dwb-controller

Set the FollowPath plugin to dwb_core::DWBLocalPlanner, define velocity and acceleration limits with vy_samples set to 1, and list critics such as RotateToGoal, Oscillation, BaseObstacle, PathAlign, and GoalDist. A complete YAML template with typical values is provided in the skill.

DWB samples discrete velocity commands and forward-simulates each one, scoring trajectories independently. MPPI samples trajectory perturbations over the full horizon and uses Boltzmann-weighted averaging. DWB is less computationally demanding but produces less-optimal trajectories in complex environments.

Oscillation typically comes from insufficient Oscillation critic scale, too many angular velocity samples, or a high angular acceleration limit. Increase the Oscillation scale, reduce vtheta_samples, or lower acc_lim_theta to damp the wobble.

The total equals vx_samples times vy_samples times vtheta_samples, so 20x1x20 yields 400 trajectories. DWB typically evaluates 200 to 600 per cycle, and enabling short_circuit_trajectory_evaluation reduces CPU cost.

DWB suits moderately complex environments as a middle ground between MPPI, which is heavier but more optimal, and Regulated Pure Pursuit, which is lighter but simpler. It is also familiar to users of the ROS1 dwa_local_planner.