Rugged Rover Update: Real SLAM, Nav2, and a Raspberry Pi 5
The rover has moved a long way from the early motor-control experiments. It is now running as a full ROS 2 Jazzy robot on a Raspberry Pi 5, with real sensors, real odometry, and Nav2 starting to behave like something I can build on.
Hardware and Firmware
The low-level platform is now based around a Teensy 4.1 running micro-ROS firmware. It handles encoder feedback, motor commands, battery voltage monitoring, status/debug output, and communication with the Sabertooth motor controller.
A few important firmware improvements went in:
- Reconnect handling for the micro-ROS agent
- Cleaner debug output so serial logs do not interfere with normal operation
- Corrected motor gearbox parameters after measuring encoder counts directly
- Battery voltage publishing and critical battery guard behavior
- Platform debug messages for runtime visibility
The motor tuning process was very hands-on: command a speed, watch encoder feedback, adjust the control behavior, and repeat until the rover moved smoothly enough for odometry and SLAM to become meaningful.
Sensors
The current sensor stack includes:
- RPLIDAR S2 for 2D laser scans
- SparkFun Razor IMU
- Intel RealSense D435 for depth sensing experiments
- Wheel encoders for odometry
The RPLIDAR and D435 uncovered the usual real-robot lessons: USB power, USB bandwidth, port stability, and device startup order all matter. The lidar became much more reliable once the Raspberry Pi power setup was fixed.
SLAM and Navigation
SLAM Toolbox is now producing usable maps. The biggest breakthrough was getting wheel odometry stable first. When odometry was wrong, SLAM was messy no matter how much I tuned the mapping parameters.
The process that worked best:
- Tune wheel odometry using measured travel and rotation tests.
- Run SLAM without loop closure until the map is locally stable.
- Reintroduce loop closure once odometry is trustworthy.
- Bring Nav2 in after the map and transforms are behaving.
Nav2 is now running on the real rover, and I have been tuning the local planner, angular velocity limits, costmaps, and transform timing. It is not finished, but it is now at the stage where the problems are tuning problems rather than fundamental bringup problems.
Simulation
The Unity simulation is still part of the workflow. It gives me a faster place to test launch files, topic remapping, fault injection, and navigation behavior before moving back onto hardware.
The real robot and simulation are now close enough that configuration drift is becoming a thing I actively watch for.
Repository Cleanup
I also spent time making the repository easier to reproduce:
- Added Raspberry Pi 5 bootstrap setup for Ubuntu 24.04 and ROS 2 Jazzy
- Updated bringup documentation
- Added common launch and debug commands
- Moved formatting/linting toward ROS 2 tooling
- Cleaned package metadata and copyright headers
Next Step
The goal now is to finish this rover as a dependable baseline platform, then build another one and start experimenting with multi-robot mapping and coordination.
The garage has officially become the test arena.
Repo: