Zoe Doulgeri

Bio: Zoe Doulgeri is an academic researcher from Aristotle University of Thessaloniki. The author has contributed to research in topics: Control theory & Kinematics. The author has an hindex of 21, co-authored 156 publications receiving 1941 citations. Previous affiliations of Zoe Doulgeri include Chalmers University of Technology & Imperial College London.

Dynamics and control of a set of dual fingers with soft tips

Abstract: This paper attempts firstly to derive a mathematical model of the dynamics of a set of dual fingers with soft and deformable tips which grasps and manipulates a rigid object with some dexterity. To gain a physical insight into the problem, consideration is restricted to the case that the motion of the whole system is confined to a horizontal plane. Secondly on the basis of the derived model it is shown that the rotation of the object can be indirectly controlled by the change of positions of the center points of both contact areas on the object. Then, each of the center points of contact areas can be positioned by inclining the last link of each corresponding finger against the object. It is further shown that, when both forces of pressing the object becomes almost equal, the equation of motion of the object in terms of rotational angles assumes the form of a harmonic oscillator with a forcing term, which can be regulated coordinately by the relative angle between the two last links contacting with the object. It is also shown that dynamics of this system satisfy passivity. Finally, design problems of control for dynamic stable grasping and enhancing dexterity in manipulating things are discussed on the basis of passivity analysis.

Neuro-Adaptive Force/Position Control With Prescribed Performance and Guaranteed Contact Maintenance

Abstract: In this paper, we address unresolved issues in robot force/position tracking including the concurrent satisfaction of contact maintenance, lack of overshoot, desired speed of response, as well as accuracy level. The control objective is satisfied under uncertainties in the force deformation model and disturbances acting at the joints. The unknown nonlinearities that arise owing to the uncertainties in the force deformation model are approximated by a neural network linear in the weights and it is proven that the neural network approximation holds for all time irrespective of the magnitude of the modeling error, the disturbances, and the controller gains. Thus, the controller gains are easily selected, and potentially large neural network approximation errors as well as disturbances can be tolerated. Simulation results on a 6-DOF robot confirm the theoretical findings.

Prescribed Performance Tracking of a Variable Stiffness Actuated Robot

Abstract: This paper is concerned with the design of a state feedback control scheme for variable stiffness actuated (VSA) robots, which guarantees prescribed performance of the tracking errors despite the low range of mechanical stiffness. The controller does not assume knowledge of the actual system dynamics nor does it utilize approximating structures (e.g., neural networks and fuzzy systems) to acquire such knowledge, leading to a low complexity design. Simulation studies, incorporating a model validated on data from an actual variable stiffness actuator (VSA) at a multi-degrees-of-freedom robot, are performed. Comparison with a gain scheduling solution reveals the superiority of the proposed scheme with respect to performance and robustness.

Robot vision

Abstract: A scheme is developed for classifying the types of motion perceived by a humanlike robot. It is assumed that the robot receives visual images of the scene using a perspective system model. Equations, theorems, concepts, clues, etc., relating the objects, their positions, and their motion to their images on the focal plane are presented. >

Learning dexterous in-hand manipulation:

Abstract: We use reinforcement learning (RL) to learn dexterous in-hand manipulation policies that can perform vision-based object reorientation on a physical Shadow Dexterous Hand. The training is performed...

Robotic grasping and contact: a review

Abstract: In this paper, we survey the field of robotic grasping and the work that has been done in this area over the last two decades, with a slight bias toward the development of the theoretical framework and analytical results in this area.

Hands for dexterous manipulation and robust grasping: a difficult road toward simplicity

Abstract: In this paper, an attempt at summarizing the evolution and the state of the art in the field of robot hands is made. In such exposition, a critical evaluation of what in the author's view are the leading ideas and emerging trends is privileged with respect to exhaustiveness of citations. The survey is focused mainly on three types of functional requirements a machine hand can be assigned in an artificial system, namely, manipulative dexterity, grasp robustness, and human operability. A basic distinction is made between hands designed for mimicking the human anatomy and physiology,and hands designed to meet restricted, practical requirements. In the latter domain, arguments are presented in favor of a -minimalistic" attitude in the design of hands for practical applications, i.e., use the least number of actuators, the simplest set of sensors, etc., for a given task. To achieve this rather obvious engineering goal is a challenge to our community. The paper illustrates some of the new sometimes difficult, problems that are brought about by building and controlling simpler, more practical devices.

Solving Rubik's Cube with a Robot Hand.

Abstract: We demonstrate that models trained only in simulation can be used to solve a manipulation problem of unprecedented complexity on a real robot. This is made possible by two key components: a novel algorithm, which we call automatic domain randomization (ADR) and a robot platform built for machine learning. ADR automatically generates a distribution over randomized environments of ever-increasing difficulty. Control policies and vision state estimators trained with ADR exhibit vastly improved sim2real transfer. For control policies, memory-augmented models trained on an ADR-generated distribution of environments show clear signs of emergent meta-learning at test time. The combination of ADR with our custom robot platform allows us to solve a Rubik's cube with a humanoid robot hand, which involves both control and state estimation problems. Videos summarizing our results are available: this https URL