controller-comparison

Choose based on environment and CPU budget: MPPI for narrow spaces and dynamic obstacles where trajectory optimization matters, DWB for open areas with scattered obstacles, and RPP for simple straight corridors where low CPU and predictability matter most.

MPPI with RotationShimController is recommended for indoor patrol robots. It handles narrow doorways, sharp 90-degree turns, and dynamic obstacles like people and pets, assuming the platform has sufficient CPU headroom.

MPPI with 2000 batch samples and no footprint checking takes about 20ms per cycle on x86, fitting a 50ms budget. On ARM boards like Jetson Nano or Raspberry Pi 4, expect 2-3x longer compute times, so reduce batch size or disable footprint checking.

Only the FollowPath plugin name and controller-specific parameters change; the controller_server config, costmaps, and behavior tree stay the same. Start with the baseline MPPI configuration, then tune critic weights iteratively with visualization enabled.

Yes, Nav2 supports multiple controller plugins in the controller_plugins list, such as MPPI for general path following and Graceful for docking. The behavior tree selects which controller handles each segment via the controller_id field in the FollowPath action.

Avoid MPPI when CPU is severely constrained, when paths are mostly straight corridors where RPP suffices, or when you need highly predictable auditable behavior, since MPPI's weighted-average trajectory can shift between cycles.