Show filters Hide filters

Refine your search

Publication Year
Publisher
1-13 out of 13 results
Change view
  • Sort by:
05:32 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2014

Human-Aware Motion Planning for Mobile Robots in Social Encounters

This video presents the implicitly coordinated motion model (ICMM), a novel motion model for the prediction, planning and coordination of agent trajectories in multi-agent encounters. It explicitly incorporates the social cooperation between humans and mobile robots. Parameters of the ICMM are identified from recorded actual encounters among groups of humans using methods from inverse optimal control. The agents' trajectories are optimized using the Timed-Elastic-Band framework [1,2] considering multiple conflicting objectives such as fastest path, minimal spatial separtion among agents, (kino-)dynamic constraints but also global proxemic aspects such as coherent motion of social groups and a prefered side of passing each other. The recorded dataset contains 73 recorded encounters with up to five humans and a total of 283 individual trajectories. Technical note: the program running in this video has been compiled in debug mode. Compilation with release settings results in a speedup factor of 7. Time-line: 00:09 Parallel trajectory optimization in alternative homotopy classes 00:50 Parallel trajectory optimization with dynamic homotopy class exploration 01:37 TEB selection and implictly coordinated motion model (ICMM) 02:24 Simulations of social encounters 04:11 Using the ICMM on a mobile robot
  • Published: 2014
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
04:49 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2011

Situated learning of visual robotic behaviors

Proof of concept experiments to learn situated behaviors using Artificial neural networks. Timeline: 0:06 Demonstration phase: Semi-automatic demonstrations 0:37 System Architecture 0:49 Features extracted 0:51 Experiment 1 1:38 Experiment 2 3:29 Experiment 3 4:47 Credits and contacts This video is the supplement to the paper: "Situated Learning of Visual Robot Behaviors", 4th International Conference on Intelligent Robotics and Applications (ICIRA 2011), Aachen, Germany.
  • Published: 2011
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
03:25 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2012

Learning Mobile Robot Behaviour Dynamics

This video shows the mobile robot behavior dynamics learned from demonstration examples. The mobile robot is equipped with an omnidirectional camera with a 360 degree horizontal and 75 deg vertical field of view directed towards the bottom to capture the floor. An ensemble of experts segmentation scheme distinguishes the omnidirectional image into floor and non-floor regions. Three indoor robotic behaviors viz. corridor following, obstacle avoidance and homing are tele-operated to the robot during the demonstration phase by a teacher during which the omnidirectional image and the corresponding executed actions are recorded. The individual behaviors are then represented by a dynamic system that couples the perception and the performed action. Thus every behavior possesses a behavioral dynamics and the variables that characterize this dynamics are called behavioral variables. Here the behavioral dynamics are represented using Gaussian Mixture Models parameters of whose are identified from demonstrations. The recorded behavioral variables are Corridor Following : Rotational velocity, lateral offset of the robot to the corridor (alpha) and orientation error of the robot to the center of the corridor (beta) Obstacle avoidance : Rotational velocity, sinusoid of the next traversible safe direction sin(theta ) and the cosinus of the orientation of the nearest obstacle times the inverse of its distance cos(theta ).1/d Homing : Rotational velocity, distance to the goal point, orientation to the goal point The Homing or docking zone are marked by two red circles whose midpoint of the virtual line connecting the centroids is the docking/homing point. The three behaviors are coordinated manually either by behavior arbitration (subsumption architecture) or by command fusion (weighted summation). The video shows the learned behaviors individually performing the task and finally the fused behavior architecture navigating through the indoor environment and docking to the final goal point. 0:00 Corridor following 0:27 Obstacle avoidance 1:02 Homing 1:39 Behavior coordination via command fusion 3:20 Credits
  • Published: 2012
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
01:58 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2011

Traversing seen and unseen corridors with Artificial Neural Networks/Context matching

