In-hand dexterous manipulation of an object is the ability to change the configuration (position and/or orientation) of an object held in the hand. This is an ability that has allowed humans to use tools and interact with the environment effectively.
For the past decades, robotics researchers have worked to provide dexterous manipulation skills to the robots by designing robotic hands that mimic the human hand and by developing applications that allow performing autonomous manipulation or teleoperation in harsh environments. Despite the progress made, managing the uncertainties that exist in the real world is one of the problems that still need to be worked on. Many existing manipulation methods for controlling robotic hands require a priori information about the object and high-fidelity sensors that are typically limited only to laboratory settings.
The main objective of this thesis is to develop strategies for the dexterous manipulation of unknown objects, using the tactile information generated during the grasp of the object and the manipulation process itself. In manipulation applications based on tactile information, the robotic hand has access only to tactile and proprioceptive data, in addition, no a priori information is known about the manipulated object. This reflects real-world applications, where there is uncertainty in the models of the objects that are commonly manipulated in daily activities, as well as in the sensorial measurements.
In this dissertation, novel manipulation strategies based on heuristic and gradient optimization methods are proposed.
Three quality indexes are selected to measure the goodness of the grasp during the manipulation, related to the configuration of the hand, the quality of the grasp, and the configuration of the object. Starting from a given initial grasp, the manipulation strategies are able to improve one quality index or a combination of them. The manipulation strategies are validated with real experimentation using robotic hands equipped with tactile sensors, allowing the execution of practical applications, such as object recognition, force optimization, and telemanipulation.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados