Event
Proposes a Unified Formal Framework for Embodied Foundation Models
1 reports · 1 sources
Xinyi Hong, Mingyang Ren, and others proposed on arXiv a stochastic-kernel-based, architecture-agnostic probabilistic formal framework for uniformly describing embodied foundation models. The framework has two parts: characterizing models via conditional output distributions and action realization kernels, and inducing task-conditioned robot policies; and connecting pretraining, post-training, and policy deployment. The authors instantiated it on OpenVLA and DreamZero, covering two paradigms: autoregressive action token prediction and joint video-action generation. The paper was submitted as v1 on October 16, 2026.
Report timeline
- Connecting VLA and world-action models with a unified model-to-policy formal framework
The paper proposes an architecture-agnostic probabilistic formal framework based on stochastic kernels for uniformly describing embodied foundation models. The framework has two parts: characterizing models with conditional output distributions and action-realization kernels to induce task-conditioned robot policies, and connecting pretraining, post-training, and policy deployment. The authors instantiate it on OpenVLA and DreamZero, covering both autoregressive action-token prediction and joint video-action generation paradigms.
arXiv cs.RO ·