RedFlow Turns Robot Failures Into Action-Level Corrections
The paper proposes converting failed actions into localized corrections for flow-matching vision-language-action policies.
Semiconductor Domain: edge inference
General multimodal model shows emergent embodied reasoning
An exploratory mobile-robot study reports spatial memory and action consequence reasoning from a general multimodal model.
