Cloak: Zero-Shot Cross-Embodiment Manipulation by Masking the End-Effector from the VLA

2026-06-22

AI Summary

Existing vision-language-action (VLA) models struggle to achieve zero-shot transfer across diverse robot embodiments. This work proposes Cloak-VLA, a method that leverages known robot geometry during training to dynamically render and occlude the end-effector in wrist-mounted camera views, thereby compelling the model to learn embodiment-agnostic visual representations. Without requiring additional data or generative models, Cloak-VLA enables zero-shot transfer from a single gripper-trained policy to various unseen embodiments—including anthropomorphic five-fingered hands—while preserving task performance on the source embodiment. To the best of our knowledge, this is the first approach to achieve effective cross-embodiment zero-shot policy transfer without compromising original-task performance.

Abstract

We present Cloak, a training recipe that endows a Vision-Language-Action (VLA) model with zero-shot cross-embodiment transfer by cloaking the end-effector from its own wrist camera. The end-effector occupies a large and consistent region of the wrist view and masking it allows for embodiment-agnostic visual reasoning. Cloak renders a mask in simulation from the robot's known geometry, accurately and in real time, with no segmentation or generative models. During training, we augment the mask so the model generalizes to embodiments unseen at training time. We demonstrate the recipe with Cloak-VLA, a VLA trained with Cloak on a single parallel-jaw gripper dataset. No data of new embodiments is ever collected. Cloak-VLA transfers zero-shot to various unseen embodiments, including another gripper, another arm, and a five-fingered hand, while preserving the source embodiment's performance. By decoupling the wrist view from its own embodiment, Cloak allows data to outlive the hardware it was collected on.

Problem

• zero-shot
• cross-embodiment
• manipulation
• Vision-Language-Action
• end-effector

Innovation

• zero-shot cross-embodiment
• end-effector masking
• Vision-Language-Action (VLA)
• embodiment-agnostic manipulation
• simulation-based masking