Embodied_AI_Paper_List: A Survey-Backed Reading List for Embodied AI

What the repository is, and the gap it fills

Embodied AI research is spread across robotics venues, vision conferences and arXiv preprints, and the vocabulary shifts between subcommunities. A person entering the field has no obvious single place to learn which simulators exist, which datasets are used for manipulation versus navigation, and how sim-to-real work relates to embodied agent work. This repository addresses that by pairing a paper list with a survey. The survey is titled "Aligning Cyber Space with Physical World: A Comprehensive Survey on Embodied AI" and the README states it was accepted by IEEE/ASME Transactions on Mechatronics in 2025. The same README describes the survey's scope as covering representative embodied robots and simulators, then four research targets: embodied perception, embodied interaction, embodied agent, and sim-to-real adaptation, followed by a discussion of Multi-modal Large Models in virtual and real embodied agents. The intended reader is a researcher or graduate student building a mental map of the field, not an engineer looking for a library to install.

The seven-section taxonomy and why chronological ordering matters

The table of contents lists seven sections: Books & Surveys, Embodied Simulators, Embodied Perception, Embodied Interaction, Embodied Agent, Sim-to-Real Adaptation, and Datasets. That structure mirrors the survey's four research targets with two additions (simulators up front, datasets at the end) plus a bibliography section. The update log records a deliberate ordering decision: on 2024.08.19 the maintainers wrote that papers are arranged in chronological order "to make readers focus on newest works." This is a real editorial choice with consequences. Within a section, the newest entries appear first, so a reader scanning the top of Embodied Agent sees 2026 work before foundational 2023 papers. Chronological ordering is good for tracking a moving frontier and bad for building a curriculum, because nothing in the list signals which older paper introduced a concept the newer ones assume. The Datasets section, added 2024.08.31 and described as being continuously updated since 2024.09.08, is the part most likely to be useful to someone who already knows the concepts and needs to pick data.

Entry format and what each listing actually contains

Each entry follows a consistent shape visible in the Books & Surveys section: a bolded title, an arXiv identifier with year, an author list, and one or more bracketed links. The survey entry itself carries a [Paper] link to the arXiv PDF. A book entry, "Multimodal Large Models: The New Paradigm of Artificial General Intelligence" by Yang Liu and Liang Lin, published by the Publishing House of Electronics Industry in 2024, instead carries a [Page] link to hcplab-sysu.github.io/Book-of-MLM. So the link target varies by resource type. What is absent is equally consistent: no per-paper summary, no difficulty rating, no indication of which papers are surveys versus primary results, no venue tag beyond the arXiv year. A reader cannot tell from the list alone whether an entry is a 40-page review or a six-page workshop paper. That is a limitation of the format, and it means the list works best as a discovery index that you then resolve against arXiv yourself.

Getting the list and the survey onto your machine

There is no package to install. The repository is consumed by cloning it or by reading it on GitHub. The README's own badges point at the survey PDF stored in the repository at the path EmbodiedAI_Review.pdf on the main branch, and at the arXiv abstract page for 2407.06886. The repository also ships image assets referenced from the README, including EmbodiedAI.jpg, teaser.png and Survey.png, which means a clone carries the figures used in the survey overview. Because the primary language field is unset and no build tooling is described, there is no documented command to generate a website or a BibTeX file from the list. If you want citations, you extract them from the arXiv identifiers by hand or with your own script. The README states that updates happen regularly, with one log line saying the project is updated weekly, so a periodic git pull is the only maintenance action the material supports.

Contribution path and the maintenance signals in the update log

The README invites issues, pull requests and email to two addresses, liuy856@mail.sysu.edu.cn and chen867820261@gmail.com. That is the whole contribution mechanism: no CONTRIBUTING file, no entry template, no CI described. The update log is the best evidence of how the project is actually maintained. It shows a burst of activity in July and August 2024 around the first release, the survey's acceptance recorded on 2025.05.27, and then a gap until 2026.03.11, when the log says the paper list was updated with "latest 2025-2026 papers across all categories." A separate 2026.03.31 entry announces a different project, Physical Agent Operation System, rather than an update to this list. So the practical cadence is closer to periodic sweeps than to weekly maintenance, despite the 2024.08.02 line claiming weekly updates. Anyone citing this list should note the date of the sweep they used.

Where this list is the wrong tool

Three cases. First, if you need to run something. This repository contains no code, no simulator, no evaluation harness. The simulators it lists are external projects reached through links. Second, if you need machine-readable metadata. There is no BibTeX file, no JSON export, no CSV, and no documented schema, so any integration into a reference manager or a literature database is manual work you own. Third, if you need coverage guarantees. The list is curated by one lab's reading of the field, and the taxonomy is the survey's taxonomy. Work that does not fit the four research targets, or that the maintainers did not encounter, will not appear. A systematic review with explicit inclusion criteria cannot be built on a hand-maintained list, and the README makes no claim that it can. The absence of a stated license for the repository contents is a separate problem: the material does not say under what terms the list itself may be redistributed, even though the linked papers carry their own arXiv terms.

How it differs from Papers with Code and Awesome lists

The obvious comparison is Papers with Code, which indexes papers alongside implementations and reported benchmark results. The difference in approach is that Papers with Code derives its structure from what authors and users submit, organized around tasks and leaderboards, while this repository derives its structure from a single survey's argument. That makes this list more coherent as a narrative and less complete as an index. The second comparison is the general "awesome" list pattern, a flat collection of links grouped by loose headings. This repository is closer to that pattern than to Papers with Code, but with three distinctions: the grouping follows a published survey, the ordering within groups is chronological by explicit decision, and the list is tied to a specific peer-reviewed artifact that a reader can check claims against. If you want leaderboards and code links, Papers with Code is the better fit. If you want a defensible reading order that someone has already argued for in print, this repository is the better fit.

Who should adopt it, and what to check first

Adopt it if you are a new graduate student or a researcher moving into embodied AI from an adjacent area and want a structured starting point with a citable survey behind it. Adopt it if you are writing a related-work section and want to check whether you have missed a category, particularly the sim-to-real and dataset categories that are easy to under-search. Do not adopt it as a dependency, a dataset source, or a benchmark reference; it provides none of those. Do not treat its ordering as a curriculum without cross-checking dates, because the newest-first arrangement buries foundations. Before you rely on it, verify two things: that the section you need was touched in the 2026.03.11 sweep, and that the survey PDF at EmbodiedAI_Review.pdf on the main branch matches the arXiv version you cite. The repository states no license for its own contents, so confirm reuse terms with the maintainers before mirroring it.

Editorial conclusion

Use this repository if you are starting an embodied AI literature review and want a curated, survey-backed entry point whose ordering you can trust more than a search engine's. Do not use it if you need runnable code, benchmark harnesses or maintained datasets: it links to papers rather than shipping implementations, and the README does not state a license for the list itself. Before relying on it, open the arXiv PDF at arXiv:2407.06886 and check whether the section you care about has been refreshed since 2026.03.11, the date of the most recent paper-list update recorded in the update log.

Official sources

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