Proof of concept experiments to navigate a corridor environment using Two level feature matching architecture. Layer 1 classifies the scenario from the shape and appearance of the environment into Corridor (C) or Open room (O) or Cluttered (L) environment. Layer 2 deploys scenario specific model to predict action and correspondingly navigate the environment. The two models used are: 1. Artificial Neural Networks and 2. Context matching and prediction. Timeline: 0:00 Scenario: Known corridor with Artificial Neural Networks 0:17 Scenario: Known corridor with Context matching and prediction 0:39 Scenario: Unknown corridor with Artificial Neural Networks 1:02 Scenario: Unknown corridor with Context matching and prediction 1:18 Scenario: Transition between trained and an untrained corridor using Artificial Neural Networks This video is the supplement to the paper: "Scenario and context specific visual robot behavior learning" presented at the 2011 IEEE International Conference on Robotics and Automation (ICRA2011), May 9-13 2011, Shanghai International Convention Center, Shanghai, China. For more information please visit: http://www.rst.e-technik.tu-dortmund....
  • Published: 2011
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
01:35 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2011

TUDOR - A multi-link-flexible robot arm catching thrown softballs

The video shows the three degree of freedom multi-flexible link robot arm TUDOR (Technische Universität Dortmund omnielastic robot) catch balls thrown by a human. The intrinsic compliance of flexible link robots offers novel opportunities for safe human robot interaction as well as force control. Nevertheless the fast joint motions induce significant structural vibrations and the kinematic chain shows load and configuration dependent static deflections. These properties aggravate the precise positioning of flexible link robots. In the video a stabilizing decentralized strain feedback controller rapidly damps the structural vibrations while an artificial neural network has been trained to solve the inverse kinematics problem for varying configurations and payload. The trajectory of the segmented ball is estimated from measurements obtained via a Microsoft Kinect RGB-D-sensor. The measurements are fused with a motion model including Newton friction via an extended Kalman filter. The camera is mounted next to the laboratory entrance. The demonstrator has a success rate of more than 66%, provided that the ball trajectory intersects the planar subspace of the robot workspace, in which the ball can be caught. Timeline: 00:08 Ball catching without vibration control 00:32 Ball catching with decentralized vibration damping control 01:00 Ball tracking and catch location prediction with the Kinect References: Malzahn, J., A. S. Phung und T. Bertram: A Multi-Link-Flexible Robot Arm Catching Thrown Balls. 7th German Conference on Robotics, 21./22.05.2012 ,pp. 411-416 Mai 2012 For more information on the project please visit: http://www.rst.e-technik.tu-dortmund....
  • Published: 2011
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
00:56 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2011

TUDOR vibration damping of a multi-link-flexible robot

The underlying concept is based on an independent joint control strategy. The joint angles of each actuator are controlled by a cascaded position controller with an inner velocity and a motor-current loop. Strain measurements on each link are fed back to the input of the velocity control loop at the preceding joint. Additionally a zero-vibration input shaper filters the command signal. Timeline: 00:07 Step motion from [0°, 0°, 0°] to [0°, 45°, -45°] 00:23 Step motion from [0°, 45°, -45°] to [0°, 135°, 45°] 00:39 Hit from an additional test mass falling onto the robot at [0°, 135°, 45°] References: - Malzahn, J., A. S. Phung, F. Hoffmann und T. Bertram: Vibration Control of a Multi-Flexible-Link Robot Arm under Gravity, In IEEE International Conference on Robotics and Biomimetics, Phuket (Thailand),07.-11.12.2011, pp. 1249-1254 Dezember 2011 - Malzahn, J., M. Ruderman, A. S. Phung, F. Hoffmann und T. Bertram: Input Shaping and strain gauge feedback vibration control of an elastic robotic arm, 2010 IEEE Conference on Control and Fault Tolerant Systems (Systol'10), pp. 672-677, Nice, France Oktober 2010 For more information on the project please visit: http://www.rst.e-technik.tu-dortmund....
  • Published: 2011
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
01:48 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2011

Traversing unknown foyer and cluttered environments with Artificial Neural Networks/Context matching

