Distributed estimation of the inertial parameters of an unknown load via multi-robot manipulation

International audienceIn this paper, we propose a distributed strategy for the estimation of the kinematic and inertial parameters of an unknown body manipulated by a team of mobile robots. We assume that each robot can measure its own velocity, as well as the contact forces exerted during the body manipulation, but neither the accelerations nor the positions of the contact points are directly accessible. Through kinematics and dynamics arguments, the relative positions of the contact points are estimated in a distributed fashion, and an observability condition is defined. Then, the inertial parameters (i.e., mass, relative position of the center of mass and moment of inertia) are estimated using distributed estimation filters and a nonlinear observer in cooperation with suitable control actions that ensure the observability of the parameters. Finally, we provide numerical simulations that corroborate our theoretical analysis. I. INTRODUCTION Last decades have seen raising attention to cooperative manipulation by teams of robotic agents, due to its relevant applications in several fields, such as search and rescue and disaster recovering, cooperative transportation, and service robotics. In [1], for example, a path planner for cooperative manipulation is proposed. It consists of two components; a global path planner, responsible of obstacle avoidance, and a local manipulation planner, responsible of the object's manipulation via the control of its position. In [2], the control of robotic teams with the aim of combining optimal goal regulation and relaxation of the formation rigidity constraint is carried out. Cooperative manipulation has also been dealt with in aerial applications [3], [4], using cables, or other interaction tools. However, the majority of the works does not focus on an accurate dynamic modeling of the payload, which is either modeled as a point mass or as a rigid body with unknown dynamical parameters. In [5], the problem of cooperatively manipulating an object on a plane by a team of non-holonomic wheeled robots is dealt with. The problem is solved by defining the dynamics of the whole system (i.e., the robot team and the object) and through a decomposition technique into a task and a null space. The dynamic model of the object is accounted for, but its inertial parameters are assumed to be known, as well as the positions of the contact points between the robots and the object. An accurate and on-line estimation of the inertial parameters of the payload is very useful for at least tw