Algorithmen zur Online-Trajektoriengenerierung und Multi-Sensor-Integration in Robotersystemen
Zusammenfassung der Projektergebnisse
The goals of the research project have been threefold: (i) extending the existing on-line trajectory generation framework; (ii) making it applicable for usage in hybrid switched-systems, and (iii) demonstrating highly reactive robot motions in way that humans are not able to touch the robot’s end-effector even if they intend to do so. In order to reach the first two aims, the dynamic models were integrated into the existing online trajectory generation framework. This allows the instantaneous generation of robot motion trajectories while taking into account the dynamic capabilities of robotic system over the entire trajectory. The basic idea of the new method is to compute the local dynamic model in order to provide a locally optimal trajectory within one single control cycle (typically 1 ms). While moving a long this initial trajectory, the trajectory is updated during the next control cycles, and the dynamic model in incorporated to compute maximum acceleration capabilities. Experiments with a seven-degree-of-freedom robot arm show the efficiency and underline the relevance. Equipped with a heavy payload, the system can react instantaneously to unforeseen events and deploy its maximum dynamic capabilities during the reactive motion. The achievement was also the basis for the third part of this project. In order to demonstrate the applicability of the proposed extensions, a scenario was setup, in which a robot had to full a set of tasks, while a human tried to touch the robot. A new collision avoidance method for robot manipulators equipped with an exteroceptive depth sensor was implemented and simulation as well real-world experimental results were achieved. The core of the algorithm is an approach to evaluate the distances between the robot and the dynamic obstacles in its workspace, which is based only on simple and efficient computations on depth space data. These distances are used to generate repulsive vectors, which are processed so as to obtain a repulsive velocity vector in space that is forwarded to an online trajectory generation algorithm. This algorithm instantaneously computes a trajectory starting from the current state of motion, so that a jerk-limited motion is achieved and the dynamic capabilities of the robot are taken into account. Further improvements in terms of natural robot behavior were obtained by using also an estimation of the obstacle velocity. A different repulsive action has been designed for the end-effector and for the other control points on the manipulator in order to be able to avoid collisions while executing at best the original Cartesian motion task. A series of experiments on a seven-degree-of-freedom robot arm using a Microsoft Kinect RGB-D camera confirmed the real-time effectiveness and good performance of the method. The entire software that contains the online trajectory generation algorithms has been freely released as part of the Reflexxes Motion Libraries (http://www.reflexxes.com). The following list highlights a number of press and media articles about the work done during the period of this research fellowship. Dec. 13, 2011, The History Channel: The Universe (special issue on artificial intelligence) Oct. 13, 2011, IEEE Spectrum: Robot Masters Jenga, Next the World Oct. 10, 2011, IEEE Spectrum: How JediBot Got Its Sword Fighting Skills Oct. 7, 2011, VDI Nachrichten: Sehnsüchtiges Warten auf serienreife Service-Roboter Sep. 14, 2011, Süddeutsche Zeitung: Duell mit einem Roboter Sep. 9, 2011, The Electric Playground: JediBot Jul. 21, 2011, SPIEGEL Online: Roboter schwingen Schwerter und Bratenwender Jul. 20, 2011, WIRED: Stanford’s Lightsaber-Wielding Robot Is Strong With the Force Jul. 19, 2011, Handelsblatt: Jedibot kämpft mit dem Lichtschwert Jul. 18, 2011, IEEE Spectrum: Stanford’s ’JediBot’ Tries to Kill You With a Foam Sword
Projektbezogene Publikationen (Auswahl)
- A Depth Space Approach to Human- Robot Collision Avoidance. In Proc. of the IEEE International Conference on Robotics and Automation, pp. 338-345, Saint Paul, MN, USA, May 2012
F. Flacco, T. Kröger, A. De Luca, and O. Khatib
- JediBot — Experiments in Human-Robot Sword-Fighting. In Proc. of the International Symposium on Experimental Robotics, Québec City, Canada , June 2012
T. Kröger, K. Oslund, T. Jenkins, D. Torczynski, N. Hippenmeyer, R. B. Rusu, and O. Khatib
- On-Line Trajectory Generation: Nonconstant Motion Constraints. In Proc. of the IEEE International Conference on Robotics and Automation, pp. 2048-2054, Saint Paul, MN, USA, May 2012
T. Kröger
- Simple and Robust Visual Servo Control of Robot Arms Using an On-Line Trajectory Generator. In Proc. of the IEEE International Conference on Robotics and Automation, pp. 4862-4869, Saint Paul, MN, USA, May 2012
T. Kröger and J. Padial
