Accurate and robust localization for walking robots fusing kinematics, inertial, vision and LIDAR

Abstract

In this article, we review methods for localization and situational awareness of biped and quadruped robotics. This type of robot is modelled as a free-floating mechanical system subject to external forces and constrained by whole-body distributed rigid contacts. Measurements of the state of the robot can be made using a variety of sensor information-such as kinematics (the sensing of the joint angles of the robot), contact force (pressure sensors in the robot's feet), accelerometers and gyroscopes as well as external sensors such as vision and LIDAR. This high-frequency state estimate is then passed to the control system of the robot to allow it to traverse terrain or manipulate its environment. In this article, we describe the development of an estimator for the Boston Dynamics Atlas humanoid robot. It was later adapted to the HyQ2 quadruped, developed by the Istituto Italiano di Tecnologia. Some discussion is given as to future trends while also considering briefly the relationship with biological systems.