Physical AI

Learning and control methods that stay consistent with physical laws when models interact with energy, matter, and motion.

Physical AI studies intelligence that is constrained by the real world: conservation laws, sensor noise, latency, and hardware limits. Work in this track asks how models can plan, predict, and act without violating the physics of the systems they control.

Typical topics include physics-informed learning, hybrid mechanistic/neural models, sim-to-real transfer, identification of dynamical systems, and evaluation protocols that measure physical consistency rather than only predictive accuracy.

Submissions should state the physical assumptions, the operating envelope of the system, and how failure modes are detected when those assumptions break.

All papers in this track must follow the PIJ Ethics Code.