Ruprecht-Karls-Universität Heidelberg

Poster Abstracts:

Posters I 13.05.2014, 15:40 - 17:00:

  • EU-Project
    WALKMAN
    Authors:
    P. Kaiser, D. Gonzalez-Aguirre, N. Vahrenkamp, J. Borràs, T. Asfour
    Abstract:

    One of the most fundamental research questions in robotics is how to provide robots with the ability to autonomously interact with an unknown scene. This is a high level problem that includes perception of the scene, recognition of the floor map and scene objects, association to possible actions, the ability to perform such actions in dynamically stable movements, and not only interact with the objects without losing balance, but to be able to increase stability by leaning on walls or objects. Humanoids are well known for their ability to perform well in scenes suited for humans, but they also have complex kinematic structures that require additional dynamic constraints to balance, compared to other mobile robots. While recent robotic research has solved partially some of the listed problems, many others still require major research and the process as a whole remains a challenging open problem.

    In the context of the WALKMAN European project, we propose the combination of several techniques for perception and manipulation of the environment to solve the above problem. Our preliminary results allow the extraction and segmentation of multiple view surfaces in the environment, including recognition of empty spaces, walls, floors and other patch planes, together with their localization with respect to the robot. We are working in a first layer of basic association of such entities to affordances for basic interaction. We plan to enhance this process to provide object recognition, allowing further links to affordances to form object action complexes.

    In parallel, we are applying previous group results on grasping to locomotion and manipulation with humanoids to naturally integrate upper and lower body movements. On one side, we intend to generate configurations of the robot in contact with the recognized planes in the scene (whole-body grasping) considering all the extremities of the body. On the other side, we plan to apply results on reachability maps that provide information not only about what entities in the scene can be reached, but also about quality manipulability measures of how such elements can be reached and manipulated.

    All together, we are integrating concepts of vision, artificial intelligence, locomotion and manipulation, working towards a full solution to the general problem of robot autonomy in unknown environments.

  • EU-Project
    Koroibot
    Authors:
    Debora Clever, Kathrin Hatz, Katja Mombaur
    Abstract:

    Human movement, as for example human gait, can be considered as an optimal realization of some given task. However, the criterion for which the naturally performed human motion is optimal, is generally not known.

    In this work, we formulate an inverse optimal control problem to study the relevance of four different optimization criteria in human locomotion. As a walking model we use an actuated three dimensional spring loaded inverted pendulum, with one leg in single and two legs in double support. Using a direct all-at-once approach, the weighting of the optimization criteria and the position of the footsteps are optimized in such a way, that the center of mass trajectory of the resulting optimal state fits real motion capture data as good as possible.

    Numerical experiments show, that whereas the so called capture point seems to have a great impact on human walking, minimization of the vertical center of mass movement does not show any relevance at all.

  • EU-Project
    Koroibot
    Authors:
    Yue Hu
    Abstract:

    Coming soon

  • EU-Project
    CODYCO
    Authors:
    Serena Ivaldi, Roberto Calandra, Alexandros Paraschos, Elmar Rueckert, Jan Peters
    Abstract:
    TBD
  • EU-Project
    H2R
    Authors:
    Karsten Berns, Steffen Schütz, Jie Zhao
    Abstract:

    The goal of H2R project is to demonstrate human-like gait and posture in a controlled compliant biped robot as a result of a combination of the most relevant motor control and cognitive mechanisms found in humans.

    In order to achieve this goal, a threefold process will be adopted:

    1. Understanding the key biological principles from human experiments.
    2. Translating the formalized concepts into human-like bipedal robots.
    3. Creating new benchmarking schemes for validating the robotic performance.

    Studying Humans
    In healthy humans, walking and posture emerges naturally from a hierarchical organization and combination of several biomechanical and neuromotor mechanisms. The H2R project aims to unveil the most crucial principles, such as synergistic control, cognitive prediction, and muscle dynamics, for their posterior inclusion in robotic real-life platforms.

