Whole body motion generation with centroidal dynamics of legged robots using sequential bounds tightening of McCormick envelopes

Abstract

In this paper we introduce a Sequential Convex Programming (SCP) algorithm for the motion generation with the centroidal dynamics of legged robots using a sequential bounds tightening of McCormick envelopes strategy to cope with the nonconvexity of the problem (related to bilinear terms). Therefore, the proposed SCP algorithm is initialized with relaxed McCormick envelopes and then their bounds are sequentially tightened around the current estimate of the solution enforcing this way convergence to a feasible point. The SCP algorithm solves a quadratic program at each iteration by an interior point method. Additionally, the proposed SCP algorithm is alternated with an inverse kinematics algorithm to achieve the whole body motion generation. Finally, extensive numerical experiments show the effectiveness of the proposed algorithm in generating highly agile motions such as trotting, bounding, stotting and running for humanoid and quadruped robots.