Previous | 
Home
Section: New Results
Robotics
Robotics
Kinematics of wire-driven parallel robots
Participants : Laurent Blanchet, Jean-Pierre Merlet [correspondant] .
The kinematics of wire robot is a complex problem because it involves both the geometrical constraints and the static equilibrium constraints as only positive tensions in the wire are possible. A major issue, that has not been addressed in the literature [16] , [15] , is that for a robot having wires the forward kinematic problem (FK) (determining the possible pose(s) of the robot knowing the wire lengths, a problem that is crucial to solve for controlling the robot) cannot be solved by assuming that all wires are under tension as the current pose of the robot may be such that only a subset of the wires may be under tension. Hence the FK problem has to be solved for all robots that may be derived from the initial one by removing 1 one to wires, each solving leading to a set of possible poses for the platform. Solving the FK for 1 wire is trivial, while for 6 wires the FK solving may be based on the already complex FK of parallel robot with rigid legs. For 2 wires it can be shown that the FK solutions can be found by solving a 12th order univariate polynomial, while for 3 wires we have shown last year by using an elimination procedure that the solutions are obtained by solving a 158th order polynomial. A very recent result of this year is that for 4 wires the order of this polynomial is 216, while no known result has been established for 5 wires (note that for 3 to 5 wires the size of the system of equations that has to be solved for the FK is larger than the one for the FK of 6-dof robot with rigid legs, a problem that has required 20 years to be solved).
Drawbacks of the elimination approach is that it does not take into account 1) that the solution should be mechanically stable, 2) that the wire tensions at the solution(s) must be positive. Hence, assuming that all solutions may be computed by the elimination approach, an a-posteriori analysis has to be performed to sort out the solutions that verify 1) and 2). We have proposed this year an efficient method to determine if a solution was mechanically stable [13] . But another major issue with the elimination method is that it leads to high order polynomial that cannot be safely numerically solved. To address this problem and 2) we are considering a numerical algorithm based on interval analysis, that consider also the tension as unknowns, hence allowing to search only for solution(s) with positive tensions [21] .
Another issue for wire-driven parallel robots is the concept of redundancy. Having more wires than dof to be controlled is interesting for increasing the workspace of the robot. But it is believed that redundant wires may also be used to better distribute the load among the wires. Unfortunately we have shown for the robot (all wires connected at the same point on the platform) with non elastic wires that whatever there will be at most 3 wires under tension simultaneously [25] and consequently that tension management is not possible (with 3 wires the tensions is uniquely determined). If the wires are elastic, then tension management is possible but the positioning error is very sensitive to errors in the stiffness model [24] . Hence new method for tension management should be devised and we have explored some possibilities [23] . Still there is some magic in wire-driven parallel robots: in spite of all the uncertainties prototypes work quite well, a phenomenon which has been explained through a sensitivity analysis [24] , [22] .
Finally we address the management of modular robots, whose geometry can be adapted to various tasks and different objects to be manipulated, especially for very large scale robot [28] , that may be used in industry for maintenance and logistics (see the Cablebot project in section 8.2.1 ).
Robot Calibration
Participants : Thibault Gayral, David Daney [correspondant] , Jean-Pierre Merlet.