    Developing robots
    Current biologically motivated bipeds rely on the concepts of passive walking, compliant actuation and neural-based control to enhance stability, versatility and efficiency. The goal of H2R is to put together the most promising control and biomechanical principles currently available in real-life and simulated platforms. This process will result in a new biped with improved walking and postural abilities.

    Benchmarking scheme
    Benchmarking research in robotics is inherently difficult since results are typically reported only for a specific robotic system and a self-chosen set of tasks. This makes it extremely difficult to compare the results with other systems developed in different labs. The H2R project wants to involve the international community in defining a benchmarking scheme that can reach a global consensus and be universally adopted.

  • EU-Project
    H2R
    Authors:
    Vittorio Lippi, Thomas Mergner, Georg Hettich, Lorenz Asslander
    Abstract:
    A bio-inspired modular system for robot postural control is presented. The system is based on the disturbance estimation and compensation, DEC, concept, that internally reconstructs external disturbances using multisensory integration. The DEC concept has been developed from neuroscientific model interpretations of human postural responses to external disturbances. Its simulations in a sensorimotor control model reproduced the basic features of the human data. This also applied when the model was re-embodied into a robot and tested in the human laboratory. Outstanding features of the model are low loop gain, tolerance of time delays, and automatic adjustment to changes in external disturbance scenarios. The proposed system uses DEC modules in a modular control architecture that allows the user to control robots with several DOF and joints. Each DOF is controlled by a DEC module and connected with functionally synergistic modules in the neighboring joints. In the control layout, the connections distribute vestibular and visual information from the 'head' to the 'joints' where they are fused with joint angle and torque information. Further connections provide each DEC module with information on the kinematic state of the supporting joint and the orientation of the supported distal links with respect to the gravitational vertical. The modular design keeps the complexity of the control low and error robustness high compared to monolithic control architectures. The control framework is implemented as a software library for Matlab/Simulink. The use of the library and some interesting behavioral features of Posturob II, a humanoid human-size robot, are briefly presented.
  • EU-Project
    Koroibot
    Authors:
    Maximilien Naveau, Olivier Stasse, Christian Vassallo
    Abstract:

    Since the 25th of November 2013, the french robotic laboratory LAAS-CNRS is part of the European project : Koroibot. The project aim to teach two-legged robots a stable, robust "human" way of walking. In that context we, as the Gepetto team, are studying the behaviour of the human walking during an obstacle race by the use of motion capture. The experiments are done in collaboration with the Universities of Tubingen (Germany) and Renne (France). Another main task is to find a way to implement the human behaviour in the robot HRP2. For that purpose, we need to study the data from the experiments and to increase the stability of the current walking pattern generator to be able to improve the behaviour of the robot.

  • EU-Project
    Koroibot
    Authors:
    Nikolaus Vahrenkamp, Ömer Terlemez, Christian Mandery, Stefan Ulbrich, Martin Do, Tamim Asfour
    Abstract:

    Master Motor Map (MMM) is a conceptual framework for perception, visualization, reproduction, and recognition of human motion in order to decouple motion capture data from further post-processing tasks, such as execution on a humanoid robot. Employing MMM makes it easy to map motions between different kinematics independently and uniformly as well as to analyze certain dynamic aspects of the considered motion.

    The MMM reference model consists of a particular kinematic structure enriched with pre-defined segment properties (anthropometric data) e.g. mass distribution, segment length, moment of inertia, etc.

Postersession II 14.05.2014, 15:10 - 16:30:

  • EU-Project
    Romeo-2
    Authors:
    Mehdi Benallegue, Florent Lamiraux
    Abstract:

    Keeping balance requires to ensure fulfillment of criteria on contact forces and the position and acceleration of the center of mass. These parameters are then controlled for humanoids to maintain equilibrium. But due to modeling errors, external perturbations and especially robot's non-modelled flexibility, their actual values can differ from the control output, and these parameters have to be corrected and stabilized. Most solutions for stabilization today, use force measurements on contact points, they rely on a model of the flexibility dynamics and a fine calibration of force sensors to estimate the actual state of the center of mass. Our research aims at bringing the problem to a kinematic perspective and use only the geometrical information provided by the IMU and the contact points to build a flexibility estimator which feeds a simple CoM stabilizer. This research would enable to stabilize robots which don't have force sensors, this would also help to understand the role of human vestibular system to keep balance.