Proof of concept experiments to navigate an unknown foyer using Two level feature matching architecture. Layer 1 classifies the scenario from the shape and appearance of the environment into Corridor (C) or Open room (O) or Cluttered (L) environment. Layer 2 deploys an Artificial Neural Network specific to the classified scenario. Timeline: 0:00 Scenario: Unknown foyer with Artificial Neural Networks 0:39 Scenario: Cluttered environment with Artificial Neural Networks 1:10 Scenario: Cluttered environment with Context matching and prediction This video is the supplement to the paper: "Scenario and context specific visual robot behavior learning" presented at the 2011 IEEE International Conference on Robotics and Automation (ICRA2011), May 9-13 2011, Shanghai International Convention Center, Shanghai, China.
  • Published: 2011
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
02:08 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2011

TUDOR initial visual servoing experiments

The video demonstrates 2D visual servoing for the multi-flexible-link robot arm TUDOR. The visual servoing scheme utilizes the robot jacobian of the equivalent rigid arm. Vibration damping is achieved by an underlying cascaded independent joint controller with augmented strain feedback (http://www.youtube.com/watch?v=pmnX4w...). The objective of the visual servoing controller is to keep the centroid of a square pattern in the image center. The controller also compensates for the pose error caused by deflections due to an additional payload of 600 gram. The camera is a Microsoft Kinect, although the depth information has not yet been incorporated in the controller. Exploiting the potentials of the RGB-D sensor for the control flexible-link robots is subject to current research. Timeline: 00:06 Experiment I: Centering a moving square pattern 00:40 Experiment II: Change in payload without visual servoing 01:27 Experiment III: Change in payload with visual servoing
  • Published: 2011
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
01:15 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2011

TUDOR tip position control with Neural Networks

The forward and inverse kinematic model of a multi-flexible-link robot arm for varying payloads are each approximated by using artificial neural networks. The tip position is predicted from the joint angles and strain signals. The strain measurements allow the reaction to changes in the payload. Thus, the kinematic models can be applied in case of varying payloads. The closed loop controller corrects the joint angles at the target pose based on the pose predicted by the forward model and archives an average pose error of less than 3 mm. Timeline: 00:10 Deflection of TUDOR after adding 600g payload 00:25 Tip position control of TUDOR after adding 600g payload 00:42 Relaxation of TUDOR after removing 600g payload 00:57 Tip position control of TUDOR after removing 600g payload References: - Phung, A. S., J. Malzahn, F. Hoffmann und T. Bertram: Data Based Kinematic Model of a Multi-Flexible-Link Robot Arm for Varying Payloads, In IEEE International Conference on Robotics and Biomimetics, Phuket (Thailand),07.-11.12.2011, pp. 1255-1260 Dezember 2011 - Malzahn, J., A. S. Phung, F. Hoffmann und T. Bertram: Vibration Control of a Multi-Flexible-Link Robot Arm under Gravity, In IEEE International Conference on Robotics and Biomimetics, Phuket (Thailand),07.-11.12.2011, pp. 1249-1254 Dezember 2011 For more information on the project please visit: http://www.rst.e-technik.tu-dortmund.de/cms/de/Forschung/Schwerpunkte/Robotik/TUDOR neu/index.html
  • Published: 2011
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
03:00 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2013

