Development of an Omnidirectional Walking Engine for Full-sized Lightweight Humanoid Robots

In this paper, we propose and demonstrate an omnidirectional walking engine that achieves stable walking using feedback from an inertial measurement unit (IMU). The 3D linear inverted pendulum model (3D-LIPM) is used as a simplified model of the robot, the zero moment point (ZMP) criterion is used as the stability criterion, and only the feedback from the IMU is utilized for stabilization. The proposed walking engine consists of two parts; the omnidirectional gait generator, and the stability controller. ZMP equations, derived based on the 3D-LIPM, are used in the omnidirectional gait generator. The solutions of the differential equations are directly used which reduces the computation cost compare to other existing methods. Two kinds of feedback controllers are implemented for the stability controller; one is the indirect reference ZMP controller, and the other is the indirect joint controller. The walking engine is tested on a lightweight, fullsized, 21-degree-of-freedom (DOF) humanoid robot CHARLI-L (Cognitive Humanoid Autonomous Robot with Learning Intelligence, version Lightweight) which stands 141 cm tall and weighs only 12.7 kg. The design goals of CHARLI-L are low development cost, lightweight, and simple design, which all match well with the proposed walking engine. The results of the experiments present the efficacy of our approach. INTRODUCTION Bipedal locomotion for humanoid robots is a challenging task, especially for tall, adult sized robots. Though there are many approaches to achieve stable walking, currently the most successful and practical implementations utilize the zero moment point (ZMP) criterion ([1-6]). Some of these robots are capable of climbing stairs or even running ([7-9]). Most of the successful humanoid robots in this size class are very expensive, heavy, and complicated in design, making them difficult to be used as an affordable and safe robotic platform for research and education. CHARLI-L (Cognitive Humanoid Autonomous Robot with Learning Intelligence, version Lightweight) is a low-cost humanoid robot developed in the Robotics and Mechanisms Laboratory (RoMeLa) at Virginia Tech ([10]). To achieve the design objectives of a lightweight and low cost system, only an inertial measurement unit (IMU) is used (vs. additional forcetorque (F/T) sensors) as the main feedback sensor for bipedal locomotion. Additionally, since the on-board, single-board computer utilizes most of its computational power for vision processing, motion planning, and autonomous behaviors, a walking engine with low computational cost that utilizes only an IMU for feedback was needed. For miniature humanoid robots with a height of less than 60 cm, such as the DARwIn (Dynamic Anthropomorphic Robot with Intelligence) series humanoid robots [11], omnidirectional walking can be achieved based on an open loop motion planner and thus requires much less computational power than full-sized humanoid robots. For CHARLI-L, the challenge is to develop a walking engine that utilizes low computational power for a full-sized humanoid robot. This is possible partially due to the fact that CHARLI-L is lightweight and has no F/T sensors which can simplify the algorithms making it less computational intensive than that for a heavier full-sized humanoid robot. To achieve this goal, a simplified model of biped robots such as the linear inverted pendulum model (LIPM) ([12]) is used for the walking engine. LIMP models the robot as a point mass at a constant height while connected to the ground by a