AgenticROS Just Got Much Easier to Install, Run, and Demo
Over the past few days, we’ve made a big push to make AgenticROS easier to try, easier to configure, and easier to use with both real and simulated robots.
Reminder: AgenticROS creates a glue between AI Agents (OpenClaw, Claude, and Gemini) and ROS2 making Physical AI simple and fun.
The biggest update is this:
>npx agenticros
That’s now the starting point (https://www.npmjs.com/package/agenticros).
With a single command, AgenticROS can install itself on a ROS-based system and launch an interactive setup wizard. No manual repo cloning. No stitching together setup scripts by hand. The CLI walks you through first-time setup, robot configuration, OpenClaw integration, skills, simulation, logs, health checks, and clean shutdowns.
The new agenticros command also works as a real CLI, so you can run things directly:
- agenticros init
- agenticros up real
- agenticros up sim-amr
- agenticros up sim-arm
- agenticros skills
- agenticros status
- agenticros doctor
- agenticros down
This matters because robotics demos often fail before the robot ever moves. The hard part is usually the setup surface area: ROS workspace builds, config files, transport modes, gateway plugins, API keys, background processes, logs, and matching the right launch files to the right environment.
AgenticROS now gives that whole workflow one front door.
New Simulation Demos
We also added simulation support so developers can experiment with AgenticROS without needing physical robot hardware on day one.
There are now two main simulation paths:
1. AMR Simulation
You can launch a simulated 2-wheel autonomous mobile robot in Gazebo and RViz:
>agenticros up sim-amr
The AMR simulation includes ROS-side topics for movement, sensors, camera/depth data, lidar, odometry, and TF. That means the same AgenticROS tools used against a real robot can be exercised end-to-end against a virtual one.
This is especially useful for testing agent workflows like:
- listing ROS topics
- publishing velocity commands
- reading sensor data
- taking camera snapshots
- sampling depth
- testing follow-me style behaviors
- validating agent-to-ROS connectivity
2. Robotic Arm Simulation
We also added a simulated 6-DOF robotic arm path:
>agenticros up sim-arm
The arm is UR5e-shaped and exposes per-joint position command topics through ROS/Gazebo bridges. It can be launched with Gazebo and RViz so developers can test arm command flows, inspect TF and robot state, and start building toward more advanced manipulation workflows.
This gives AgenticROS a second major robot category beyond mobile bases: manipulation.
Why This Update Matters
AgenticROS is about making ROS-powered robots accessible to AI agent platforms.
The project already connects ROS 2 robots to tools like OpenClaw, Nvidia's NemoClaw, Claude Code, Claude Desktop, Gemini CLI, and MCP-based agent workflows. But for that ecosystem to be useful, developers need to get from “fresh machine” to “robot responding” quickly.
That’s what this update is about.
You can now:
- install with npx agenticros
- configure through a guided wizard
- launch a real robot stack
- launch AMR simulation
- launch robotic arm simulation
- manage skills
- check system health
- tail logs
- stop everything cleanly
All from one CLI.
A Better On-Ramp for Agentic Robotics
The goal is to make AgenticROS feel less like a research wiring project and more like a usable robotics developer platform.
If you have ROS 2 and Node.js 20+, you can now try AgenticROS with:
>npx agenticros
From there, choose real robot, AMR simulation, or arm simulation.
This is a big step toward making agentic robotics easier to build, test, demo, and share.
More soon. The next layer is making these simulated and real robot workflows increasingly skill-driven, so agents can move from low-level ROS tool use toward higher-level robot behaviors.