Physical AI: bridging the sim-to-real divide toward embodied, ethical, and autonomous intelligence

Share:

Article / Journal

Author(s) / editor(s):
Partha Pratim Ray

Year: 2026

Keywords: Physical Artificial Intelligence, Generative Physical AI, Embodied intelligence, World-model learning, Simulation fidelity
Language(s): English

Abstract:
The article positions Physical Artificial Intelligence as a distinct paradigm in which intelligence is embodied and enacted through continuous, closed-loop interaction with the physical world, rather than confined to digital or symbolic domains. It synthesizes advances in robotics, differentiable simulation, neuromorphic computing, multimodal world models, and autonomous control into a unified framework. The authors introduce three major contributions: a PDE-assisted generative–physical framework that embeds physical laws into world models; a six-level capability-based taxonomy for classifying Physical AI systems; and a quantitative evaluation framework with metrics spanning efficiency, safety, energy use, sim‑to‑real fidelity, uncertainty, and sustainability. The paper also reviews key enabling technologies and outlines future directions, including federated autonomy, affect-aware interaction, and large-scale robotic coordination and governance.

https://doi.org/10.1007/s44379-025-00050-y

Post created by: Virginia Signorini

Back to overview