Swept laser based 3D pose detection of the swinging robot Acroboter Roland Reginald Zana Department of Applied Mechanics Budapest University of Technology and Economics Budapest, Hungary zana_r@mm.bme.hu Ambrus Zelei MTA-BME Research Group on Dynamics of Machines and Vehicles Budapest, Hungary zelei@mm.bme.hu Abstract—Domestic robots have been attracted growing interest. The Acroboter is a ceiling based crane-like robot concept, which utilizes the entire cubic volume of indoor environments. The benefits of cranes and drones are combined in our concept: the fan actuators make the robot agile in the horizontal direction, and the hoist rope provides the suspending force without power input when the robot is still. The lightweight structure is composed of 3D printed components and carbon balsa composite sandwich base plate. Our novel pose detection approach is to attach the HTC Vive measurement unit rigidly to the swinging unit of the robot. The working principle is based on sweeping beams of laser and time measurement providing high accuracy. We demonstrate in experiments that this novel pose detection concept is feasible for robots and it is accurate and fast enough for achieving stable trajectory tracking control of our crane-like manipulator. Keywords—cable suspended robot, model predictive control, trajectory tracking, underactuation, multibody dynamics, pose estimation I. INTRODUCTION Robots and robot control algorithms are widespread and greatly developed: the structure and the control of industrial robots haven’t change in the past few decades. However, new robots with specific purposes are continuously developed, such as legged robots, flexible robots, human- friendly lightweight robots, cable tethered robots, ceiling based robots and flying/underwater robots [1]. Model predictive control (MPC) is an efficient approach for the motion control of these state-of-the-art robots because of its accuracy, good trajectory tracking performance and relatively low computational effort. In case of mechanical systems, MPC basically means that the torques of the servo drivers are computed in such a way that results the prescribed output motion. The complete dynamical model of the controlled robotic structure is included in the core of the computation algorithms. This approach is often referred as computed torque control (CTC) method in robotics [1], [2]. A great portion of robots are underactuated; the development of their MPC is mathematically more challenging than the control of fully actuated robots. By definition, underactuated dynamical systems have less number of independent control inputs than degrees-of- freedom (DoF) [3], [4], [5]. One of the most expressive underactuated examples is the mathematical pendulum model of a gantry crane, where the position of the upper mounting point of the suspension cable can only be directly controlled, while the hanging payload performs a swinging motion [6]. Even so, the positioning of the payload is the control task. The problem of underactuation occurs in several other real-life applications, such as flexible robots, unmanned air and underwater vehicles, robotic hands, legged locomotion systems and tethered robots. Necessarily, underactuated systems have internal dynamics of which the motion is not specified by the control task. The model predictive control of underactuated systems cannot be achieved without the calculation of the internal dynamics. Thus, the dynamics of the full controlled mechanical system has to be considered and the resulting system of differential-algebraic equations (DAE) must be handled in the motion control algorithm [7], [8]. This is in contrast with fully actuated systems, where the control inputs can be expressed on purely algebraic way. Besides underactuation, the complexity of the mechanical structure of robots can grow high. Multibody systems may possess high DoF and closed kinematic loops. A widely used approach for the mathematical description of these systems is to use redundant descriptor coordinates and equation of motion in DAE form [9]. The simulation techniques of multibody systems are elaborated in the literature; however their efficient MPC algorithms can be further developed. Our goal is the further development of the underactuated test robot shown in Fig. 1 [10]. The robot is applicable for the test and quantitative benchmark of the novel underactuated DAE model based motion control algorithms. Our robot possesses similar structure of cranes; employs fan actuators and fast cable winches for actuation; and uses sweeping laser beams and inertial measurement units (IMU) for pose estimation. Fig. 1. Prior prototype of the Acroboter robot platform