LAP: Language-Action Pre-Training Enables Zero-shot Cross-Embodiment Transfer

TL;DR

LAP introduces a natural language-based pre-training method for vision-language-action models, enabling zero-shot transfer to new robot embodiments and improving manipulation success rates without embodiment-specific fine-tuning.

Contribution

The paper presents LAP, a novel approach that aligns robot actions with language, allowing zero-shot transfer and outperforming prior models in unseen robot tasks.

Findings

LAP-3B achieves over 50% zero-shot success on new robots.

LAP outperforms previous VLAs by roughly 2x in success rate.

LAP enables efficient adaptation and unified action prediction and VQA.

Abstract

A long-standing goal in robotics is a generalist policy that can be deployed zero-shot on new robot embodiments without per-embodiment adaptation. Despite large-scale multi-embodiment pre-training, existing Vision-Language-Action models (VLAs) remain tightly coupled to their training embodiments and typically require costly fine-tuning. We introduce Language-Action Pre-training (LAP), a simple recipe that represents low-level robot actions directly in natural language, aligning action supervision with the pre-trained vision-language model's input-output distribution. LAP requires no learned tokenizer, no costly annotation, and no embodiment-specific architectural design. Based on LAP, we present LAP-3B, which to the best of our knowledge is the first VLA to achieve substantial zero-shot transfer to previously unseen robot embodiments without any embodiment-specific fine-tuning. Across multiple novel robots and manipulation tasks, LAP-3B attains over 50% average zero-shot success, delivering roughly a 2x improvement over the strongest prior VLAs. We further show that LAP enables efficient adaptation and favorable scaling, while unifying action prediction and VQA in a shared language-action format that yields additional gains through co-training.