This is a small Vision-Language model running on an inexpensive phone chipset. Executes reasoning task while controlling UAV (drone) in sim world. The task is to find the human doctor near the medic truck. If desired, the drone could be instructed to hit the doctor or the truck. SuperfocusAI

"If desired" - very challenging for ape brains to switch between descriptive and normative mode. Descriptive = "IF DESIRED, yes the UAV can do X" Normative = "NOOOOOO! ... bad bad bad" Obviously the OP is about CAPABILITIES and is DESCRIPTIVE. If this discussion is challenging for you please stop reading stuff I write.

equal parts deeply impressed and scared!!

crazy choice of demo scenario here

wtf, this is cursed IDF tech. BTW things of this nature should be banned, why kill the doctors?

"hit" is a poor choice of words here, this isn't the IDF.

We had this at DJI ten years ago

No one had V-L models even 5 years ago. L = Language, like LLM

Wrong, we had object recognition already back then. Could easily do what you say in your post.

Object recognition is not the same as a VL model

That's not the point. The capabilities you mention have been there for a decade. Go watch a DJI Spark video.

ok i know its application

Compression of capability onto cheap hardware is where edge AI gets interesting. A VLM doing real-time reasoning on a phone chipset for drone control was unthinkable a year ago.

@chesterzelaya

this new GTA is mid

this tech is currently built by all armies of the world. but not to hit the doctor.

Excuse me?