Vela: Scaling Vision-Language-Action Models with Adaptive Action Curve Parametrization
Vela models robot actions as adaptive continuous trajectories and reports gains on simulation and real manipulation tasks.
Vela is a vision-language-action foundation model that represents future robot behavior as continuous trajectories rather than fixed-rate action chunks. It uses a compact spline-based action representation, motion-dependent temporal support, and a shared interface for heterogeneous embodiments, so a fixed output budget can adapt temporal resolution. The authors pretrain on large-scale multi-embodiment robot data and report promising results on LIBERO-X, EBench, and two real long-horizon tasks: egg-cake cooking and potato shredding.
- Vela represents robot motion as continuous spline trajectories instead of fixed-rate action chunks.
- Motion-dependent temporal support lets a fixed output budget change resolution.
- A shared action interface covers heterogeneous robot embodiments.
- Evaluated on LIBERO-X, EBench, and real egg-cake and potato-shredding tasks.
From the original article
1 min read at huggingface.co
Most vision-language-action models represent future motion as fixed-rate action chunks, tying temporal resolution and prediction horizon to a fixed output budget.
By Yifan Li, Jiaxu Wang, Dongming Wu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.