neal2020github/awesome-embodied-agents

This repository is a curated reading list, in the style of the Awesome on GitHub series, focused on a specific corner of AI research: embodied agents that use vision language models or large language models as their reasoning core. By embodied the README means systems that exist in some kind of physical or simulated environment, such as a robot in a room or a virtual avatar that has to move around, look at things, and decide what to do next. The scope, written in a hidden HTML comment, lists what is in and what is out. The list includes embodied memory and retrieval, vision language model based reasoning and planning, exploration and navigation agents, embodied question answering, manipulation agents that use a model as a planner or skill selector, and benchmarks for measuring these systems. End to end vision language action policies, world models, low level robot control, diffusion policies, and generic language model agents with no perception of an environment are typically out of scope. The table of contents groups entries into ten sections: Survey, Benchmark, Agent System, Memory, Reasoning and Planning, Active perception, Navigation, EQA, Manipulation, and Related Lists. Each section is a bulleted list of papers, sorted by year in descending order, with links to the arXiv PDF and, where available, a project page or code repository. Surveys cover topics such as safety in embodied AI, self evolving embodied AI, foundation model driven robotics, and earlier surveys back to 2021. The Benchmark section is heavy with 2026 entries, including names like ESI-Bench, RoboMemArena, NavTrust, AsgardBench, and SpaMEM, each measuring a different capability such as spatial reasoning, memory, or trustworthy navigation. The README invites pull requests for papers and resources not yet listed. The full README is longer than what was shown.