特異点を避けつつ高速収束する実時間逆運動学の高次微分ソルバ
Higher-Order Kinematic Derivatives for Aspect-Preserving Real-Time Numerical Inverse Kinematics
直列ロボットの高次運動学微分を用い、特異点への接近を予測してステップを制限することで、姿勢の連続性を保ちながら4次・5次収束する実時間逆運動学ソルバを実現した。
詳しい要約
1. どんなもの？
2. 先行研究と比べてどこがすごい？
3. 技術・手法の肝は？
4. どうやって有効だと検証した？
5. 議論はある？
6. 次に読むべき論文は？
※ AIが要旨から生成した要約です。正確性は原文をご確認ください。
著者: Steffan Lloyd
分類: cs.RO
原文アブストラクト
This paper leverages the higher-order kinematic derivatives of serial robots to achieve efficient, aspect-preserving real-time inverse kinematics. First, a family of higher-order solvers is derived, with a necessary orientation-error correction map, yielding the QuIK4 and QuIK5 methods with fourth- and fifth-order convergence. Second, an aspect-preserving step limiter extrapolates the Jacobian determinant along the step to predict and avoid singularities, shortening steps only when they head toward a singularity rather than when the current iterate is ill-conditioned. Third, a Hermite-interpolation certifier verifies the aspect of each answer, and a certified restart algorithm recovers failures from proven same-aspect seeds. The proposed methods are shown to outperform prior methods, reducing failures by up to four orders of magnitude and nearly eliminating aspect switching at tracking-level errors. A C++ library is released with ROS 2 integration.