The Platonic Representation Hypothesis argues that vision and text are two projections of the same underlying world. As vision and language models scale, their representations converge toward this shared reality. We ask: Can physical robot motion be viewed as another projection of that same world? Our answer is yes. We find that more capable embodied models better align their vision and language representations with the structure of motion trajectories. 🚀 Based on this, we introduce Sensorimotor Alignment (S²): A Platonic Proxy for Embodied Policy Success. 📈 First, we find that S² is positively correlated with downstream robot task success across model families, in both simulated and real-world settings—making S² a training-free predictor of robot policy success. 🔍 Furthermore, we use S² to explain, at the representation level, several phenomena in VLAs and WAMs that were previously observed only through closed-loop results—such as why VLAs struggle to follow language instructions and why Fast-WAM can skip test-time future generation. 💡 We believe Sensorimotor Alignment offers a low-cost, representation-level view of robot foundation models that closed-loop success rates alone cannot reveal. More: 🌐 Blog & Video: yunfeixie233.github.io/s2/ 📄 Paper: arxiv.org/abs/2610.31670 💻 Code: github.com/yunfeixie233/sens… I’m sincerely grateful to Howard H. Qian, Shiyi Lan, Kaiyu Hang, and Chen Wei @Chen_Wei for their guidance and collaboration.