A performer's arms over one action as one trajectory optimisation, in TrajOpt's formulation (Schulman et al. 2014, "Motion planning with sequential convex optimization and convex collision checking", IJRR 33:1251): the arms' joint-space waypoints over the reach, the hold and the return are the variables, the objective and the constraints are HumanArmTrajectoryModel's, and the problem is solved by penalty sequential convex optimisation with a trust region (TrajOpt's Algorithm 1, HumanArmOptimiser) from a seed whose hold is solved by inverse kinematics and whose reach and return are a sampling planner's clear paths (HumanArmSeed), its waypoints refined where they do not bound what lies between them (HumanArmRefine). Two arms that move over one span are one problem, their probes kept clear of each other as they both move: planning one after the other is incomplete where their paths are coupled (Erdmann and Lozano-Pérez 1987, Algorithmica 2:477). A problem whose first arm is the left is solved as its mirror image, the arms on the other side, so mirrored scenes give mirrored paths. A problem still infeasible at the last penalty is refused.
HumanArmTrajectoryModel
HumanArmOptimiser
HumanArmSeed
HumanArmRefine
A performer's arms over one action as one trajectory optimisation, in TrajOpt's formulation (Schulman et al. 2014, "Motion planning with sequential convex optimization and convex collision checking", IJRR 33:1251): the arms' joint-space waypoints over the reach, the hold and the return are the variables, the objective and the constraints are
HumanArmTrajectoryModel's, and the problem is solved by penalty sequential convex optimisation with a trust region (TrajOpt's Algorithm 1,HumanArmOptimiser) from a seed whose hold is solved by inverse kinematics and whose reach and return are a sampling planner's clear paths (HumanArmSeed), its waypoints refined where they do not bound what lies between them (HumanArmRefine). Two arms that move over one span are one problem, their probes kept clear of each other as they both move: planning one after the other is incomplete where their paths are coupled (Erdmann and Lozano-Pérez 1987, Algorithmica 2:477). A problem whose first arm is the left is solved as its mirror image, the arms on the other side, so mirrored scenes give mirrored paths. A problem still infeasible at the last penalty is refused.