Editorial: Towards Real-World Deployment of Legged Robots

摘要

part of its weight up while walking. This novel design provides the advantage of being extremely safe and 18 low-cost. The design of the robot incorporates tendon wires to actuate the articulations attached to their 19 thin linkages. The robot employs a data-driven neural network approach for performing walking behaviors.Employing such approach enables to handle the underactuated nature of the robot since BALLU2 has only 21 2 actuated degrees of freedom per leg. In order to determine the transition function for walking, the authors 22 perform a correlation study to chose significant state variables. Then, they train a neural network to learn 23 transition times on both single and double support phases. Walking is accomplished in the real system at a 24 speed of 0.18m/s using an RGBD camera for onboard state estimation.