Proceedings


     
  Short Abstracts




RoboBrain: Knowledge Engine and Learning from Human Signals 
Ashesh Jain, Cornell University, Ithaca, NY, USA

Abstract: Robots working in human environments receive informative signals in various modalities such as perception, natural language, kinesthetic feedback etc. It is important to develop algorithms which allow robots to learn from these weak and noisy signals. Furthermore, instead of robots learning in isolation, a knowledge-engine that collectively represent concepts learned by different robots is required. In this talk I will present the RoboBrain knowledge engine, the human signals it learns from, and its applications.






Achieving Cyclicity on Repetitive Tasks for Safe Human-Robot Coexistence 
Massimo Cefalo, Università di Roma "La Sapienza", Rome, Italy

Abstract: In this talk we consider the problem of planning safe motions for redundant robots which are subject to repetitive task constraints. Traditionally, task-constrained motion planning is realized through kinematic control techniques. These are on-line motion generation schemes that use a generalized inverse of the task Jacobian (e.g., the pseudoinverse), possibly with the addition of internal motions that do not perturb the execution of the task (null space motions) and are therefore used for local optimization. However, researchers readily identified a critical issue of these schemes when used on repetitive tasks: in general, closed paths in the task space do not result in closed paths in the joint space. This is clearly a drawback in all applications where the robot workspace is shared with humans. If the robot moves differently from cycle to cycle while executing a repetitive task, it is essentially unpredictable and the human-robot coexistence cannot be considered safe. Such behavior is particularly undesirable in human-robot interaction scenarios, because it ultimately hinders the legibility of the robot movements by the human. Researchers identified a quite restrictive structural condition that the generalized inverse must satisfy in order to possess the cyclicity (also called repeatability) property. However, even when a cyclic generalized inverse is used, there is no space left for additional objectives such as obstacle avoidance. An exception are the kinematic control schemes that rely on task augmentation, the archetype of which is the Extended Jacobian method. Nevertheless, even using these method, it is impossible to guarantee that the solution paths are collision-free, essentially due to the fact that obstacle avoidance cannot be effectively encoded in an additional equality task. We show a recent development in constrained motion planning that solves the problem producing cyclic, collision-free paths in the configuration space and guarantees continuous satisfaction of the task constraints. In particular, the solution that we will explore relies on a randomized bidirectional search in the task-constrained configuration space and a loop closure algorithm. Planning experiments on a simple 3R planar robot, the KUKA LWRIV 7-dof manipulator and the KUKA Youbot mobile manipulator show the effectiveness of this method.






Collision Detection, Trajectory Planning, and the Design of the Robot 
Vahid Sotoudehnejad, University of Western Ontario, London, Ontario, Canada

Abstract: The accuracy of external force estimation and collision detection in robotic manipulators depends on the precision of the available sensors, the accuracy of the available dynamic model, and the state of the robot. The formulation of optimal trajectories for a given task with regards to the design of the robot permits further improvement in the precision of estimated external forces. A metric can also be defined to compare different manipulators and different trajectories in terms of safety and force estimation.






Safety Issues in Robotic Assistance for Manual Welding 
Mustafa Suphi Erden, Heriot-Watt University Edinburgh, United Kingdom

Abstract: In this talk I will present my scheme of robotic assistive by impedance compensation with application to manual welding and will highlight the safety issues with such physical human-robot interaction application. Robotic assistance by impedance compensation is based on the hand-impedance measurements I have performed across professional and novice manual welders. These measurements show that professional welders apply more impedance compared to the novice welders, which in turn indicates that novice welders need compensation for their hand impedance. The robotic assistive scheme estimates the direction of intended hand movement in real-time and compensates the hand impedance in the directions perpendicular to the estimated direction. In this way, the hand tremor in the perpendicular directions is suppressed by the robot whereas the movement along the intended direction is let free. Making this assistive scheme operational for actual welding requires various welding specific arrangements of the overall interactive control scheme for sake of both user-friendliness and safety. These arrangements can be planned and programmed only with a good understanding of the welding process as applied by professional welders. This process is not only about melting the metal with the welding torch, but also about transferring the torch from place to place, precisely positioning the tip of the torch in the starting point, turning on the welding arc, turning off the arc in some instances and turning on again, which means moving the torch away, bringing it back and restarting welding, and finally placing the welding torch stably away from the welded material in the end of the process. These actions are to be performed in the vicinity of a melted metal (very high temperature), with a sharp and hot edged welding torch, while the subject is wearing a helmet that obscures the vision of the surrounding. All these imply safety issues to be considered while programming the interactive control, which should both allow an easy manipulation for the subject and ensure safety.






Control Tools for Safe Human-robot Coexistence 
Andrea Maria Zanchettin, Politecnico di Milano, Milano, Italy

Abstract:Trajectory generation and control of an industrial manipulator for safe human-robot collaboration are challenging tasks because of two conflicting requirements: ensuring workers’s safety and quickly completing the task assigned to the robot. In this talk, useful control tools specifically design to ensure safety will be presented. Particular emphasis will be given to their applicability in industrial settings, as well as their relationship with industrial (ISO and ANSI/RIA) standards. Different and realistic applications will show how these tools can be effectively applied to both closed and open robot control architectures.






Online Motion Reshaping based on Dynamical Systems: A Contribution to Human-Robot Co-existence 
Matteo Saveriano, Technische Universität München, Munich, Germany

Abstract: A fruitful human-robot co-working requires close and safe interaction, in which partners can understand and quickly adapt their mutual behaviors. To guarantee a safe interaction, the robot has to modify its motion path on-line, without penalizing task execution. This talk presents our recent work on real-time generation of safe and feasible trajectories in a human-robot interaction scenario. Assuming the robot motion is generated by first-order dynamical systems, we present a sensor based architecture capable to adapt, in real-time, the robot's trajectories to different human behaviors. This reshaping approach does not affect the equilibrium points of the dynamical systems, hence it is guaranteed that the robot will achieve the task which is being executed. The talk describes the theoretical foundations of our approach as well as the implementation in realistic scenarios.






     
  Extended Abstracts




Motion Planning with Safety Constraints and High-Level Task Specifications
Seyedshams Feyzabadi and Stefano Carpin



Motion Safety with People: an Open Problem
Thierry Fraichard



Robot Collision Avoidance Using an Environment Model for Capacitive Sensors
Alwin Hoffmann, Andreas Schierl, Andreas Angerer, Matthias Stüben, Michael Vistein, and Wolfgang Reif



Contact-based Gesture Recognition on an Omni-directional Mobile Robot for a Robot Companion
Kwan Suk Kim and Luis Sentis



Human Factor Multi-Layered Cost Map Approach for Computing Human-Like Paths
Yoichi Morales, Takahiro Miyashita, and Norihiro Hagita



Force Control of the iCub Humanoid for One-foot Balancing and Safe Interaction
Daniele Pucci and Francesco Romano and Silvio Traversaro and Jorhabib Eljaik, and Francesco Nori



Using a Contact-based Inducement for Efficient Navigation in Congested Environments
Moondeep C. Shrestha, Alexander Schmitz, Erika Uno, Yuta Yokoyama, Hayato Yanagawa, Keng Or, and Shigeki Sugano



Real-world Applications of Human-robot Interaction
Christian Vogel, José Saenz, and Norbert Elkmann