D&A Turns Any VLM into a Zero-Shot Robot Reward Model
A University of Washington and NVIDIA team introduced D&A, which decouples description from assessment so that any VLM (GPT, Gemini, Claude, etc.) can serve as a zero-shot robot reward model without training, achieving SOTA on three benchmarks.
2026-11-03 ~ 2026-11-04 · 3 related posts
- D&A Turns Any VLM into a Zero-Shot Robot Reward Model, SOTA on 3 Benchmarks — DJiafei · 2026-11-03
- D&A turns any robot task video into an off-the-shelf reward model with zero training — KyleMorgenstein · 2026-11-04
- D&A turns any VLM into a zero-shot robot reward model with SOTA on three benchmarks — KyleMorgenstein · 2026-11-04