ICRA RoboMaster AI Challenge simulator — device market

Install

Three ways to run this device. Each one serves its fake backend: no instrument, no simulator install and no upstream checkout.

OpenAI4S (one command)

shell

openai4s lab add robots/yizhengzhang1__robomaster-simulator

Installs the device into the OpenAI4S Lab on its fake backend and starts it. First time? See setup below.

Or, in the workbench, open the Lab tab → Market → Install.

Standalone MHS endpoint

shell

git clone --depth 1 https://github.com/PKU-YuanGroup/OpenMHS.git && cd OpenMHS
PYTHONPATH=MHS/sdk/src python3 -m mhs serve robots/yizhengzhang1__robomaster-simulator --port 7400
# in a second terminal, from the same directory:
PYTHONPATH=MHS/sdk/src python3 -m mhs describe --endpoint http://127.0.0.1:7400

Serves the device as an MHS endpoint on port 7400 from an OpenMHS clone (Python 3.10 or later, standard library only); describe prints what it offers.

Any MCP host (Claude Desktop, Cursor, …)

mcp configuration

{
  "mcpServers": {
    "robomaster-simulator": {
      "command": "python3",
      "args": [
        "-m",
        "mhs",
        "serve",
        "robots/yizhengzhang1__robomaster-simulator",
        "--mcp"
      ],
      "env": {
        "PYTHONPATH": "/path/to/OpenMHS/MHS/sdk/src"
      }
    }
  }
}

Add this server to the host's MCP configuration and replace /path/to/OpenMHS with the path of your OpenMHS clone. The host starts the server and talks to it on stdin and stdout.

What it does

Measures

read_referee_state

Adjusts

Nothing declared.

Actions

pause_physics: Pause Gazebo physics with the acknowledged /gazebo/pause_physics service
unpause_physics: Resume Gazebo physics with the acknowledged /gazebo/unpause_physics service
start_game: Ask the referee node to start or stop its game loop; the response only acknowledges receipt of the request
command_turret: Publish one turret angle in radians on /jackal0/turret_position; the publication carries no execution acknowledgement
command_base_velocity: Publish one body twist on /jackal0/cmd_vel; the mecanum controller applies its own velocity and acceleration limits and has no execution acknowledgement

Names as the driver declares them. Units, limits and the safety declaration are in the package's REFERENCE.md.

First-time setup

The OpenAI4S workbench with the OpenMHS lab, from a source checkout:

shell

# OpenAI4S workbench with the OpenMHS lab (source checkout)
git clone --recursive https://github.com/PKU-YuanGroup/OpenAI4S.git
cd OpenAI4S && ./setup.sh && source .venv/bin/activate
openai4s lab add robots/yizhengzhang1__robomaster-simulator
openai4s serve        # open the Lab tab

The standalone endpoint and the MCP server need only an OpenMHS clone and Python 3.10 or later. More in the market's getting-started guide and on the OpenAI4S host page.

Source and evidence

Upstream: yizhengzhang1/robomaster-simulator
Pinned commit: ff085e4109a9
Licence: LGPL-3.0
Stars: 150
Evidence: contract_checked on the fake backend
Backends: fake

The driver package MHS/drivers/robots/yizhengzhang1__robomaster-simulator wraps the upstream code at the pinned commit. Its REFERENCE.md documents the full contract: every value, action, unit, limit and the safety declaration.