Zhejiang University's VLA-Delta Teaches Robots Explicit Physical State Change

1 reports archived1 independent sourcesupdated 11/15/2026, 10:51:00

Synthesis & Latest Updates

1 Sources Cross-Validated

Zhejiang University's ACES Lab OmniAI team introduces VLA-Delta, which turns task progress, spatial displacement and contact changes into explicit pretraining supervision so the representation behind action generation learns both task semantics and visual motion. Built on a 2B vision-language backbone, it reaches 98.4% average success on LIBERO and 68.0% on five real dual-arm robot tasks.

LATEST/Zhejiang University's ACES Lab OmniAI team introduces VLA-Delta, which turns task progress, spatial displacement and contact changes into explicit pretraining supervision so the representation behind action generation learns both task semantics and visual motion. Built on a 2B vision-language backbone, it reaches 98.4% average success on LIBERO and 68.0% on five real dual-arm robot tasks.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

量子位 (微信公众号)T2Nov 15, 10:51·72 pts
Zhejiang University's VLA-Delta Teaches Robots Explicit Physical State Change
- Learnable Physical State Difference Queries encode current-to-future state change for the action expert
- Three complementary supervisions: structured change slots, dense visual motion and difference text
- 2B backbone hits 98.4% on LIBERO and 68.0% average across five real dual-arm tasks
View Related Reports & Analysis Original Source