ROS2 fundamentals — the 10-minute version of a 6-hour tutorial

The single best mental model for ROS2: everything is a process talking to other processes over a deterministic message bus.

The five primitives

Nodes are processes. Each does one thing — read from a sensor, run a controller, drive a motor. Nodes can be Python or C++. They start and stop independently.

Topics are named channels for one-way streams. A camera node publishes to /camera/image_raw; any node that wants frames subscribes. Topics use a pub-sub model — many publishers, many subscribers, all loosely coupled.

Services are request/response. Useful when you want a definite reply: "give me the current state of the gripper," "spawn a turtle at (3, 4)."

Actions are services for long-running tasks with feedback. "Navigate to this point" returns periodic progress updates and a final success/fail. Most robot motion uses actions.

Parameters are per-node config — set at startup, can be changed at runtime. Like environment variables for nodes.

DDS — the bit that's different from ROS1

ROS2 runs on DDS (Data Distribution Service), an industrial pub-sub middleware standard. DDS does the discovery (nodes find each other automatically), the serialization (turning structs into bytes), the QoS (reliability, durability, latency). You don't write any of this — you write nodes; DDS handles the wiring.

This is the single biggest improvement over ROS1: no more roscore central master. Drop a new node onto the network and it just shows up.

The minimum command set

• ros2 run <pkg> <node> — start a node
• ros2 topic list / ros2 topic echo /foo — see what's flowing
• ros2 service call ... — invoke a service
• ros2 launch <pkg> <file> — start many nodes from one config
• colcon build — build the workspace

Spend an evening with these five commands on a turtlesim and you've crossed the ROS2 starting line.

Then read ROS2 and Nodes & Topics for the full reference.