Experiments on force control of a multi-flexible-link robot

Structural elasticity represents an undesired effect in a variety of technical systems such as fire rescue turntable ladders, concrete pumps, cherry pickers, cranes and robots. Oscillations prolong settling times and static deflections reduce accuracy. Avoiding structural elasticity therfore most often is a design criterion. However, in this video we intend to show the other side of the coin by exploiting the potential of the elastic properties to sense contact forces. Elasticity is intentionally introduced in an experimental structure and accounted for in the control of the mechanism. The control concept sufficiently mitigates the oscillations as shown in the beginning of the video and position accuracy in the presence of varying payloads can be improved e.g. by means of visual servoing as exemplified in another video (https://www.youtube.com/watch?v=V2NnEU6yGEA). The control concept behind the video follows an independent joint control strategy. The joint angles of each actuator are controlled by a cascaded position controller with an inner velocity and a motor-current loop. The torques acting on the individual joints due to oscillations, gravitational influences and physical interactions of the robot with it's environment are inferred via strain measurements on each link. This information is fed back to each independent joint motion controller to actively influence the reflected joint compliance while simultaneously damping oscillations. Oscillations may occur because of high joint accelerations as well as unforeseen but also planned interactions with the environment. These oscillations are damped regardless of their source. The control concept allows to shape the reflected compliance such that the probability of breaking even fragile objects in case of accidental collisions is significantly reduced. Time-line: 00:12 Oscillation damping during step motion from [0°, 0°, 0°] to [0°, 45°, -45°] 00:27 Damping oscillation due to external impacts 00:37 Passive compliance test at the tip using a soft-ball. With just passive compliance it is clearly visible that the soft-ball gets compressed. 00:58 Active compliance test at the tip using a soft-ball. The compression of the ball is hardly visible. 01:16 Active compliance tests at different points along the structure using a soft-ball. Conventional robots can be equipped with force/torque sensors at the tip. Force control laws enable a user to grab the robot at this sensor and guide it to another desired position. In contrast, the example shows that the flexible links allow the robot to be grabbed along the structure to perform this guidance. 01:34 Pushing the robot at the tip using a feather. 01:45 Accidental collision with a feather in the path and no force control. The robot tries to reach the commanded joint configuration at all cost and breaks the feather. 02:00 Accidental collision with a feather in the path and *activated* force control. The controller limits the force exerted on the feather and stops the robot. Once the feather is removed from the path, the robot approaches the desired joint configuration. 02:12 Accidental collision with a Christmas ball in the path and no force control. Similar to the feather experiment without force control the Christmas ball breaks if the end-effector destination corresponding to the desired joint values lies within the ball. 02:37 Accidental collision with a Christmas ball in the path and *activated* force control. Again, the force controller reduces the exerted forces and saves the Christmas ball.
  • Published: 2013
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
02:41 TU Dortmund, Lehrstuhl für Regelungssystemtechnik No linguistic content; Not applicable 2014

Exploiting Link Elasticity in a Conventional Industrial Robot Arm

Conventional industrial robots are intended for fast and precise manipulation of heavy payloads. The optimization of these objectives results in the bulky design of today's conventional industrial robots, aiming at the maximization of precision through structural rigidity. The video demonstrates that the question, whether a robot arm is rigid or not, basically depends on how close you wish to look at it. A human can deflect the endeffector of a typical conventional industrial robot by hand without major efforts. The video illustrates the structural oscillations and deflections in the order of 2 mm resulting from moderate manual pushes. The deflections and oscillations originate from a combination of the actually present joint as well as link elasticity. While the joint elasticity due to the harmonic drive gears as well as the drive belts are surely dominant, the link elasticity is also measurable. The presented work employs optical strain sensors -- so called Fiber-Bragg-Grating sensors -- for this purpose. The optical fibers are glued onto the links and their working principle is briefly sketched in the second part of the video. In the third part of the video the link elasticity is exploited to make the conventional industrial robot backdriveable. The demonstrated experiment is a physical interaction with the robot. The human touches the arm at arbitrary points along the structure in order to reconfigure the arm posture as desired. The techniques used in the video have been developed and investigated in previous works on the multi-elastic-link arm TUDOR (watch our previous video with TUDOR: http://youtu.be/kJPuenyxeps). The experiments shown in this video represent a straight forward transfer of these techniques to a conventional industrial robot. The strain dynamics modeling of elastic link robots is presented in: Malzahn, J., R. F. Reinhart and T. Bertram: Dynamics Identification of a Damped Multi Elastic Link Robot Arm under Gravity, IEEE International Conference on Robotics and Automation, Honkong, China, 2014 The usage of the strain dynamics model for interaction control is explained in: Malzahn, J. and T. Bertram: Collision Detection and Reaction for a Multi-Elastic-Link Robot Arm, IFAC World Congress, Cape Town (South Africa), 2014 Video outline: 00:10 Demonstration of elasticity in a conventional robot arm 00:33 Link deflection measurement principle 01:30 Experiment: physical interaction with a conventional robot arm Note: The experiments shown in the video have been conducted by professionals. For your own safety: NEVER STAY INSIDE THE WORKSPACE OF AN INDUSTRIAL ROBOT IN OPERATION! For more information on the project please visit: http://tinyurl.com/TUDORRobot
  • Published: 2014
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: No linguistic content; Not applicable
01:48 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Silent film 2011