Experimental calibration of a high-accuracy space telescope
A collaborative work began in October 2010 with Thales Alenia Space on the calibration of the mechanical structure of a space telescope. Its architecture is based on a parallel manipulator (of the active wrist 6-PUS type, which has been patented by COPRIN) and is used to correct the relative position of two mirrors. The aim is to reach a micrometer accuracy in order to obtain a suitable quality of the images provided by the telescope. Thus, a complete model of the space telescope needs to be developed and validated through calibration. Since high velocity is not required in such an application, the dynamic effects can be neglected and only geometric and/or static calibration has to be considered. Moreover, measurements for calibration were performed in a clean room under controlled pressure, temperature and humidity conditions to minimize the influence of the non-geometric errors. Thus, two possible static inaccuracy sources were identified and modeled: one from the deformation of the mobile platform and the other resulting from the behavior of the flexure joints. Three incremental models of the flexure joints were developed and compared: a spherical joint model, a model issued from the beam theory and a stiffness model. Results of calibration using an accurate measurement system of photogrammetry showed that the flexure joints can be modeled by perfect spherical joints due to the small workspace of the telescope. Concerning the mobile platform deformation, two models were developed. Good accuracy results were obtained for both models. The developed models allowed us to explain how the model errors are directly accounted in the parameter identification during calibration. This resulted in different sets of identified parameters which all enable a good positioning accuracy. Those differences were explained and results of calibration allow a proper choice of the model of the mobile platform deformation. Considering this model, a positioning accuracy of some micrometers was finally reached after calibration with only position and orientation measurements of the mobile platform, which should allow the calibration of the telescope in space [33] . This is currently under study using interferometric measurements on the prototype of the space telescope.
Calibration of a cable-driven robot
To improve the accuracy of a cable manipulator, it is necessary to identify the uncertainties of its model. The cable robots, studied in the ANR funded project Cogiro (see section 8.1.1.2 ), are theoretically redundantly actuated: the number of powered wires is larger than the number of degrees of freedom of the manipulator (however see section 6.2.1.1 about the reality of this redundancy).
In 2011 an over-constrained prototype was self-calibrated (the identification of the parameters does not need additional external measurement), under some assumptions on the cable properties [17] , [29] . We will apply our recent calibration methods on the large scale robot prototype developed for the Cogiro project at the very end of this year.
Cable properties
Quite often cable-driven robot analysis assume mass-less and non-elastic wires. We proposed a method based on interval analysis to judge the validity of this assumption for a particular robot in a specific workspace. Our aim is to use this method in order to determine a a region within the robot workspace for which the hypothesis is valid and consequently for which self calibration of the robot is possible. Indeed, the assumption on the cable properties is not acceptable over the full workspace of the large scale robot developed in the Cogiro project. Still a self-calibration is possible if calibration poses are chosen within a specific subpart of the workspace. A more efficient calibration approach is in progress with additional measures and a more complex model (static and elasticity). The results has been published in [18] , [30] .
Assistance robotics
Participants : David Daney, Claire Dune, Jean-Pierre Merlet [correspondant] , Yves Papegay, Odile Pourtallier.
As mentioned earlier in the report we have started in 2008 a long term strategic move toward assistance robotics, with the objectives of providing low-cost, simple to control, robotized smart devices that may help disabled, elderly and handicapped people in their personal life, provide also assistance to family and caregivers while allowing doctors to get better and objective information on the health state of the end-user. Our credo is that these devices have to be adapted to the end-user and to its everyday environment (by contrast with the existing trend of focusing on a "universal" robot, to which the end-user and its environment have to adapt). As for cost reasons we intend to use only standard hardware adaptation has to be taken into account at the very early stage of the system design and uncertainties in the physical instances of our systems are also to be considered.
For validation purposes we have developed a flat in order to explore various full scale scenarii that cover a part of the daily life of an elderly, to develop specific assistance devices and to test them (pictures of this assistive flat are available at http://www-sop.inria.fr/coprin/prototypes/main.html ). Our activity in this field is concentrated on transfer, manipulation, walking monitoring, rehabilitation and the use of virtual reality. We are also investigating how such complex environments with multiples smart agents, quite heterogeneous from a computing viewpoint, but that have to cooperate, may be programmed. All these topics are in accordance with the one of the large scale initiative PAL (http://pal.inria.fr ) of which we are an active member.
Transfer and manipulation
Participants : François Chaumette [Lagadic] , Jean-Pierre Merlet, Rémy Ramadour.
