What this guide proves
A trained operator can localize a tracked robot on a known small map, dispatch one conservative goal and explain the expected fallback when localization, scan data or host control fails.
Hardware and scope boundary
Nav2 is an application layer. It depends on a validated COG-RPI5-8GB host, tracked motion chain, controller timeout behavior, map, odometry, TF tree and either MS200 or RPLIDAR C1 scan data. No component specification turns this into a safety-rated autonomous system.
Prepare before applying power
Use the saved, repeatable map from the SLAM guide. Establish a low maximum speed, accessible power-disconnect method, open route and operator observation. Confirm map frame, odom frame, base frame, scan topic and control-topic ownership before starting Nav2.
Procedure
- Bring up localization with the correct map and observe its pose estimate while the robot remains still. Set or confirm the initial pose only from a known physical reference.
- Inspect global and local costmaps, scan observation source, footprint and inflation parameters. These values must match measured chassis geometry and the actual scan-plane limitations.
- Send one short goal through RViz or the approved application interface. Observe command velocity, planned path, actual chassis motion and immediate stop method at all times.
- Repeat the same goal after a stack restart and after moving the robot back to the known starting point. Record whether pose convergence and path behavior are repeatable.
- Test safe failure handling in a controlled way: stop the command source or driver according to the approved procedure and verify controller-side timeout rather than improvising a fault injection.
Pass criteria and record
The robot repeatedly localizes, follows one short route at the recorded conservative limits, stops predictably and leaves a log that ties map, transform, scan driver, controller firmware and Nav2 parameters together.
Do not proceed when
Stop if localization jumps, costmaps do not reflect known obstacles, command ownership is unclear, controller timeout is unverified, or the environment contains hazards outside the validated sensing envelope.
Continue from here
Expand one environmental variable at a time. For manipulation or 3D obstacle context, validate RGB-D perception independently rather than assuming Nav2 sees every object.