR00T success from 64% to 85% on SimplerEnv

POSTECH · hf · 2026-08-31

POSTECH proposes Intention Distillation (INDI): during training a frozen VLM teacher distills behavior-level intent from the instruction, a coarse action summary, and execution video; the deployed VLA recovers this multimodal intent at an intermediate decoder layer to organize action prediction.

Results:

Analyses show the recovered latent captures behavior objective and execution progress.

Original post →

More from Embodied

Embodied channel →