Ways to Speed up Planning¶

This page attempts to answer a commonly asked question: The planning algorithm is SLOW, what can we do?!

• Create a databases.linkstatistics model and tune the xyzdelta that controls the joint resolutions/weights. Can also manually tweak the joint resolutions and weights via .KinBody.SetDOFWeights and .KinBody.Joint.SetResolution.
• Each planner has the following parameters in its Planner.PlannerParameters structure that can be tweaked:

• Planner.PlannerParameters._fStepLength - max step length for each tree extension
• Planner.PlannerParameters._nMaxIterations - max number of iterations to search

• Each planner usually has a post-processing planner that does the smoothing/retiming of the path. The parameters to control are

• Planner.PlannerParameters._sPostProcessingPlanner
• Planner.PlannerParameters._sPostProcessingParameters

_sPostProcessingParameters = "parabolicsmoother";
_sPostProcessingParameters = <_nmaxiterations>100</_nmaxiterations>";

• The smoothers has several other parameters, which are written in the TrajectoryTimingParameters structure inside the rplanners plugin
• Add as many pairs of links of the robot that will never collide as “adjacent links” (both OpenRAVE XML and COLLADA specifications support this).
• Temporarily disable KinBody objects via KinBody.Enable() (False) for kinematic bodies that are too far from the robot or robot could never collide with them.
• Use primitive shapes (boxes/cylinders/spheres) as much as possible!