Feasibility-Aware Plan Adaptation in Humanoid Gait Generation | IEEE Conference Publication | IEEE Xplore

Feasibility-Aware Plan Adaptation in Humanoid Gait Generation

Publisher: IEEE

Abstract:

Most available schemes used for humanoid walking rely on the separation into a planning phase, typically offline, and a Model Predictive Controller (MPC). Moreover, in or...View more

Abstract:

Most available schemes used for humanoid walking rely on the separation into a planning phase, typically offline, and a Model Predictive Controller (MPC). Moreover, in order for the MPC to work in real time, simplifying assumptions are made both on the template model and on the constraints so that the underlying optimization problem is a Quadratic Programming (QP). The planner is unaware of the underlying humanoid dynamics and of any disturbance acting on the robot. We present an online Feasibility-Aware Plan Adaptation (FAPA) module which can locally adapt footsteps (positions, timings and orientation) in such a way that it guarantees feasibility of the subsequent Intrinsically Stable MPC (IS-MPC) stage. We present two versions of the proposed scheme: one with a fixed regions assignment for placing the footstep and another one where the regions are selected automatically through mixed-integer programming. Simulation results show the effectiveness of the FAPA scheme.
Date of Conference: 12-14 December 2023
Date Added to IEEE Xplore: 01 January 2024
ISBN Information:

ISSN Information:

Publisher: IEEE
Conference Location: Austin, TX, USA

Funding Agency:


I. Introduction

Humanoid robot locomotion is a complex task that involves multiple concurrent activities. It is usually tackled by breaking it down into several subproblems and solving each of them more or less independently. The first component is in general a footstep planner, which determines a sequence of footstep, e.g., leading the robot to some desired location. This sequence of footsteps must be kinematically realizable at least in terms of step lengths. The humanoid dynamics are usually accounted for in a second stage, typically based on Model Predictive Control (MPC), using a simplified robot model which is used to generate Center of Mass (CoM) trajectories. MPC, in its basic form, allows to perform realtime footstep position adaptation [1] and obtain reactive stepping so to reject pushes and impacts. However, in order to be able to formulate the optimization problem as a Quadratic Program (QP), constraints should be kept linear. For this reason, most schemes only adapt footstep positions, leaving out footstep orientation and step timing.

References

References is not available for this document.