robot_template — Cookiecutter Scaffold

Cookiecutter 3-mode out-of-the-box

The fast lane to a new robot. Generates a complete package that drops straight into the robot_agent runtime and is immediately controllable from the robotapp dashboard.

pip install cookiecutter
cookiecutter https://github.com/mtbui2010/robot_template
cd <package_name>
make install && make run
# → uvicorn on :8001, ready to drive from https://robot.aistations.org

What it emits

<package_name>/
├── Makefile                  install · run · cli · doctor · terminate
├── pyproject.toml            [project.scripts] <package_name> = ...
├── .vscode/launch.json       F5 → debug uvicorn
└── <package_name>/
    ├── main.py               create_app('<package_name>', data_dir=...)
    ├── __main__.py           CLI entry → robot_agent.cli:main
    ├── configs/
    │   ├── skills_config.py  SKILL_CONFIGS — auto-populated from your list
    │   ├── tasks.py          ARM_CONFIGS, ENV stubs
    │   └── guide.py          planner guide stub
    ├── skills/
    │   ├── __init__.py       auto_wrap_skills(SKILL_CONFIGS, pkg=...)
    │   └── <skill>.py        one runnable mock per skill name you gave
    └── template_skills/      three reference patterns, NOT auto-registered
        ├── grip_pyconnect.py     Pattern 1 — pyconnect NodeAgent (90 % of cases)
        ├── grip_pure_ros2.py     Pattern 2 — raw rclpy + custom QoS / feedback
        └── grip_external.py     Pattern 3 — separate process / language / host

Prompts

Prompt	Default	Purpose
project_name	My Robot	Human-readable name
package_name	derived	Importable Python package
description	A robot agent ...	pyproject.toml summary
author	Your Name	pyproject.toml metadata
version	0.1.0	Initial version
skills	find,pick,place	One mock file per skill
default_port	8001	Makefile port pin
robot_agent_relpath	../robot_agent	Editable-install path
pyconnect_relpath	../pyconnect	Editable-install path

Why a template instead of forking kcare_robot?

kcare_robot is a real implementation — 23 skills tuned for a specific KAAIR arm, Femto Bolt head, and D405 wrist camera. Forking it means inheriting (and then pruning) those quirks. The template gives a clean baseline that honours the same contract with robot_agent:

# Everything robot_agent needs from your package:
SKILL_CONFIGS: dict[str, tuple[str, str]]   # name -> (module_path, func_name)

That’s it. One dict, one convention, infinite robots.

Three ways to drive a fresh robot — out of the box

HTTP

make run → uvicorn on the port you chose → open the hosted dashboard, paste the URL, drive it.

CLI

<package_name> find::apple — auto-registered as a console-script.

Python

from <package_name>.skills.find import find; find('apple')