Conceptos Básicos
Efficient tools for robot control optimization.
Resumen
This work presents tools for efficient sequential hierarchical least-squares programming tailored to robot control and planning. The approach relies on approximations of non-linear hierarchical least-squares programming to a hierarchical form using Newton's method or the Gauss-Newton algorithm. A threshold adaptation strategy ensures optimality of infeasible constraints, enhances numerical stability, and avoids regularized local minima. The NADM2 solver based on nullspace projections of active constraints shows faster computation times than other solvers. The proposed methods are evaluated on test-functions and trajectory optimization for robotic systems.
Estadísticas
NADM2 consistently shows faster computations times than competing off-the-shelf solvers.
The proposed solver is computationally more efficient when the number of iterations is limited.
Sparse nullspace projections eliminate structured constraints in optimal control scenarios.
Turnback algorithm efficiently computes a basis of the nullspace without costly matrix factorization.