# AI use cases for EMERGE Policy (EMERGE-Policy/EMERGE-Policy) Repo: [EMERGE-Policy/EMERGE-Policy](https://github.com/EMERGE-Policy/EMERGE-Policy) | Use cases: 4 | Page: https://dedupedai.com/use-cases/emerge-policy-emerge-policy/ EMERGE-Policy is a Python embodied-agent framework where a primary agent plans and calls specialized perception, action, and verification models to control robots in LIBERO simulation. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Run a robot task from a browser console: https://dedupedai.com/use-cases/emerge-policy-emerge-policy/run-robot-task-from-browser/ - Run one robot task headlessly and get JSON back: https://dedupedai.com/use-cases/emerge-policy-emerge-policy/run-single-task-headless/ - Reset and switch simulation scenes from the terminal: https://dedupedai.com/use-cases/emerge-policy-emerge-policy/reset-and-switch-scenes-in-tui/ - Evaluate the agent on LIBERO and its robustness variants: https://dedupedai.com/use-cases/emerge-policy-emerge-policy/benchmark-policy-on-libero-variants/ ## How to run a robot task from a browser console with EMERGE Policy Repo: [EMERGE-Policy/EMERGE-Policy](https://github.com/EMERGE-Policy/EMERGE-Policy) | Difficulty: Intermediate | Time: 3 hours | Basis: From the README Start the Web console, create a conversation, and give the robot a task with live camera views. **Who it is for:** A researcher or operator who wants a visual workspace for daily robot task runs in simulation without writing code. **The problem:** Managing model services, controllers, and scene state by hand for each experiment is slow and easy to get wrong. **The outcome:** The robot attempts the typed task in an isolated scene, with live camera views, conversation, and task progress visible in the browser. ### Steps 1. Clone the repository with `git clone --recurse-submodules https://github.com/EMERGE-Policy/EMERGE-Policy.git` and then run `git submodule update --init --recursive`. (inferred, check the README) 2. Create and activate the environment with `conda create -n EmergePolicy python=3.12 -y` and `conda activate EmergePolicy`. 3. Start the model services for OpenPI VLA with `OPENPI_GPU=0 VGGT_GPU=1 SAM3_GPU=2 bash scripts/model_server/start_external_model_servers.sh --services openpi,vggt,sam3`. (inferred, check the README) 4. Launch the browser console with `emerge web`. 5. Open http://127.0.0.1:8080/, complete the first-run API setup, create a conversation, and enter a task. ### Prompt to paste into your agent ```text In the EMERGE-Policy repo, follow the README Quick Start to start the Web console with `emerge web`, create a conversation, and submit the task "". Report the scene that was created and the task progress shown in the console. Verify that the conversation has its own isolated scene and workspace before reporting success. ``` ### Caveat The first run needs several large model checkpoints and GPUs assigned to the model services, so setup takes much longer than the launch command itself. 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/emerge-policy-emerge-policy/run-robot-task-from-browser/ ## How to run one robot task headlessly and get JSON back with EMERGE Policy Repo: [EMERGE-Policy/EMERGE-Policy](https://github.com/EMERGE-Policy/EMERGE-Policy) | Difficulty: Intermediate | Time: 3 hours | Basis: From the README Send one natural-language task to a running Controller and read the JSON result from stdout. **Who it is for:** A developer or evaluation engineer who needs to call robot tasks from a script or batch job. **The problem:** Scripts need a non-interactive way to run a task and capture the outcome without opening a terminal interface. **The outcome:** The task runs against the Controller and the result is printed to stdout as JSON, with workspace files stored in ~/.Emerge/workspace. ### Steps 1. Complete the environment setup with `conda activate EmergePolicy` from the repository root. 2. Start the model services with `OPENPI_GPU=0 VGGT_GPU=1 SAM3_GPU=2 bash scripts/model_server/start_external_model_servers.sh --services openpi,vggt,sam3`. (inferred, check the README) 3. Start the Controller in one terminal with `python -m robot.controller --driver libero_mujoco --driver-config dev/libero_mujoco_driver_sample.json`. (inferred, check the README) 4. In a second terminal, run `python -m Emerge.cli.headless "Put the red block in the basket"`. 