HCPLab-SYSU/Embodied_AI_Paper_List: A Survey-Backed Reading List for Embodied AI

What the Embodied AI paper list actually is

The repository is a reading list, not a tool. Its top level contains five entries: README.md, EmbodiedAI.jpg, EmbodiedAI_Review.pdf, Survey.png and teaser.png. There is no package manifest, no source directory, no test suite and no configuration file. Everything a reader consumes lives inside README.md or in the two images and the PDF that the README embeds.

The intended audience is narrow and identifiable: researchers and graduate students entering embodied AI who need a starting bibliography, and reviewers who want to check whether a paper they are citing appears in a peer-reviewed survey. The README states the list accompanies "Aligning Cyber Space with Physical World: A Comprehensive Survey on Embodied AI", published in IEEE/ASME Transactions on Mechatronics in 2025. The survey is the authority; the repository is its index.

That framing matters for adoption. If you want a library to install and call, this is the wrong artifact. If you want a curated map of a fast-moving field with a citable survey behind it, the format is appropriate.

How the table of contents maps the field

The mechanism is a single Markdown table of contents with seven anchors: Books & Surveys, Embodied Simulators, Embodied Perception, Embodied Interaction, Embodied Agent, Sim-to-Real Adaptation and Datasets. Each anchor links to a section further down the same file, and each section is a flat list of entries.

An entry follows one shape. A bolded title, then an arXiv identifier and year, then an author list, then a link. For example, the README lists "Self-evolving Embodied AI", arXiv:2602.04411, 2026, by Tongtong Feng, Xin Wang and Wenwu Zhu, with a link to the PDF. There is no abstract, no venue field, no code link and no BibTeX block. The uniformity is deliberate and it makes scanning fast, but it also means the list carries no structured metadata you can query.

The four research targets named in the README's About section (embodied perception, embodied interaction, embodied agent, sim-to-real adaptation) line up with four of the seven sections, which is the clearest sign that the list is organized around the survey's argument rather than around convenience. Simulators and datasets sit outside that quartet as supporting material.

One structural choice is worth flagging. The README's update log says papers were arranged in chronological order in August 2024 "to make readers focus on newest works". Chronological ordering is good for catching recent output and bad for finding a specific older paper, since there is no alphabetical fallback and no search index beyond your browser's find function.

Installing nothing: reading the list on GitHub or offline

There is no installation. The repository has no package, no CLI and no server. The README gives two ways to consume it, and both are file operations rather than software setup.

The first is the rendered README on GitHub. The table of contents anchors jump to sections, and each entry's link opens the paper. If you prefer a local copy, clone the repository and open the Markdown in any editor; the images and the survey PDF come along with it.

git clone https://github.com/HCPLab-SYSU/Embodied_AI_Paper_List.git
cd Embodied_AI_Paper_List
ls

The `ls` output should match the top-level entries: EmbodiedAI.jpg, EmbodiedAI_Review.pdf, README.md, Survey.png and teaser.png. If you see a package.json or a setup.py, you have cloned something else.

The second route is the survey PDF itself, which the README links as EmbodiedAI_Review.pdf in the repository and as arXiv:2407.06886. The README also points to a companion book, "Multimodal Large Models: The New Paradigm of Artificial General Intelligence", hosted at hcplab-sysu.github.io/Book-of-MLM/.

Contributions go through the repository's issue tracker or by email to the two addresses listed in the README, and pull requests are explicitly welcomed. That is the entire contribution workflow.

Where the paper list falls short

The most concrete limitation is that the README is the database. There is no JSON, no CSV and no YAML export, so any analysis you want to run (counting papers per year, deduplicating across sections, tracking a topic over time) means parsing Markdown by hand or writing a scraper against a format that can change without notice.

Freshness is uneven and self-reported. The update log records a large refresh on 2026-03-11 covering "latest 2025-2026 papers across all categories", and a note from 2024-09-08 saying the Dataset section is being continuously updated. The last push to the repository was on 2026-06-10. Entries dated 2026 do appear, including arXiv:2602.04411 in Books & Surveys, so the list is not frozen, but the log does not promise a cadence and the README's own 2024-08-02 line about updating "weekly" is not corroborated by later entries.

Coverage is also asymmetric. Books & Surveys is dense and current. Datasets is acknowledged as a work in progress. If your work depends on a specific subarea, check that subarea's section before assuming the list is complete for it.

Finally, the entry format omits the things a practitioner usually needs: no venue, no code repository link, no citation count, no summary. A paper that shipped a widely used simulator looks identical in the list to one that did not.

Alternatives and how they differ in approach

The obvious comparison is the awesome-list pattern, such as Awesome-Embodied AI style collections. The difference is editorial posture. An awesome list is typically a community-maintained link dump organized by topic, with contributors adding entries ad hoc. This repository is anchored to a single peer-reviewed survey with a named author list (Yang Liu, Weixing Chen, Yongjie Bai, Xiaodan Liang, Guanbin Li, Wen Gao and Liang Lin), and the sections mirror that survey's structure. You get coherence and a citable source; you give up the breadth that comes from many independent contributors.

A second alternative is a general literature database with programmatic access, where you search by keyword and export BibTeX. Those give you query, filtering and citation export. This repository gives you a human-curated ordering and a survey's judgement about what matters. The two are complementary rather than competing, and the practical move is to use the repository to find the shape of the field and a literature database to pull citations.

A third alternative is the survey PDF alone. Reading arXiv:2407.06886 gives you the argument and the taxonomy; the repository gives you the raw list without the prose. If you only need the narrative, skip the repository. If you need to scan titles quickly, the README is faster than a PDF.

Maintenance, licence and what to check before citing

The repository is not archived, and the last push was on 2026-06-10. That is the only maintenance signal available. There are no releases, and the README does not document a versioning scheme, a deprecation policy or a rollback path, so there is nothing to upgrade and nothing to roll back. Your cost of staying current is re-reading the README or re-pulling the repository.

The repository does not state a licence. No LICENSE file appears among the top-level entries, and the README does not name a licence identifier. The survey paper itself carries its own publisher terms through IEEE/ASME Transactions on Mechatronics, and the linked arXiv preprint has arXiv's terms. If you plan to redistribute the list, mirror the PDF or reuse the images (EmbodiedAI.jpg, Survey.png, teaser.png), confirm the terms with the maintainers first via the contact addresses in the README. This is a factual gap in the repository, not a legal opinion.

For citation, the README gives the survey title, venue, year and author list, plus arXiv:2407.06886. Cite the survey, not the repository, when the claim comes from the survey's analysis.

Editorial conclusion

Adopt this repository if you need a single chronological entry point into embodied AI literature and you are willing to read the README as a plain document, because there is no build step, no CLI and no API to learn. Do not adopt it if you need a maintained software package, a benchmark harness or machine-readable metadata; the repository ships images, a PDF and a README, and nothing else at the top level. Before relying on it, verify three things yourself: that the sections you care about have entries newer than the ones you already have, that each arXiv identifier resolves to the paper you expect, and whether the licence terms are stated anywhere in the repository, since no licence file appears among the top-level entries.

Frequently asked questions

Official sources

1. HCPLab-SYSU/Embodied_AI_Paper_List on GitHub
2. Issues
3. README