Skip to main content

​
Do I need to know ROS2?

Probably not — the Agent and Skill SDK covers most projects without you ever touching a topic or service. Come down to this layer when you want raw sensor data, custom motion control, or to plug MARS into an existing ROS2 system. Everything underneath is standard ROS2, open and yours to hack on.

​
What is ROS2?

ROS2 (Robot Operating System 2) is a middleware framework for robotics. It provides:
  • Topics: Pub/sub messaging for streaming data (camera images, odometry, commands)
  • Services: Request/response calls for one-off operations (turn on lights, get status)
  • Actions: Long-running tasks with feedback (navigate to a point, execute a trajectory)
MARS runs ROS2 Humble on Ubuntu 22.04, with Zenoh as the DDS (networking) layer.

​
Architecture Overview

The Innate OS has two main package groups:

​
Brain (the “thinking” layer)

  • brain_client: The main orchestrator that runs agents, loads skills, and bridges to the Innate agent (cloud agent)
  • brain_messages: Custom message types for the brain system
  • manipulation: Behavior server for arm control using learned policies

​
Mars Bot (the “body” layer)

  • mars_bringup: Hardware initialization (cameras, battery, UART)
  • mars_arm: Arm kinematics and motion planning (MoveIt2)
  • mars_nav: Navigation stack (Nav2, SLAM)
  • mars_cam: Camera drivers (OAK-D via DepthAI)
  • mars_msgs: Custom messages for MARS hardware
  • And several more for simulation, logging, and Bluetooth provisioning

​
Common Use Cases

“I want to read raw camera data” → Subscribe to /oak/rgb/image_raw (sensor_msgs/Image) “I want to manually drive the robot” → Publish to /cmd_vel (geometry_msgs/Twist) “I want to check battery level” → Subscribe to /battery_state (sensor_msgs/BatteryState) “I want to move the arm to a specific position” → Use the /goto_js service or MoveIt2 interfaces (see advanced docs)
MARS navigation is built on Nav2 plus MARS-specific wrappers.
  • /cmd_vel drives base motion
  • /odom and /scan are core for localization and obstacle avoidance
  • App navigation modes map to ROS2-level behavior mode changes
To check that navigation signals are flowing: 
ros2 topic echo /odom
ros2 topic echo /scan
For how these modes look from the app side, see Capabilities.

​
Manipulation stack

MARS manipulation combines arm interfaces, controllers, and learned policy execution.
  • Arm state is exposed through joint-state topics
  • Motion commands can be sent through arm services/interfaces
  • Learned manipulation policies are deployed as skills in the higher-level stack
To check arm state: 
ros2 topic echo /joint_states
ros2 topic echo /mars/arm/state

​
Visualization

You can visualize ROS2 data with either RViz or Foxglove. If you are on Linux, RViz is the standard option. If you are not on Linux, use Foxglove with the Foxglove Bridge running on MARS.

​
RViz (Linux)

Use RViz to inspect frames, point clouds, LiDAR, trajectories, and robot state directly from a Linux ROS2 workstation.

​
Foxglove Bridge (non-Linux)

If you do not have Linux, run Foxglove Bridge on the robot and connect from Foxglove. Typical flow:
1

SSH into MARS

Connect to the robot over SSH on the same network.
2

Start Foxglove Bridge

Run the Foxglove Bridge node on MARS so topics are exposed over WebSocket.
3

Open Foxglove and connect

In your browser, open app.foxglove.dev and connect to the robot bridge endpoint.

​
Going Deeper

Use these ROS2 sections: