Awesome-VLA-Papers organizes a collection of seminal research papers concerning Vision-Language-Action (VLA) models. This repository aims to provide a structured overview of the advancements in this field, particularly focusing on models leveraging action tokenization. The primary objective is to offer a readily accessible resource for researchers and practitioners seeking to understand the evolution and current state of VLA technology, covering both foundational and recent work.
This repository stands out by focusing specifically on the survey's scope, offering a curated list of high-impact papers within the VLA domain. The organization into 'Foundation Models' and 'Vision Foundation Models' provides a clear structure for navigating the literature. The inclusion of direct links to papers, code, and models facilitates immediate access to relevant resources. The continuous updating with the latest research ensures the repository remains a current reference.
- Foundation Models: Covers foundational language models like Transformers, BERT, and GPT series, which are crucial for VLA. These models serve as building blocks for understanding language context and generating textual descriptions.
- Vision Foundation Models: Includes key vision models like ViT and CLIP, forming the basis for visual understanding within VLA systems. These provide representations of visual data used in conjunction with language.
- Image Generation Models: Features diffusion models like DALL-E, GLIDE, and Stable Diffusion, pivotal for generating images from text prompts in VLA applications.
- VLA Model Architectures: Highlights models like GLIP and DINO that directly address the intersection of vision and language, including techniques for visual grounding and joint learning.
- Recent Advancements: Regularly updated with recent papers like Gemma and Gemini models, reflecting the rapidly evolving landscape of VLA research.
- Code & Model Links: Provides direct links to GitHub repositories for code and Hugging Face model hubs for pre-trained models, simplifying experimentation and implementation.
- Comprehensive Coverage: Encompasses a wide range of topics within VLA, including pre-training techniques, image generation, and multimodal understanding.
The repository is actively maintained and frequently updated with new publications. The inclusion of recent papers and links to associated code and models indicates ongoing relevance. The consistent additions of papers suggest active community contribution and support. The structured organization and clear presentation further contribute to its reliability as a reference resource.
This repository benefits researchers, developers, and students interested in Vision-Language-Action models by providing a curated and up-to-date list of key papers. It addresses the need for a centralized resource to navigate the rapidly evolving field of multimodal AI. By offering direct links to papers, code, and models, it simplifies research and development efforts, providing a valuable starting point for exploring VLA technologies and their applications.