My current research focus is in the areas of human-robot interaction (HRI) and motion planning. In particular I am interested in designing algorithms that enable robots to navigate in a socially acceptable fashion among human pedestrians in real world scenarios. My research draws conclusions from psychology and sociology studies and incorporates them into motion planning algorithms in order to achieve human-like robot behaviors.
In the past I have also been involved in robotics research in the areas of Robotic Grasping, Manipulation and Robot Hand Design. A more detailed description of past research projects can be found next.
At NTUA, I was mainly involved in the following research projects:
Grasp Synthesis Algorithms for Multifingered Robot Hands
In the context of my diploma thesis: “Grasp Synthesis Algorithms for Multifingered Robot Hands”, I considered the problem of deriving optimal grasps for two different classes of robot hands: i) multi-DOF, decterous robot hands (such as the DLR-HIT II) and ii) synergistically actuated, underactuated hands. In both cases, the severe hardware limitations of the hand mechanisms were taken into account (i.e., joint limits and the ability of the mechanism for force exertion along the desired directions). This problem was considered for a 15 DOF robot hand and also for the case of a hypothetical synergistic underactuated hand, whose kinematics were derived after determining the principal components of the human grasp (by conducting PCA to a data set collected during human grasping experiments).
Sequential Improvement of Grasp based on Sensitivity Analysis
The first part of this work was published in the proceedings of ICRA 2013 and I was happy to present it in my first conference oral session with a title: “Sequential Improvement of Grasp based on Sensitivity Analysis” [presentation slides] and was also featured to NTUA’s contributions to the project contribution to the project “The Hand Embodied”. In this work, I considered a 15-DOF dexterous robot hand such as the DLR-HIT II in a grasping force optimization problem. Initializing from a random force closure grasp, the algorithm proposed, perturbs the contact points and the wrist position/orientation through sequential optimal grasps wrt force distribution and the Manipulability Volume until a joint limit is reached or a singular hand’s configuration is detected. The whole approach was based in a Mathematical Programming technique commonly referred to as Sensitivity Analysis to perform a Post Optimality Analysis in an initial locally optimal grasp. The main novelty of this algorithm is that it only requires local knowledge of the grasped object’s geometry at the contacts. Hence, it can be generalized for objects of unknown geometry, through the use of tactile/vision/force sensor equipment.
Figure showing small perturbations of the grasp points that lead to significant improvements of Grasp Quality.
Task-Specific Grasp Selection for Underactuated Hands
The second part of my thesis started as an extension of this work, concerning the determination of the initial optimal grasp to be fed in the aforementioned algorithm. However this time we focused on a different class of robot hands that has been under extensive research these days, i.e., the class of Synergistic Underactuated hands. These hands are designed in a bioinspired fashion, importing elements from the design of the most perfect manipulator known in nature, the human hand. This started as an effort to simplify the design of robot hands which is currently severely constrained by the size of current actuators. Besides, it was supported by a series of neuroscientific studies which showed that the majority of everyday life human grasps can only be composed using 1 synergistic degree of freedom. My basic contribution was a grasp optimization scheme for underactuated hands which is able to produce force closure grasps that are optimal with respect to the task to be executed, while taking into consideration hardware and geometric limitations. The whole idea encapsulates the concept of task-specificity, by determining mechanically optimal force closure grasps (force distribution, ability for force exertion) that also favor a desired given task execution. Given the rise of synergistic underactuated hands during the last decade, this paper was formulated in a way compatible with their latest theoretical advances. This work was published in the proceedings of ICRA 2014 and was presented by my coauthor Minas Liarokapis in an oral session with a title “Task-Specific Grasp Selection for Underactuated Hands” [presentation slides].
Figure showing optimal grasp configurations for a hand with the kinematic design of DLR-HIT II but synergistically actuated using a) 1 b) 2 c) 3 and d) 15 (degenerate case) degrees of actuation. The arm configuration was derived anthropomorfically following the work of Minas Liarokapis.
Teleoperation of the AIBO ERS7 Robot Dog via an RGBD vision based system (Microsoft Kinect)
Earlier in my studies, I worked on an HRI project in the NeuroRobotics group of the CSL. My goal was to map common natural human gestures to selected robot movements. Upon installing open source drivers for the communication of the Kinect sensor with the PC, I used open source software such as OpenKinect, OpenNI and RGBDemo, which can track a human’s skeleton and provide real-time data for its movements in space. At the same time, through a simple MATLAB interface, human movement is first mapped to a corresponding AIBO movement in URBI language and consequently transmitted via WLAN to the robot. A video showing the whole system (and me) in action can be found here.