# AI use cases for Evo Bot (ABC12345anouys/EvoBot)

Repo: [ABC12345anouys/EvoBot](https://github.com/ABC12345anouys/EvoBot) | Use cases: 4 | Page: https://dedupedai.com/use-cases/abc12345anouys-evobot/

EvoBot is a Python agent and reusable skill library that runs manipulation tasks in the LIBERO robot benchmark, reporting 25 of 30 tasks passed.

This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL.

- Reproduce the 25 of 30 LIBERO evaluation result: https://dedupedai.com/use-cases/abc12345anouys-evobot/reproduce-libero-30-task-evaluation/
- Watch one LIBERO task learn across retry attempts: https://dedupedai.com/use-cases/abc12345anouys-evobot/watch-one-libero-task-learn-over-attempts/
- Resume the LIBERO agent runner and skip passed tasks: https://dedupedai.com/use-cases/abc12345anouys-evobot/resume-libero-agent-runner-across-suites/
- Regenerate LIBERO task specs with deterministic ordering: https://dedupedai.com/use-cases/abc12345anouys-evobot/regenerate-task-specs-without-api-key/

## How to reproduce the 25 of 30 LIBERO evaluation result with Evo Bot

Repo: [ABC12345anouys/EvoBot](https://github.com/ABC12345anouys/EvoBot) | Difficulty: Advanced | Time: 3 hours | Basis: From the README

Run all 30 LIBERO tasks from the three suites and check the pass count against the README.

**Who it is for:** A robotics researcher who wants to confirm the reported 83 percent success rate on the LIBERO benchmark before building on it.

**The problem:** Published benchmark numbers are hard to trust until the same tasks run on your own machine with the same settings.

**The outcome:** A full verification run over libero_spatial, libero_object and libero_goal that reports how many of the 30 tasks passed.

### Steps

1. Create the environment with `conda create -n darwin python=3.10 -y && conda activate darwin`
2. Install the project with `pip install -e .` and the base dependencies with `pip install mujoco h5py numpy pyyaml`
3. Install LIBERO following its official instructions, then set `export PYTHONPATH=$PWD:/path/to/LIBERO` and `export MUJOCO_GL=egl`
4. Run the full verification with `bash scripts/verify_30.sh 8 verify`
5. Compare the pass count with the README result of 25 of 30 and read the attempt records in logs/libero_agent_progress.json

### Prompt to paste into your agent

```text
Work in the EvoBot repo at <repo path>. Set up the Python 3.10 environment and PYTHONPATH from the README, then run `bash scripts/verify_30.sh 8 verify`. Report the number of passed tasks out of 30 and list any failed task names. Verify that logs/libero_agent_progress.json has an entry for every task you ran.
```

### Caveat

The README describes a 30-task run at up to 8 attempts per task, and the LIBERO source and MuJoCo setup must be installed separately before any run works.

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/abc12345anouys-evobot/reproduce-libero-30-task-evaluation/

## How to watch one LIBERO task learn across retry attempts with Evo Bot

Repo: [ABC12345anouys/EvoBot](https://github.com/ABC12345anouys/EvoBot) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README

Run a single LIBERO task with a persistent simulator and see how failures change the next attempt.

**Who it is for:** A researcher studying how a skill-based agent recovers from failure, who wants to inspect one task closely.

**The problem:** Reading summary scores hides how the agent adjusts parameters, grasp positions or methods after each failed attempt.

**The outcome:** A run on one task, such as libero_spatial:0, that logs each attempt, stops at the step cap and keeps the video unless disabled.

