Earlier, the range for RRT* was set to 0.0 which, internally in OMPL, ended up being set to 7.079055 for ADA (determined through the debug logs). This resulted in some plans that made needlessly large motions, like reaching across the user and overshooting in order to go from the resting configuration to the staging configuration.
Smaller range increases planning time, but results in more concise motions. range: 3.5 had better motions but some still overshot. range: 1.75 had good motions, but often failed to find a plan within 2-2.5 secs. range: 2.5 empirically seemed like a good compromise.
This PR also updates the moveit configuration to handle workspace bounds ada_feeding#82, by making the initial position the home position, and rotating MoveIt to have visibility inside the walls.
Further Improving Plan Quality
As an aside, here are some tuning steps we should consider doing in the future to further improve planning:
Consider changing the termination criteria to ExactSolution. If that empirically results in similar plans, then stick with that and bump up the planning time, so that the planner returns as soon as it finds a solution as opposed to continuing to optimize as long as it has time.
Description
Motivated by ada_feeding#72.
Earlier, the
rangefor RRT* was set to 0.0 which, internally in OMPL, ended up being set to 7.079055 for ADA (determined through the debug logs). This resulted in some plans that made needlessly large motions, like reaching across the user and overshooting in order to go from the resting configuration to the staging configuration.Smaller
rangeincreases planning time, but results in more concise motions.range: 3.5had better motions but some still overshot.range: 1.75had good motions, but often failed to find a plan within 2-2.5 secs.range: 2.5empirically seemed like a good compromise.This PR also updates the moveit configuration to handle workspace bounds ada_feeding#82, by making the initial position the home position, and rotating MoveIt to have visibility inside the walls.
Further Improving Plan Quality
As an aside, here are some tuning steps we should consider doing in the future to further improve planning:
ExactSolution. If that empirically results in similar plans, then stick with that and bump up the planning time, so that the planner returns as soon as it finds a solution as opposed to continuing to optimize as long as it has time.