Many groups are researching a biped walking robot, although they have different objectives in mind such as applications of modern control theory, the study of mechanisms, or practical use to medical fields. The authors and others are engaged in studies of biped walking robots, with "human form" as a key word, from two points of view: one is human engineering, and the other is toward the development of anthropomorphic robots. The authors and others have obtained the following results to date. In 1984, the authors and others succeeded in achieving a dynamic biped walking of 1.3 [s/step] by the use of a hydraulic biped walking robot, WL-10 RD (Waseda Leg-No.10 Refined Dynamic). From 1986 to 1994, the authors developed hydraulic biped walking robots of the WL-12 series that compensated for lower limbs moment using an upper body and realized not only fast dynamic biped walking (0.54 [s/step] with a step length 0.3 [m]) but also walking on an unknown surface. In 1995, the authors developed an electrical powered biped walking robot WL-13, in which each leg joint is driven antagonistically via a rotary-type, nonlinear spring mechanism, and realized quasi-dynamic walking (7.68 [s/step] with a 0.1 [m] step length). In the current research concerning a biped walking robot, however, there is no developed example of a life-size biped walking robot which can perform manipulation and locomotion by dynamically coordinating arms and legs. Therefore, the authors proposed the construction of a biped humanoid robot that has a hand-arm system, a head system with visual sensors, and antagonistic driven joints using a rotary-type non-linear spring mechanism, on the basis of WL-13. We designed and built it. In addition, as the first step to realize the dynamically coordinated motion of limbs and trunk, the authors developed a control algorithm and a simulation program that generates the trunk trajectory for a stable biped walking pattern even if the trajectories of upper and lower limbs are arbitrarily set for locomotion and manipulation respectively. Using this preset walking pattern with variable muscle tension references corresponding to swing phase and stance phase, the authors performed walking experiments of dynamic walking forward and backward, dynamic dance and carrying, on a flat level surface (1.28 [s/step] with a 0.15 [m] step length). As a result, the efficiency of our walking control algorithm and robot system was proven. In this paper, the mechanism of WABIAN and its control method are introduced.