  • EU-Project
    MOBOT
    Authors:
    Davide Corradi, Khai-Long Ho Hoang
    Abstract:

    The MOBOT project aims at developing mobility assistance devices to support the motions of elderly people with mobility impairments, in particular when walking, standing up and sitting down. The work described on this poster concerns the development of a method that allows an online adaptation of the shape and actuation of the mobility assistant in a way that optimally supports the motion of the subject in every instant.

    To this end, the device should try to imitate the support given by a professional human carer, by compensating lack of balance and strength to avoid falls and other dangerous situations. Help should be provided only if needed to avoid developing an over-reliance of the user on the device.

    The challenges of the project lie in the interaction between a controllable system (rollator) with a not controllable and not fully known biomechanical system (user), and in the biomechanical modeling of elderly people with mobility impairments, whose dynamic parameters are not extensively studied in literature and whose walking dynamics differ from those of healthy young people.

    The first step to reach the goal is to develop models of the device and of the user. Then, an online optimal control that allows the device to compensate lack of strength and balance in the user will be developed. In this regard, Nonlinear Model Predictive Control (NMPC) will be used. Finally, the control will be implemented on the real device and integrated with the software modules provided by the other partners of the project. Offline optimal control simulations during the relevant scenarios (walking, standing up...) will also be performed to get an insight on objective functions and contraints in the motions of elderly people.

  • EU-Project
    MOBOT
    Authors:
    Khai-Long Ho Hoang, Davide Corradi, Katja Mombaur
    Abstract:

    Based on the kinematic data gathered during Motion Capture experiments, methods are developed to identify specific gait patterns of the MOBOT user group. They are classified according to the temporo-spatial parameters that allow for a reliable estimation of the gait performance of an individual patient. The resulting mobility classes can be used to adapt the sensitivity of the safety features of the assistive device.

    The identification of stable and unstable walking motions enable us to develop control strategies that provide optimal assistance and safety for the user. Considering the multiple ground-patient and device-patient contacts we develop criteria to quantify the stability of the subject as well as rate the feasibility of its desired trajectory while allowing for both dynamic locomotion and maximal security margins in the case of emergency.

  • EU-Project
    JSPS fellowship project
    Authors:
    Jovana Jovic, Adrien Escande, Gentiane Venture, Eiichi Yoshida
    Abstract:

    We propose a method for estimation of subject specific physically feasible body segment inertia parameters. The method is based on hierarchical quadratic programming optimization approach. The method is validated by performing the experiments with a human subject and comparing the results of the estimated ground reaction forces and moments with those obtained using state of the art anthropometric data bases. The results of the study showed that the proposed method is able to successfully estimate the ground reaction forces and moments. The method was also able to detect additional masses placed on the subject during the experiments contrary to commonly used methods based on anthropometric tables.

  • Cooperating - Instituts
    ORB-IWR - LAAS-CNRS
    Authors:
    Henning Koch, Katja Mombaur, Philippe Souères
    Abstract:

    The research field of motion generation is far spread. Typically one would distinguish between, methods that are mainly based on heuristics (e.g. pattern generators) and those that are optimisation based (optimal control).

    Pattern generators are based on highly simplified models (linear inverted pendulum, lumped mass models), operate sufficiantly quickly for realtime applications and are easy parametrisable.

    Optimal Control methods usally depend on more complex models, are therefore computationally much more expensive. Even they rarely perform quickly enough to permit real-time applications, these methods give a much more intuitive access to complex motion characteristics.

    In the following we propose an optimal control optimisation approach that depends solely on the model dynamics of the system and the governing physical events of the motion at hand.