### Steps

1. Set up the environment and PYTHONPATH as in the README quick start
2. Run `PYTHONPATH=$PWD:/path/to/LIBERO MUJOCO_GL=egl python -u scripts/ipc_learn.py libero_spatial:0 --max-attempts 30 --sock /tmp/libero.sock` (inferred, check the README)
3. Watch the attempt output and note the failure reason the agent reports for each failed try
4. Check the progress file logs/libero_agent_progress.json for the recorded result of the task

### Prompt to paste into your agent

```text
In the EvoBot repo at <repo path>, run the task libero_spatial:0 with scripts/ipc_learn.py and --max-attempts 10. Report how many attempts were used, whether the task passed, and the failure reasons you saw. Verify that the progress entry for libero_spatial:0 matches the outcome you report.
```

### Caveat

The step cap defaults to 7000 simulation steps and the wall-clock timeout only triggers when the program stalls, so a single attempt can run for a long time on a slow 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/abc12345anouys-evobot/watch-one-libero-task-learn-over-attempts/

## How to resume the LIBERO agent runner and skip passed tasks with Evo Bot

Repo: [ABC12345anouys/EvoBot](https://github.com/ABC12345anouys/EvoBot) | Difficulty: Intermediate | Time: 2 hours | Basis: From the README

Run the agent over all three LIBERO suites so that tasks already passed are skipped on each rerun.

**Who it is for:** A researcher running a long batch of LIBERO tasks on a machine that may be interrupted.

**The problem:** Restarting a full benchmark batch wastes time when many tasks already passed in an earlier session.

**The outcome:** A batch run that processes only unfinished tasks and records progress for each one.

### Steps

1. Set up the environment and export `PYTHONPATH=$PWD:/path/to/LIBERO` and `MUJOCO_GL=egl` as in the README
2. Run `PYTHONPATH=$PWD:/path/to/LIBERO MUJOCO_GL=egl python -u -m darwin.agents.libero_runner --suites libero_spatial,libero_object,libero_goal` (inferred, check the README)
3. After the run, open logs/libero_agent_progress.json and count the tasks marked as passed
4. Rerun the same command later to continue from the unfinished tasks

### Prompt to paste into your agent

```text
In the EvoBot repo at <repo path>, run darwin.agents.libero_runner for the three suites with the PYTHONPATH and MUJOCO_GL settings from the README. Report how many tasks were skipped as already passed and how many were newly run. Verify the counts against the entries in logs/libero_agent_progress.json.
```

### Caveat

The README says the progress file keys follow a pattern, but the exact skip logic beyond passed tasks is not described, so check the progress file yourself after each rerun.

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/abc12345anouys-evobot/resume-libero-agent-runner-across-suites/

## How to regenerate LIBERO task specs with deterministic ordering with Evo Bot

Repo: [ABC12345anouys/EvoBot](https://github.com/ABC12345anouys/EvoBot) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README

Rebuild subgoal orders for LIBERO tasks using only the automatic sort, with no model calls.

**Who it is for:** A developer who wants reproducible subgoal orders for LIBERO tasks and has no DeepSeek API key.

**The problem:** Model-generated task decompositions can differ between sessions, which makes debugging harder.

**The outcome:** Task specification YAML files under darwin/skills/configs/task_specs/ regenerated with the dependency-based order.

### Steps

1. Activate the environment from the README setup
2. Run `python scripts/gen_task_specs.py --no-llm` to regenerate all 30 tasks without calling a model
3. Open one file in darwin/skills/configs/task_specs/ and check that the source field reads deterministic
4. Run a single task to confirm the subgoal order runs without error, for example with `python scripts/gen_task_specs.py --tasks libero_goal:7 --no-llm` (inferred, check the README)

### Prompt to paste into your agent

```text
In the EvoBot repo at <repo path>, run `python scripts/gen_task_specs.py --no-llm`. Report how many task spec files were written and confirm that each has source set to deterministic. Verify by opening one spec and comparing its subgoals list with the reference_order field.
```

### Caveat

The README says the runtime does not call a model, but regenerating specs with the model needs DEEPSEEK_API_KEY and falls back to deterministic ordering if the key is missing.

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/abc12345anouys-evobot/regenerate-task-specs-without-api-key/