This document gives an excellent explanation of tuning ROS parameters: http://kaiyuzheng.me/documents/navguide.pdf
This summary assumes you’ve read that. I’m just consolidating some information from that document.
NOTE: The only way to find all of the parameters is to use rqt_reconfigure, because many parameters are not in the files, but just set to defaults.
rosrun rqt_reconfigure rqt_reconfigure
or
~/catkin_ws/src/ArloBot/scripts/runReconfigure.sh
filename
settings inside file
costmap_common_params.yaml
robot_radius: 0.25
Padding it a bit helps it to avoid bumping into door frames
NOTE: If your robot is always overshooting goals, especially rotationally, your acceleration parameters are probably wrong. Read the above doc for some insight on how to test and tune those.
dwa_local_planner_params.yaml
The top of the file has all of the max/min speed and acceleration settings.
move_base_params.yaml
base_local_planner: "dwa_local_planner/DWAPlannerROS"
base_global_planner: "global_planner/GlobalPlanner"
global_planner_params.yaml
old_navfn_behavior: false
use_quadratic: true
use_dijkstra: true
use_grid_path: false
global_planner_params.yaml
See page 5 for good pictures and 6 for warnings.
lethal_cost: 253
neutral_cost: 66 #NOTE: I wonder if setting this to 1 would make it take wider corners?
cost_factor: 0.55
dwa_local_planner_params.yaml
sim_time: 4.0
Sim Time is how far into the future DWA simulates each possible path. High values are CPU intensive.
Too low and the paths created are not of good quality.
Too high and it creates long inflexible paths, although it keeps updating them so it should work.
Turtlebot uses like 1.7. Page 8 of the referenced PDF suggests 4.
On Arlobot, a high sim_time seems to work if the robot is headed in the right direction.
However, a good way to show the problems is start the robot facing in the opposite direction:
It starts to rotate toward the global path, and then at some point, it plans a wide arc to “drive to” the global path, instead of just rotating to it. This usually throws the robot out into the room where it is colliding with everything.
So a very short sim_time, like 1.0, seems to keep it just sticking to the global path.
dwa_local_planner_params.yaml
vx_samples: 20
vy_samples: 1 # This don’t apply if your robot cannot move sideways.
vtheta_samples: 40
These can increase CPU load, but can improve quality. See Turtlebot’s parameters for a minimum.
sim_granularity: 0.025 # Not in the file, but probably in rqtreconfigure. Default is fine. See Page 9 for details.
dwa_local_planner_params.yaml
As of Melodic the defaults for path_distance_bias, goal_distance_bias, and occdist_scale have changed due to bugs in how the previous versions calculated https://github.com/ros-planning/navigation/pull/763/files
#path_distance_bias: 64.0 # How closely to follow the global path.
#goal_distance_bias: 20.0 # Weight given to goal regardless of path
#occdist_scale: 0.02 # Weight given to avoiding obstacles
Too high of occdist_scale will cause the robot to cycle indecisively.
With a low sim time, a path_distance_bias of 64.0 seems to work.
Higher helps in some situations, but in others it will spend more time getting “back to the path” than moving along it.
oscillation_reset_dist: 0.05 # 0.05 - how far to travel before resetting oscillation flags
If the robot is not getting anywhere, it will initiate recovery behaviors.
This is the distance it must progress to be considered “not oscillating” anymore.
dwa_local_planner_params.yaml
yaw_goal_tolerance: 0.3 # 0.05
xy_goal_tolerance: 0.15 # 0.10
# latch_xy_goal_tolerance: false
This is how we get that blue coverage of the local planner area to push the robot into the middle of the room. See page 12 for an excellent picture. costmap_common_params.yaml
inflation_layer:
enabled: true
cost_scaling_factor: 2.58
inflation_radius: 1.75
global_costmap_params.yaml
local_costmap_params.yaml
resolution: 0.01
In theory setting it too low can cause gaps of “unknown” space in the maps, because the laser cannot fill them in,
but when I set it to 0.05 I got gaps in the wall instead!