  • EU-Project
    LOCOMORPH
    Authors:
    Rico Möckel, Soha Pouya, Christophe Maufroy, Peter Aerts, Helmut Hauser, Auke Jan Ijspeert, André Seyfarth
    Abstract:

    We present the Locomorph robot system - a modular system for the creation of legged robots suitable for the validation of locomotion models and studying locomotion of legged animals.

    Dynamic models have been proven to be a powerful tool for studying the locomotion of legged animals. Among the many models with different level of complexity the SLIP model has been shown to be a valid abstract model for studying running in animals. But we and others as well have been presenting and studying dynamic models that model animal legs in more detail. The search for a valid template ? a model that can generally explain and predict locomotion of legged animals at a level of abstraction that does not involve too detailed features ? is not satisfactory solved. Robots are valuable tools for the validation of abstract locomotion models that typically do not realisticly model physical effects like for instance friction and slipping.

    We built a new modular robot system to create legged robots that allow a direct transfer from popular templates like the SLIP model to the robot - a feature that for instance is not well supported by a robot with a segmented leg design. Our goal was the creation of a flexible, high performance modular robot system that can be used for the validation of a variety of models and to study the locomotion capabilites of these templates rather than of a specific animal. We believe that through this abstraction of animal leg mechanics with the Locomorph robot system we will be able to study legged locomotion gaining an abstract understanding that applies to a variety of animals rather than just a specific one.

  • EU-Project
    ECHORD-GOP
    Authors:
    Martin Felis, Benjamin Reh, Johannes Schlöder, Katja Mombaur
    Abstract:

    The generation of the best possible path that does not violate any constraints imposed by the environment is an ubiquitous task for robots. Currently there is no algorithmic approach available that allows to address this problem for very complex dynamic robot systems in cluttered changing environments in real time. Instead there are two established but still quite separated research areas that both address a part of the problem, namely path planning and numerical optimal control. However the treatment of a huge number of environmental constraints giving rise to many local minima makes the problem very hard, if not impossible, to solve. In our ECHORD experiment GOP we developed a framework that combines two method and testet it successfully on an industrial robot arm, namely the KUKA KR5 sixx R850.

  • EU-Project
    Lola
    Authors:
    Thomas Buschmann, Arne-Christoph Hildebrandt, Robert Wittmann
    Abstract:

    The main drawback of current biped walking machines is their limited ability to handle previously unknown situations during practical use in real-time. These could be large disturbances or the complexity of generating walking patterns in cluttered environments. In order to face these challenges we present two methods

    1. to enable collision-free movements in cluttered environments and
    2. to increase the robustness against unknown disturbances.

    First, we present a method which generates locally optimized trajectories online during the feedback control to dynamically avoid known obstacles. This method successfully combines a local potential field method with a heuristic based on height and width of an obstacle to avoid collisions.

    Second, we developed an estimation model to predict the robot's behaviour which enables the computation of stabilizing motions. The robot is approximated by a three-mass model with two degrees of freedom, unilateral compliant contacts and a closed-loop stabilizing control. With this model it is possible to predict the robot's behavior during the next step for a given measured state.

    The two methods are successfully integrated into the real-time control system of our biped LOLA. Simulations and experimental results demonstrate the effectiveness of the proposed methods.

  • Cooperating - Instituts
    CNRS (Centre National de la Recherche Scientifique) - AIST (National Institute of Advanced Industrial Science and Technology)
    Authors:
    Eiichi Yoshida, Abderrahmane Kheddar
    Abstract:

    CNRS-AIST JRL (Joint Robotics Laboratory), UMI3218/CRT is a joint laboratory located at AIST at Tsukuba, Japan and recognized as the Unite Mixte Internationale (UMI, International Joint Research Unit) for CNRS and concurrently as a Collaborative Research Team (CRT) for AIST, and defined as UMI/CRT. The researchers from both countries are closely collaborating to pursue the means of increasing robot's functional autonomy, using a humanoid robot as a main platform.

Last modified by: Henning Koch on 2018-05-30
to the top of the page