University of Patras
Ανάπτυξη συστημάτων ελέγχου για συνεργασία ρομπότ-ανθρώπου σε χειρισμούς αντικειμένων
Abstract
dc:descriptionHuman-robot cooperation is a research topic that aims at the complementary combination of human skills and the properties of a robot. The most important characteristics of a robot arm is the ability to move with high precision, the repeatability, and the load capacity, allowing usage in industrial applications. The main skills of a human that are either absent from a robot or outperform its abilities, are the awareness of the environment from multiple sensors, the decision making and inference mechanisms, the fast adaptation to changes of the environment and the optimal configuration of the myoskeletal system for a task. Given the difficulty to develop a robot that approaches the skills of a human, an appropriate human-robot cooperation can bridge the gap between manual labor and hard automation, and also create new fields of applications.Initially, the interaction control between a human and a robot is studied and novel methods to regulate online the gains of the compliant control system are proposed for the improvement of objective criteria, such as the effort of the operator. For the gain regulation, methods based on computational intelligence with supervised and unsupervised learning are used. The minimum jerk model is proposed for the first time as a reference for variable compliance, under the assumption that if the robot's motion is adjusted during the cooperation according to that model, then the cooperation is more efficient for the operator. This model is extended also in rotational motion, apart from the translational. Then, a method is proposed that utilizes EMG signals for continuous regulation of the gains in order to adjust the controller to the activation levels of the operator's arm muscles. In that way, the operators are able to use their muscles as switches of the robot's compliance.An important problem in Cartesian robot control appears when it approaches singular configurations. A novel method for the restriction of the robot's motion based on performance indices is proposed, that applies repulsive forces to the operator so that there is no case of driving in such configurations. With this method of performance constraints, repulsive forces are calculated in real time based on the gradient of a local performance index of the robot arm. This gradient is approximated numerically, without requiring an analytic expression of the index. Various local performance indices are compared with respect to the operator's comfort during cooperation. The performance constraints allow even an untrained operator to collaborate with the robot and acquire direct feedback for low performance configurations that must be avoided.When the human collaborates with the robot, the stability of the robot's control system is affected. In order to avoid conservative predefined gains to the controller, a novel instability index and a gain adaptation technique are proposed. The instability index detects high frequency oscillations in the measured external force that are attributed to instability and the adaptation method adjusts accordingly the gains of the compliance controller to stabilize the robot. It is experimentally shown that the increase of the virtual damping of the compliance controller for stability, which is widely used in the literature, is inefficient for the stability of the robot and for the cooperation with a human. Instead, the increase of the virtual inertia is proposed, that is proved more appropriate than the virtual damping. Towards safe human-robot interaction, dangerous situations are investigated that can occur in the shared workspace. A robot collision detection method is proposed that is based on the estimation of the robot's dynamics with fuzzy logic and without the use of an explicit dynamic model. This method is trained to approximate the dynamic characteristics of each joint and can take into account the dynamic coupling between them. Also, a contact type distinction method is proposed in order to distinguish between voluntary contact of the operator with the robot and involuntary collisions, by analyzing the frequency characteristics of the external force in each case. Finally, robot reaction techniques are investigated in order to minimize the collision forces. The collision detection and distinction methods are compared experimentally with others from the literature and they are proved to be faster, which is very important for early robot reaction and injury avoidance. For English readers who are unable to understant Greek language, the content of this thesis is also available in the English papers listed on page 9.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Dimeas, Fotios
- Δημέας, Φώτιος
Subjects
dc:subject × 10Rights
dc:rights- Statement dc:rights
-
- BY
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/40358