A new planner helped a seven-joint robot arm navigate a maze while keeping its tool tip on a fixed plane. In a physical test with moving obstacles, ReVAMP reached collision-free routes in a median 11.95 milliseconds, compared with 146.19 milliseconds for the earlier McVAMP method. The executed path averaged 0.22 millimeters off the plane because of controller error, according to the September 24 preprint.
Some robot tasks constrain the end effector, the gripper or tool at the arm's tip. For example, it may need to stay level while carrying an object. Sampling joint positions at random often produces candidates that violate that condition, forcing planners to correct them before checking for collisions.
ReVAMP starts with a smaller task-specific space where candidate poses already meet the rule, then maps each pose to joint angles with analytic inverse kinematics, a calculation that works backward from a desired tool position. Its authors use CPU instructions to evaluate several path points in parallel. An early collision check at the tool tip can discard a path before a full robot-body check.
Both planners ran for about 60 seconds under changing-obstacle conditions on a real-time operating system with an Intel i5-6500T CPU at 2.5 GHz. The paper also reports additional evaluations using 14- and 20-degree-of-freedom systems, including a bimanual object-carrying task.
ReVAMP currently has to stay on one inverse-kinematics branch. Because an arm can reach the same tool pose with different joint arrangements, it cannot yet plan routes that require switching between those arrangements. The physical result comes from one maze setup; the authors identify branch-switching as future work.
Illustrative image: Nenad Stojković via Wikimedia Commons, used under CC BY 2.0.
