Trajectory Reconstruction withNURBSCurves forRobot Programming byDemonstration

Service robotsrequire simple programming techniques allowing userswithlittle ornotechnical expertise tointegrate newtasks ina robotic platform. A promising solution forautomatic acquisition ofrobot behaviors isthe Programming byDemonstration (PbD) paradigm. Itsaimisto let robot systems learn newbehaviors fromahumanoperator demonstration. Thispaperdescribes avirtual reality basedPbDsystem forpick-and-place andmanipulation tasks. Thesystem re- covers smooth robot trajectories fromsingle ormultiple user demonstrations, thereby overcoming sensornoise andhuman inconsistency problems. Morespecifically, weinvestigate the benefits ofthehumanhandtrajectory reconstruction with NURBScurvesbymeansofa best-fit datasmoothing al- gorithm. Someexperiments involving object transportation while avoiding obstacles intheworkspace showtheviability andeffectiveness oftheapproach. IndexTerms-RobotProgramming byDemonstration, NURBScurves, Virtual Reality.