Sensorimotor Alignment: A Platonic Proxy for Embodied Policy Success

Forum topic · 小凯 · 2026-10-24

Summary

Robotic foundation models are usually evaluated through expensive closed-loop rollouts, creating demand for cheap offline proxies of model capability. Building on the Platonic Representation Hypothesis (PRH)—the idea that as representations improve, modalities like vision and language converge toward a shared statistical model of reality—this paper asks whether motion trajectories constitute another such modality: as embodied models become more capable, should their sensory representations converge toward the structure of physical motion? The authors propose Sensorimotor Alignment (S²), a training-free offline metric that measures how well a model's sensory representation aligns with the geometry of physical motion. Experiments show that S² correlates positively with downstream robot task success, enabling low-cost ranking of models before closed-loop evaluation. S² can also identify promising vision-language model (VLM) backbones before they are fine-tuned into vision-language-action (VLA) models. The paper (arXiv:2610.31670) is authored by Yunfei Xie, Howard H. Qian, Shiyi Lan, Kaiyu Hang, and Chen Wei, and falls in the computer vision / robotics domain.

Overview

Field: CV / Robotics Authors: Yunfei Xie, Howard H. Qian, Shiyi Lan, Kaiyu Hang, Chen Wei Published: 2026-10-22 arXiv: 2610.31670

Abstract

Robotic foundation models are typically evaluated through costly closed-loop rollouts, which motivates cheap offline proxies for model capability. The Platonic Representation Hypothesis (PRH) suggests that as representations improve, different modalities such as vision and language converge toward a shared statistical model of reality. We ask whether motion trajectories form another such modality: as embodied models become more capable, should their sensory representations converge toward the structure of physical motion? To examine this, we propose Sensorimotor Alignment (S²), a training-free offline metric that measures how well a model's sensory representation aligns with the geometry of physical motion. We find that S² correlates positively with downstream robot task success across embodied model families, making it a low-cost offline proxy for prioritizing models before closed-loop evaluation. S² can further identify promising vision-language model (VLM) backbones before they are fine-tuned into vision-language-action (VLA) models.

Key Points

Closed-loop rollout evaluation of robotic foundation models is expensive; cheap offline capability proxies are needed.
Extends the Platonic Representation Hypothesis (PRH) to embodiment: physical motion may act as another converging modality.
Introduces Sensorimotor Alignment (S²), a training-free offline metric comparing sensory representations with motion geometry.
S² positively correlates with downstream robot task success.
S² helps select promising VLM backbones prior to VLA fine-tuning.

--- *Auto-collected on 2026-10-24*