Paper
VLA-Δ
A decision-centric vision-language-action architecture that replaces autoregressive decoding with a single classification token. It matches state-of-the-art results on the LIBERO benchmark and improves spatial reasoning.
Timeline (last 90 days)
Oct 19
VLA-Δ proposes a decision-centric architecture that replaces expensive autoregressive decoding with a single classification token, matching the current state of the art on the LIBERO benchmark while improving spatial reasoning.
Unverified50%