For transfer operation we are using the MARIONET-ASSIST robot (see section 6.2.1.4 ) that is installed in our flat. Currently we use 4 winches with the wires connected at the same point on the platform, hence providing 3 translational degrees of freedom. This low-cost, low-intrusivity robot has proved to be very effective for transfer operation. Apart of transfer operation robot may be used for manipulation. Adding one or several low-cost cameras (the cost being here a fundamental constraint), visual-servoing control is used to provide a whole new set of useful services such as grasping objects in order to bring them to the end-user (if they are too heavy, too far, high or low), or cleaning the table after lunch. Using a parallel crane robot, we are able to cover a large workspace, the vision-control allowing us to obtain the precision required by the manipulation of daily-life objects. The collaborative implementation of the vision and the kinematic control of the robot gives us a way to make best use of the advantages of both parts, while overcoming their respective drawbacks.
Given a region where the object of interest belongs, the first step is to detect it in an evolutive environment. A segmentation is made, robust to luminance variations and perspective projections. The vision is then used to move the platform toward a desired position relatively to the target. In order to execute this task, some carefully chosen features are measured, allowing to estimate the incremental displacement required to move the end-effector to the desired place. We use the library ViSP for both the detection and the visual-control part.
Experimental results were obtained using a platform with 3 degrees of freedom and a single camera, grasping a single object and moving it from a place to another. We used for that basic image data such as 2D moments, allowing a fast computing and yet robustness in measurements. We currently address to generalize this manipulation to other configurations and to evaluate its robustness to calibration errors and other uncertainty sources. We also are looking for a global paradigm merging both the vision-based kinematic model and the mechanical one, which could significantly improve the efficiency of the experiment, while reducing the mathematical complexity behind each kinematic model considered on its own.
Walking monitoring
Participants : Claire Dune [Handibio] , Jean-Pierre Merlet.
We use the walking aids ANG-light and ANG-II (see section 6.2.1.4 ) to monitor the trajectory of the walking aid. The on-board sensors of these aids allow to evaluate the step pattern, gait asymmetry,...during daily walking, hence providing an health monitoring system that is always available. ANG-light has been tested last year with 24 subjects that were themselves instrumented (accelerometers in the wrists and knees, pressure sensors in the shoes) and were asked to perform two different trajectories twice with/without the walking aid. The purposes were:
to obtain a “gold” standard of these indicators for non pathological walking, taking into account the normal variability of the walking pattern
to determine if indicators obtained with the walking aid may led to an accurate estimation of the indicators when the walking aid is not used
Several indicators have been determined after the analysis of these data. In a second phase the inclusion test of elderly people (30 subjects) is taking place at the CHU of Nice-Cimiez and will last until the first trimester of 2013. An analysis of the collected data, in close collaboration with the doctors, will allow to determine if the proposed walking indicators are pertinent.
Another interest of the walking aids ANG is that they allow to collect significant information for mobility during their daily use: slope and surface quality of the sidewalks and automatic detection of lowered kerbs with a ranking of their convenience. It will be interesting for a community to share such information that is collected by the community members. For that purpose we propose to use collective maps, such as OpenStreeMap, which allow for map annotation. To validate this concept we have used ANG-light to automatically annotate the map of the Inria Sophia site with pertinent information for walking aid and wheelchair users (see http://www-sop.inria.fr/coprin/prototypes/main.html#ang ).
Rehabilitation
Participants : David Daney, Mandar Harshe, Sami Bennour, Jean-Pierre Merlet [correspondant] .
The focus of our work is on analyzing knee joint motion during a walking activity. The main principle of the system is to observe relative motions of the collars attached to tibia and femur. The measurement of the motion of these collars is based on the wire actuated parallel robot architecture (using the MARIONET-REHAB robot, see section 6.2.1.4 ). To increase the reliability of our analysis, and decrease the influence of Skin Tissue Artifacts (STA), we also incorporate a passive wire measurement system, IR camera based motion capture system, accelerometers, and force sensors to measure human motions.
