Constrained CHOMP using Dual Projected Newton Method

Abstract

CHOMP is a popular trajectory optimization algorithm that uses covari- ant gradient techniques to produce high quality solutions. In its original formulation, it solves an unconstrained sequentially quadratic problem with extensions for handling equality constraints. In this paper we present an approach to solve sequentially quadratic problem with linear inequality con- straints. We present a dual projected newton method to efficiently solve this problem. The proposed method alternates between primal and dual up- dates thus leading to faster convergence than solving a constrained quadratic program at each iteration.