Development of a bipedal humanoid robot-control method of whole body cooperative dynamic biped walking

Abstract

Abstract:The authors have focused on the bipedal humanoid robot expected to play an active role in human living space, through studies on an anthropomorphic biped walking robot. As the first stage of developing a bipedal humanoid robot, the authors developed the human-size 35 active DOF bipedal humanoid robot "WABIAN" and the human-size 41 active DOF bipedal humanoid robot " WABIAN-R". The authors also proposed a basic control method of whole body cooperative dynamic biped walking that uses trunk or trunk-waist cooperative motion to compensate for three-axis (pitch, roll and yaw-axis) moment generated not only by the motion of the lower-limbs planned arbitrarily but by the time trajectory of the hands planned arbitrarily. Using these systems and the control method, normal biped walking cforward and backward), dynamic dance waving arms and hip, dynamic carrying of a load using its arms, and trunk-waist cooperative dynamic walking are achieved.

Walking Control Algorithm of Biped Humanoid Robot on Uneven and Inclined Floor

“…Researchers thus began to focus on dynamic walking of biped robots [7][8][9]. It is fast walking with a speed of less than 1 second per step.…”

Hybrid gait pattern generator capable of rapid and dynamically consistent pattern regeneration

“…The initial velocity should be determined so that after the time T ss elapses, the CoM will be in exactly same position y 0 . To calculate the initial velocity we can use formula (3) knowing that for the above to be true the velocity after T ss /2 should be zero.…”

“…The dynamics filter can comprise of any gait controller that accepts as input the motion reference and uses the Multibody System (MBS) dynamics to refine the reference in order to make the gait stable. One of the common motion references is the Zero Moment Point (ZMP) and there are two well known methods utilizing the MBS inside the pattern generation process: Preview Control developed by Kajita et al [8] and FFT based pattern generator developed by Takanishi et al [3].…”

“…In most of the locomotion control related methodologies, applied to the full-scale humanoid robots, the whole robot is simplified either to the single mass [1], [2] or to the set of masses [3]. Most of the methods eventually results in the gait pattern, which is the set of reference trajectories that completely define the spatial configuration of the robot in time.…”

“…However, since they do not include the full body dynamics in the motion generation process, they are prone to the modeling errors resulting from the self-motion and have to rely on the feedback control during the execution of motion. On the contrary, the multiobody dynamics based approach often relies on the notion of the Zero Moment Point (ZMP) and very accurate models of the system dynamics [6], [7], [8], [3]. The motion in this approach is generated by defining the ZMP reference, and forcing the controller to create the CoM -often approximated by waist -trajectory that would result in the real ZMP following the reference.…”