Zhejiang University and Alibaba DAMO unveil VLA-Corrector, a 40M-parameter module for real-time robot action correction
jiqizhixin · x · 2026-09-14
Zhejiang University and Alibaba DAMO Academy present VLA-Corrector, addressing the open-loop blind zone in VLA models where robots commit to action chunks without checking for errors between inferences.
- Adds a lightweight 40M-parameter module that monitors execution state in real time
- Detects trajectory deviation and triggers early correction without forcing the large model to recompute each step
- Makes the action horizon adaptive: longer when stable, shorter when correction is needed
- Horizon sweeps show a clear tradeoff: for π0.5, lengthening the horizon cuts policy calls by 4x
More from Embodied
- Main approaches for finetuning e2e driving models in close(ish)-loop — abursuc · 2026-09-15
- Unitree upgrades G1 humanoid after 18,000+ units sold, G1+ adds torque and sensing — CyberRobooo · 2026-09-15
- Light Origins' humanoid parkour policy picks walk, vault or climb with onboard sensing only — micoolcho · 2026-09-15
- Agility Robotics unveils Digit 5 humanoid that works alongside people, fence-free — The Decoder · 2026-09-15
- AgentSTAR uses agents for articulated object tracking; Malik hails end of "grad student descent" — CSProfKGD · 2026-09-15
- China opens world's first plant mass-producing a humanoid robot every 10 minutes — mariolefebvre · 2026-09-15