RT-Transformer: The Transformer Block as a Spherical State Estimator

TL;DR

This paper demonstrates that the core components of the Transformer architecture naturally emerge from a geometric estimation problem involving spherical states, unifying attention, residuals, and normalization.

Contribution

It introduces a geometric perspective showing that Transformer components are derived from a single spherical estimation framework, rather than independent design choices.

Findings

Attention acts as evidence aggregation in the estimation process.

Residual connections perform incremental state updates.

Normalization projects the state back onto the hypersphere.

Abstract

We show that the core components of the Transformer block -- attention, residual connections, and normalization -- arise naturally from a single geometric estimation problem. Modeling the latent state as a direction on the hypersphere, with noise defined in the tangent plane at the current estimate, yields a precision-weighted directional inference procedure in which attention aggregates evidence, residual connections implement incremental state updates, and normalization retracts the updated state back onto the hypersphere. Together, these components follow from the geometry of the estimation problem rather than being introduced as independent architectural choices.