From a 1495 mechanical knight to a model that simulates gravity.
The idea of a machine that moves with purpose is older than "AI." What is new is the ability to learn that purpose from data — and to fail, iterate, and improve in the physical world.
Leonardo's Mechanical Knight
Da Vinci designs a life-size armored automaton operated by pulleys and cables — able to sit, stand, raise its visor, and move its arms. The first Western blueprint for a humanoid machine.
Grey Walter's Tortoises
Two vacuum tubes, a photocell, and a shell. "Elsie" and "Elmer" exhibit obstacle avoidance and light-seeking — emergent behavior from a tiny physical system.
Shakey the Robot
The first mobile robot to combine perception, planning, and action. Shakey's route-finding algorithm, A*, became the basis for modern GPS navigation.
"Physical AI" Gets Its Name
Nature Machine Intelligence publishes a paper defining Physical AI as systems that fuse materials, mechanics, computation, biology, and chemistry to act autonomously in unstructured environments.
World Models Become the Engine
NVIDIA Cosmos 3, world foundation models, and open VLA stacks turn synthetic physics into training data — the "textbook" from which robots learn before touching the real world.
