Path Smoothing (Shortcutting)

Post-processing a jagged planned path to cut out detours and round off corners, so the robot moves shorter and smoother.

Path smoothing is a post-processing step in motion planning. Sampling-based planners such as RRT and PRM produce paths made of many waypoints strung into a jagged line, often taking a roundabout route with back-and-forth wiggle, which would be slow and unnatural to execute as-is. The most common technique is randomized shortcutting: repeatedly pick two random points on the path and try connecting them with a shorter straight line (or a smooth curve respecting velocity and acceleration limits); if that segment is collision-free, it replaces the original path between the two points; repeating this many times keeps shortening the path. Other approaches include removing redundant waypoints and fitting a B-spline; the open-source planning library OMPL's PathSimplifier provides shortcutting, point removal, and B-spline smoothing functions. After smoothing, the path usually still needs time parameterization to assign it speed and timing before it becomes an executable trajectory. It differs from trajectory optimization in that it only makes local improvements to an existing feasible solution — cheap to compute, but with no optimality guarantee.

ExampleHauser and Ng-Thow-Hing (2010) had an arm reach under a table to grab a cup: applying 100 random shortcut attempts, each respecting velocity and acceleration limits, to a sampling planner's jagged path cut the execution time from 9.4 seconds to 4.0 seconds.

Also called

Shortcutting, Path Simplification

Related

Rapidly-exploring Random Tree · Probabilistic Roadmap · Collision Checking · Time Parameterization · Trajectory Optimization · Open Motion Planning Library (OMPL)

Sources

OMPL: ompl::geometric::PathSimplifier Class Reference
Hauser & Ng-Thow-Hing, Fast Smoothing of Manipulator Trajectories using Optimal Bounded-Acceleration Shortcuts (ICRA 2010)