Visual robotic behavior learning

Proof of concept experiments to learn visual robotic behaviors using Instance based learning. Timeline: 0:00 Obstacle avoidance using Instance based learning 0:19 Corner situation using Instance based learning 0:37 Wandering in a corridor using Instance based learning. This video is the supplement to the paper: "Imitation Learning for Visual Robotic Learning" presented at the 19. Workshop Computational Intelligence, Dortmund, 2-4 December 2009.
  • Published: 2011
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Silent film
04:51 TU Dortmund, Lehrstuhl für Regelungssystemtechnik Original sound, no spoken text 2013

Oscillation Damping, Collision Detection and Reaction for a Multi-Elastic-Link Robot Arm

In first place, elasticity in the links of robot arms and structurally comparable mechatronic systems such as construction machines, fire rescue turntable ladders, cherry pickers or automobile concrete pumps is a highly undesired effect. It prolongs settling times and deteriorates the positioning accuracy. Therefore substantial mechanical design efforts are commonly taken to avoid link elasticity in these mechanisms. The presented work approaches from the contrary perspective and intentionally introduces intrinsic structural compliance in the links of an experimental robot platform. The motivation is to exploit the added intrinsic link compliance to reduce the overall robot weight, to cut costs, to add positioning tolerance as well as to add contact force sensing capabilities to the system. The video shows, how robust and rapid settling as well as disturbance rejection can still be accomplished by devising control algorithms [1,2] based on per link strain measurements. In addition, the derivation and identification of mathematical models that accurately describe the load and joint configuration dependent static end effector deflections allows -- through software -- for the compensation of the inaccuracy of the mechanism [3]. The feasibility of time critical and precise end effector positioning for an elastic link arm has been exemplified with ball catching experiments before [4] (http://www.youtube.com/watch?v=P4 i k...). With the mechanical imperfections compensated by the developed inner loop control software, the video demonstrates, how the intrinsic link compliance can be exploited to actively shape the apparent arm compliance, to sensitively sense contact forces, to safely react to accidental collisions as well as to enable intentional physical human machine interaction. The control scheme behind these features uses an identified model of the residual damped arm dynamics [5]. This model is way simpler to derive and identify than a holistic arm model including the oscillatory and actually infinite dimensional arm dynamics. An identified linear mapping from the strain readings acquired close to the hubs on each passively compliant link and the motor torques turns each link into load side joint torque sensors. This way the video shows that collision detection and reaction techniques originally developed by other authors for rigid or elastic joint robots can be readily adopted for the use with elastic link robots [6]. The provided results imply that link elasticity is not necessarily just a problem. In contrast, the devised control concepts are able to compensate for the machine imperfections reveal promising new perspectives. Time-line: 00:12 Introduction to the experimental setup 00:36 Exp I: Oscillation Damping: step motion 00:48 Exp II: Oscillation Damping: harmonic disturbance 02:10 Exp III: Collision detection and reaction: blunt impacts with a balloon 02:37 Exp IV: Collision detection and reaction: sharp impacts with a balloon 03:02 Exp V: Collision detection and reaction: sharp impacts with a Christmas ball 03:23 Exp VI: Collision detection and reaction: sharp impacts with a human arm 03:47 Exp VII: Interaction in zero gravity mode For more information on the project please visit:http://www.rst.e-technik.tu-dortmund....
  • Published: 2013
  • Publisher: TU Dortmund, Lehrstuhl für Regelungssystemtechnik
  • Language: Original sound, no spoken text
out of 1 pages
Loading...
Feedback

Timings

   65 ms - page object
   24 ms - search
    0 ms - highlighting
    0 ms - highlighting/15413
    0 ms - highlighting/15422
    0 ms - highlighting/15415
    0 ms - highlighting/15423
    0 ms - highlighting/15416
    0 ms - highlighting/15414
    0 ms - highlighting/15421
    0 ms - highlighting/15419
    0 ms - highlighting/15424
    0 ms - highlighting/15420
    0 ms - highlighting/15418
    0 ms - highlighting/15417
    0 ms - highlighting/15412

Version

AV-Portal 3.7.0 (943df4b4639bec127ddc6b93adb0c7d8d995f77c)