5. Read the JSON result from stdout and check the files in ~/.Emerge/workspace. ### Prompt to paste into your agent ```text In the EMERGE-Policy repo, with the Controller already running, execute `python -m Emerge.cli.headless ""` and capture stdout. Expected output is a JSON result for the task. Verify that the output parses as valid JSON before reporting it. ``` ### Caveat The README says to configure the API through Web or TUI before using Headless, so a script cannot run cold on a fresh machine. 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/emerge-policy-emerge-policy/run-single-task-headless/ ## How to reset and switch simulation scenes from the terminal with EMERGE Policy Repo: [EMERGE-Policy/EMERGE-Policy](https://github.com/EMERGE-Policy/EMERGE-Policy) | Difficulty: Intermediate | Time: 2 hours | Basis: From the README Use the terminal interface to reload the environment or pick a different scene without restarting. **Who it is for:** A robot operator working over SSH or in a terminal who runs many short scene trials in a row. **The problem:** Restarting the whole stack between scenes slows down iteration on the same task. **The outcome:** The environment reloads with /reset, or a new scene is chosen with /scene, while the robot state and plan stay visible in the TUI. ### Steps 1. Start the Controller with `python -m robot.controller --driver libero_mujoco --driver-config dev/libero_mujoco_driver_sample.json`. (inferred, check the README) 2. In another terminal, launch the interface with `emerge`. 3. Enter a task in the conversation to start planning. 4. Type `/scene` to browse and select a new scene, or `/reset` to reload the current environment. 5. Check the plan and robot state panels to confirm the environment changed. ### Prompt to paste into your agent ```text In the EMERGE-Policy repo, open the TUI with `emerge` and run `/scene` to select the scene named "", then run a task. Expected output is the selected scene loaded with plan and robot state shown. Verify that the scene name in the interface matches the requested one before continuing. ``` ### Caveat The README describes /scene and /reset as TUI commands and does not list which scenes are available, so you must browse them in the interface. 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/emerge-policy-emerge-policy/reset-and-switch-scenes-in-tui/ ## How to evaluate the agent on LIBERO and its robustness variants with EMERGE Policy Repo: [EMERGE-Policy/EMERGE-Policy](https://github.com/EMERGE-Policy/EMERGE-Policy) | Difficulty: Advanced | Time: 4 hours | Basis: From the README Follow the README evaluation guides to test EMERGE Policy on standard LIBERO, LIBERO-Plus, and LIBERO-Pro. **Who it is for:** A robotics researcher comparing embodied policies who needs repeatable benchmark numbers across LIBERO suites. **The problem:** Benchmarks are spread across several suites with separate setup, and robustness suites need extra assets. **The outcome:** The researcher runs the evaluation scripts for the chosen suite and collects results for the EMERGE Policy agent. ### Steps 1. Set up the repository, the EmergePolicy environment, and the LIBERO-Plus assets as described in the README Installation section. 2. Download the OpenPI and VGGT checkpoints into checkpoints/ as listed in the Model Checkpoints table. 3. Open the guide for the suite you need: scripts/Libero_eval/README.md, scripts/LiberoPlus_eval/README.md, or scripts/LiberoPro_eval/README.md. 4. Start the model services with `bash scripts/model_server/start_external_model_servers.sh --services openpi,vggt,sam3` using the GPU variables from the Quick Start. (inferred, check the README) 5. Follow the suite guide to run the evaluation and save the output. ### Prompt to paste into your agent ```text In the EMERGE-Policy repo, read scripts/LiberoPlus_eval/README.md and run the evaluation it describes for the suite "". Expected output is an evaluation result file or summary from that suite. Verify that the LIBERO-Plus assets are present under third_party/libero_plus before starting. ``` ### Caveat The README says LIBERO-Pro evaluation still runs on the VLA version only, so the WAM backend cannot be used for that suite. 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/emerge-policy-emerge-policy/benchmark-policy-on-libero-variants/