Recent work shows that VLMs and AI-agents fail to acheive

state-of-the-art planning capabiliteis such as symbolic plan-

ner’s. On the other hand, it has been demostrated that VLMs

can overcome the strict requirement of symbolic planners

for formal domain model description, since VLMs can ap-

ply common-sense without formal descriptions. We present

NAPC1, a new neuro-symbolic method for generic adapta-

tion of planning domain using VLMs. We demonstrate NAPC

with Neil robot2, in real-world lab, showing it can adapt to un-

foreseen situation. We also show that the method reaches the

state-of-the-arm planning capabilities of symbolic planners.