TurtleBot4 Navigation Tuning Result
A custom TurtleBot4 clone has autonomously navigated for over 10,000 hours using optimized ROS 2 Jazzy parameters.
Developer RobotDreams has published extensive results from a year-long navigation tuning project for TB5-WaLI, a custom-built TurtleBot4 clone running on an 8GB Raspberry Pi 5. The robot has achieved remarkable reliability, operating continuously for 10,434.9 hours since January 2025 with only 4 safety shutdowns. Using ROS 2 Jazzy on Ubuntu 24.04 Server with the Nav2 stack, TB5-WaLI has completed 2,043 autonomous play sessions, traveling 2,730 meters while successfully docking 29 times at 20% battery threshold.
The tuning focused on optimizing Nav2 parameters for real-world home environments. Key adjustments included increasing the inflation_radius from 0.35 to 1.25 meters and lowering cost_scaling_factor to 1.25, which helped the robot maintain safer distances from walls and reduce 'side object distractions.' The developer also resolved persistent 'sensor origin out of map bounds' warnings by adjusting voxel grid parameters and strategically repositioning goal locations in challenging areas like laundry rooms.
A significant finding was the documentation of Google Gemini's unreliable advice for ROS configuration. The developer reported that Gemini consistently hallucinated solutions, suggested fixes that created new problems, and falsely claimed to have read documentation URLs. The project also identified DDS-related CPU spikes in FastDDS that could cause navigation failures during system monitoring, highlighting the importance of stable middleware in long-running autonomous systems.
- TB5-WaLI operated autonomously for 10,434.9 hours over 14 months with 2,730 meters traveled and 29 successful dockings
- Nav2 tuning required increasing inflation radius to 1.25m and adjusting voxel parameters to eliminate sensor boundary errors
- Documented Google Gemini providing consistently faulty ROS configuration advice that created cascading problems
Why It Matters
Demonstrates practical, long-term autonomous navigation in home environments and highlights the risks of using LLMs for complex system configuration.