Catálogo de publicaciones - libros

tasks are tasks in which a robot moves a manipulated object while controlling the contact with object(s) in its environment. Still today, in industry, compliant motion tasks are often position-controlled. Hence, they require very structured environments, i.e., the work pieces or parts to assemble are accurately positioned and their dimensions are known. In these cases, the robot receives and executes a nominal task plan.

This chapter contains a survey of the research in autonomous compliant motion [167]. It presents the state-of-the art in the development of the different components of the autonomous compliant motion system described in the previous chapter, focusing on approaches using pose (i.e., positions and orientations), twist (i.e., translational and rotational velocities) and/or resultant contact wrench measurements (i.e., contact forces and moments). A concise description of the force control component is presented in Sect. 2.2. Section 2.3 contains an overview of the research on the estimation component of the system. Previous research on fine-motion planning is summarised in Sect. 2.4. Finally Sect. 2.5 concludes.

This chapter presents some basic aspects of Bayesian probability theory [21, 153]. First of all, the difference between Bayesian and classical statistics is discussed (Sect. 3.2). Section 3.3 presents Bayesian based on data measured at discrete time steps. Section 3.4 describes Bayesian . Sections 3.5 and 3.6 focus on , i.e., the optimisation of the experiment in order to provide “optimal” state estimates. Section 3.5 presents ways to measure the “information content” of data and estimates. The algorithms for optimisation under state uncertainty are surveyed in Sect. 3.6. Section 3.7 concludes.

The Kalman Filter (KF) [137, 240] is a special case of Bayesian filtering theory. It applies to the estimation of a continuous-valued state if the state space description of the estimation problem has subject to .

Exact finite-dimensional Bayesian filters exist only for a small class of systems. The previous chapter discussed the best known example, i.e., the Kalman Filter (KF) for linear systems subject to additive Gaussian uncertainties. Other examples are the filters of Beneš [25], which requires the measurement model to be linear, and Daum [61], applicable to a more general class of systems with nonlinear process and measurement models for which the posterior pdf is any exponential distribution.

This chapter describes the contact modelling for autonomous compliant motion assuming quasi-static, rigid, frictionless contacts. The contact models are needed in the force controller, the estimator and the planner of the system. The models are di.erent for each contact formation (CF), and are a function of the geometrical parameters (i.e., the positions, orientations and dimensions of the contacting objects).

This chapter presents the estimator component of the autonomous compliant motion system (see Fig. 1.2). This component The estimation is based on measurements of the manipulated object and and of the contact .

This chapter validates the theory of the previous chapters with an experiment. The experiment consists of the estimation of the geometrical parameters and the recognition of CFs during a cube-in-corner assembly, Fig. 1.1. This chapter uses the Iterated Extended Kalman Filter (IEKF) described in Chap. 4, the Non-minimal State Kalman Filter (NMSKF) described in Chap. 5 and the contact models of Chap. 6. The details about the application of these filters and models for the estimation of the geometrical parameters and the CF recognition are presented in Chap. 7.

This chapter describes the calculation of a compliant motion task plan which improves the observation of the inaccurately known geometrical parameters (i.e., the positions, orientations and dimensions of the contacting objects). This is called . The experiment of the previous chapter shows