Workspace Trajectory Generation Method for Humanoid Adaptive Walking With Dynamic Motion Primitives

Chengju Liu,Wandong Geng,Qijun Chen,Ming Liu

doi:10.1109/access.2020.2976098

Abstract

To properly function in real-world environments, a humanoid robot must be able to adapt its walking gait to new situations. In this paper, an adaptive bipedal walking control method that uses sensory feedback to modulate dynamic movement primitive (DMP) parameters is presented. This work addresses the challenge of adaptive locomotion by implementing DMPs in the workspace of a humanoid robot. This workspace formulation allows new movements to be created such that the DMP parameters, including the stride, height of the hip joint, foot clearance and forward velocity, are directly related to the walking pattern. One set of DMPs is applied to generate the foot trajectory, and a second set is used to generate the CoM (centre of mass) trajectory in an online fashion. Sensory feedback information is utilized to modify the generated CoM and foot trajectories to improve the walking quality. Furthermore, a staged evolutionary algorithm (EA) is designed to optimize the parameters of the control system to enhance the walking performance. The presented control strategy is demonstrated through simulations and real experiments that focus on the adaptation of the robot's walking pattern over sloped terrain.

Highlights

Biped locomotion is complex to analyze in general
Our work considers how to incorporate sensory information to improve the adaptability of workspace dynamic movement primitive (DMP)-based bipedal walking
The whole control architecture is set as shown in Fig. 17; the main central processing unit module, i.e., the DMP network, is used for the online generation of joint control signals, and the components indicated by dashed lines form the feedback loop that is used to modulate the DMP outputs to dynamically achieve adaptive walking on irregular ground

Summary

INTRODUCTION

Biped locomotion is complex to analyze in general. Most of the models in the literature for biped locomotion utilize hybrid dynamics and various control strategies have been proposed [1]–[8]. Unlike in the abovementioned works, the DMPs used for this purpose must be learned in the task space; we call this approach the ‘‘workspace DMP-based method’’ This idea of using DMPs in the task space, which can be related to the task parameters, was first presented by Pastor et al [35] and Rosado et al [36] in the context of robot manipulation and bipedal walking. Using DMPs, Bockmann et al realized adaptive kick motions for the NAO robot [38] In these strategies, DMPs are utilized as trajectory representations learned in the task space from a single demonstration. THE PROPOSED WORKSPACE DMP-BASED CONTROL SYSTEM This section demonstrates the structure and implementation of the presented adaptive bipedal walking control method.

DMP-BASED TRAJECTORY FORMULATION

WALKING PATTERN EVOLUTION

INCLINED TERRAIN ADAPTIVE WALKING

Findings

CONCLUSION AND FUTURE WORK