Hacker News new | past | comments | ask | show | jobs | submit
tzury | on: $96 3D-printed rocket that recalculates its mid-air trajectory using a $5 sensor
"let alone with AI" what's falling into the AI category here perhaps is the key question, since microseconds counts, and LLM are very slow!

Even the fastest "real-time" LLM frameworks currently report sub-second latencies around 120ms. This is fine for high-level mission planning (e.g., "fly to the red house") but too slow to prevent a drone from hitting a tree at 50mph (80 KM/h)[1]

Whilst the Shahed-136 kamikaze drone typically flies at a maximum speed of around 185 km/h (roughly 115 mph or 100 knots).

[1] https://arxiv.org/html/2602.19534v1 [2] https://en.wikipedia.org/wiki/HESA_Shahed_136

ninjagoo | parent
> "let alone with AI" what's falling into the AI category here perhaps is the key question, since microseconds counts, and LLM are very slow!

LLMs (Large Language Models) are far from the only type of AI around. It's a pretty broad field, and there are real-time AI systems, for example, self-driving cars, which have the response times you're thinking of. [1]

[1] https://en.wikipedia.org/wiki/Artificial_Intelligence:_A_Mod...

tzury | root | parent
hence: "what's falling into the AI category here perhaps is the key question"

YOLO is a good example for something that can work.

https://docs.ultralytics.com/models/yolov8/