# AI use cases for Awesome Code as X (kkakkkka/awesome-code-as-x) Repo: [kkakkkka/awesome-code-as-x](https://github.com/kkakkkka/awesome-code-as-x) | Use cases: 3 | Page: https://dedupedai.com/use-cases/kkakkkka-awesome-code-as-x/ A curated list of coding-agent papers and projects that use a program as the intermediate representation for robot policies, world models, scenes, and image or video edits. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Shortlist Code-as-Policy robot papers that link to code: https://dedupedai.com/use-cases/kkakkkka-awesome-code-as-x/shortlist-code-as-policy-papers-with-code/ - Test a coding agent on robot tasks with CaP-X or RLE-Bench: https://dedupedai.com/use-cases/kkakkkka-awesome-code-as-x/benchmark-coding-agent-on-robot-tasks/ - Plan a video-to-Blender world program evaluation: https://dedupedai.com/use-cases/kkakkkka-awesome-code-as-x/plan-video-to-blender-world-program-eval/ ## How to shortlist Code-as-Policy robot papers that link to code with Awesome Code as X Repo: [kkakkkka/awesome-code-as-x](https://github.com/kkakkkka/awesome-code-as-x) | Difficulty: Beginner | Time: P1H | Basis: From the README Filter the list to Code-as-Policy entries with GitHub code for a robot project. **Who it is for:** A robotics engineer who needs a starting baseline for LLM-written robot control code and has limited time to read papers. **The problem:** The Code as Policy section has many entries, and it is slow to check which ones point to a code repository. **The outcome:** A short list of 3 to 5 candidate papers, each with its paper link, project page, and GitHub link, ready for a closer read. ### Steps 1. Open the Code as Policy section of the README and read the one-line description of each entry. 2. Note the entries that list a GitHub link, for example AGP, ASPIRE, ENPIRE, and GPT-Policy. 3. Open each candidate's arXiv link and project page and check the task type it targets. 4. `git clone https://github.com/NVlabs/ASPIRE` to inspect the code of one chosen candidate. (inferred, check the README) ### Prompt to paste into your agent ```text Using the README of kkakkkka/awesome-code-as-x, list the Code as Policy entries that link to a GitHub repository, and for each one give the paper title, the arXiv link, and the GitHub link. Expected output: a table of 3 to 5 rows. Verify that every link appears verbatim in the README. ``` ### Caveat This is a list of links, not a tool, so each linked project has its own setup, hardware needs, and license that you must check. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/kkakkkka-awesome-code-as-x/shortlist-code-as-policy-papers-with-code/ ## How to test a coding agent on robot tasks with CaP-X or RLE-Bench with Awesome Code as X Repo: [kkakkkka/awesome-code-as-x](https://github.com/kkakkkka/awesome-code-as-x) | Difficulty: Intermediate | Time: PT2H | Basis: From the README Pick a benchmark from the list to measure how well a coding agent writes robot code. **Who it is for:** An ML researcher who wants to compare coding agents on robot learning or manipulation tasks before choosing one for a project. **The problem:** There is no single agreed test for whether a coding agent can write, run, and revise robot programs. **The outcome:** A chosen benchmark with its project page open, a clear list of its task types, and a plan for running the first evaluation. ### Steps 1. Read the Benchmark section and the Robot evaluation infrastructure subsection. 2. Open the RLE-Bench project page at rle-bench.github.io and note what the coding agent must do. 3. Open the CaP-X project page at capgym.github.io and compare its robot manipulation tasks with RLE-Bench. 4. Pick one benchmark and follow its own setup instructions, since the list itself gives no install steps. ### Prompt to paste into your agent ```text Read the Benchmark section of the README in kkakkkka/awesome-code-as-x and compare RLE-Bench and CaP-X for a coding agent test on . Expected output: a short comparison of task type, required inputs, and evaluation method for each benchmark. Verify that each claim is taken from the linked project page or the README. ``` ### Caveat The README describes the benchmarks only in one line each, so you must read the linked project pages for setup and metrics. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/kkakkkka-awesome-code-as-x/benchmark-coding-agent-on-robot-tasks/ ## How to plan a video-to-Blender world program evaluation with Awesome Code as X Repo: [kkakkkka/awesome-code-as-x](https://github.com/kkakkkka/awesome-code-as-x) | Difficulty: Advanced | Time: PT3H | Basis: From the README Use PhysCodeGym to test whether a coding agent rebuilds a video as a physical Blender scene. **Who it is for:** A graphics or simulation engineer who wants to test whether generated scene code keeps the physics of a real video. **The problem:** Reconstructions from video often look right but break physics, and it is hard to check this automatically. **The outcome:** A short evaluation plan that uses the PhysCodeGym repository and dataset to check a reconstructed Blender program against the physics of its video. ### Steps 1. Read the World programs and Benchmark definitions in the README to confirm what a world program is. 2. Open the PhysCodeGym GitHub repository at github.com/ShaoShuai0605/PhysCodeGym. 3. Open the dataset at huggingface.co/datasets/ShaoShuai0605/PhysCodeGym and choose one video. 4. Ask a coding agent to write a Blender program for that video, then run the physics check described in the PhysCodeGym repository. ### Prompt to paste into your agent ```text Using the PhysCodeGym entry in the kkakkkka/awesome-code-as-x README, write a plan to evaluate a coding agent that rebuilds