Movemur: Large World Models, Newtonian Physics, and the Embodied Bottleneck of Artificial General Intelligence

Abstract

Contemporary large language models achieve remarkable competence in symbolic reasoning, dialog, and the synthesis of internet-scale text and yet remain conspicuously deficient in the contact-rich, gravity-bound competence that any biological six-month-old begins to acquire while attempting to climb out of a crib. Drawing on embodied cognition, on Yann LeCun’s joint-embedding predictive architecture (JEPA) and its video extension V-JEPA 2, and on Google DeepMind’s interactive world-model line (Genie, SIMA, SIMA 2), this review argues that the principal remaining bottleneck on the path to artificial general intelligence (AGI) is not linguistic or symbolic but Newtonian: the capacity to ground perception, planning, and action within the constraints of classical mechanics as they impinge on a physical body in real time. We propose that climbing and walking robots occupy a privileged position in this research program because locomotion compresses into a single integrated task virtually every requirement for embodied intelligence under physical law while imposing failure modes that are immediate and irreversible. Building on prior surgical and grappling benchmarks, we introduce a Pauline diagnostic triad (vivimus/movemur/sumus), review recent empirical landmarks of legged-robot learning, outline a milestone ladder for locomotion-grounded embodied AGI, and frame the field’s stance toward the sim-to-real gap. The thesis is concise: until an autonomous system can climb an unfamiliar staircase, traverse rubble, and recover from a slip with adult competence, claims of general intelligence remain rhetorical. We advance this as a falsifiable hypothesis: walking and climbing are plausibly necessary, though not sufficient, conditions for it.

Received: 9 May 2026 | Revised: 10 July 2026 | Accepted: 24 July 2026

Conflicts of Interest

The author declares that he has no conflicts of interest to this work.

Data Availability Statement

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Author Contribution Statement

Enrique Díaz Cantón: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Resources, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.