ROS2 Hands-On Exercises — Learning Plan

For: Engineer who reads ROS2 logs/bags daily but hasn’t built the muscle memory
Goal: Write, debug, and tune any ROS2 node from scratch; understand TF2 and Nav2 architecture

Why This Track Exists

You have 1524 lines of ROS2 theory in zephyr/study-notes/05-jetson-ros2.md. You debug Nav2, TF lookups, and QoS issues in production every week. But theory without hands-on exercises doesn’t build the “just know it” reflex.

Goal: After 3 weeks, you can write a working ROS2 node, diagnose TF failures, and read a Nav2 behaviour tree without looking anything up.

Weekly Schedule

Week 1: Core ROS2 Patterns (3 hrs)

Nodes, publishers, subscribers, timers
Services (sync) vs Actions (async with feedback)
Lifecycle nodes — why OKS uses them
QoS profiles: reliability, durability, history — when BEST_EFFORT drops frames

Exercises: exercises/01-pub-sub-timer.md, exercises/03-action-server.md

Week 2: TF2 + Time + QoS (3 hrs)

Transform tree: map → odom → base_link → sensor_frame
lookupTransform() with timeout — why it throws
Latency and time sync: use_sim_time, stamp tolerance
QoS mismatch: the silent subscription failure

Exercises: exercises/02-tf2-broadcaster.md

Week 3: Nav2 Architecture (4 hrs)

BehaviourTree XML: Navigate, ComputePath, FollowPath, Recovery
Costmaps: global vs local, inflation radius, lethal obstacles
Controller plugins: DWB, RPP — what parameters map to OKS tuning
Planner search models: Dijkstra vs A*, Dubins vs Reeds-Shepp, and when kinematics matter
Search-space intuition: why NavFn thinks in (x, y) but Smac thinks in (x, y, θ)
Hybrid-A* tuning: how minimum_turning_radius changes feasibility
Bellman-Ford intuition: why repeated relaxation works when greedy finalization fails
Robot-motion intuition: holonomic vs non-holonomic vs underactuated
How robot_localization feeds odom to Nav2