Measurements in the global frame and collar specific local frames give precise data to reconstruct collar (and thus, knee joint) motion. The system developed already incorporates the optical motion capture, inertial measurement units and the wire sensors for comprehensive coordinated measurements of the motion of the knee.We have performed preliminary trials on three subjects for walking motion.
In the past year, we worked on processing the data to obtain pose and orientation information of knee joint. Data obtained from the trials was analyzed and post-processing steps were implemented to reduce noise and errors. In order to perform sensor fusion, we implemented a probabilistic estimation based method to estimate the pose.
The results from these analysis have allowed us to identify the merits of our approach and also helped us identify improvements that are needed. We have also identified the possible changes to our mathematical model that could allow use of interval analysis tools along with probabilistic estimation methods. We have identified the changes needed to the hardware setup that will help reduce the sensor noise and error. These changes once implemented will allow us to improve the usability of the system and also point us towards newer areas for further investigation, including, for example, effect of sensor placement, collar design, and interval based extended Kalman Filters for pose estimation[8] , [9] .
Virtual reality
Virtual reality has proved to be an effective mean for dealing with rehabilitation, provided that motion is added to the 3D visual feedback. The MARIONET-VR robot, together with our motion base (see section 6.2.1.4 ) may provide very realistic motion in an immersive room and we will start in 2013 a collaboration with the VR4I and REVES project-teams on this issue. A first task will be to install a moving walking treadmill in the immersive room at Sophia and to combine motion of the treadmill with 3D viewing.
Programming
In our opinion there will not be a single assistance device that will be able to offer all the required help that may be needed but rather numerous redundant smart agents that are able to perform very efficiently (at low cost and with a small energy consumption) a set of specific tasks. Such agents must be able to communicate and possibly will have to cooperate in some cases (e.g. after a fall). They will be heterogeneous from a computer view point as the used agents will change according to technological advances and to the trajectory of life of the end-users. If the manual programming of a single agent is possible (although quite complex for some of them) the overall system cannot be managed in that way: we need an unifying framework for this development. For that purpose we are currently investigating the use of HOP, a multi-tier programming language for the diffuse Web developed in the INDES project-team. Already one of our wire-driven parallel robot MARIONET-SCHOOL has been programmed in HOP and can be seen as a web resource.
Prototypes
Participants : Julien Alexandre Dit Sandretto, David Daney, Claire Dune, Jean-Pierre Merlet [correspondant] .
Experimental works are a key point in the field of service robotics: it allows for validating concepts, getting feedback from the end-users and discovering new problems. We have extensively developed prototypes this year (pictures and videos of our prototypes are available at http://www-sop.inria.fr/coprin/prototypes/main.html or on our YouTube channel http://www.youtube.com/user/CoprinTeam/videos?flow=grid&view=0 )
Wire-driven parallel robots
The MARIONET family is now constituted of
MARIONET-REHAB (2004-): using up to 7 linear actuators it is mainly used for rehabilitation and health monitoring although it is also a very fast pick-and place robot
MARIONET-CRANE (2008-): a very large 6-dof rescue crane, portable and autonomous that can be deployed in 10 minutes and has a lifting capability of 2 tons
MARIONET-ASSIST (2011-): a robot deployed in our flat with up to 6 winches, that is used for transfer operation and health monitoring, with a lifting ability of about 2000 kg
MARIONET-VR (2011-): using up to 6 linear actuators it will be used in the Sophia immersive room for simulation and rehabilitation. It allows to fully lift a person and has been physically installed in the immersive room this year although it is not fully operational
MARIONET-SCHOOL (2012-): a set of very low-cost robots that are intended to be used for dissemination. They fit in a small suitcase and can be deployed on a table or over a full classroom. We believe that such robots may be used by roboticists but also by researchers from other domains that may use the motion of the robot(s) to illustrate scientific concepts. Currently we have 3 such robots (one using Lego components, one with step motors and one with servos)
Walking aids
The Assistive Navigation Guide (ANG) family is based on commercially available Rollators. with several objectives (we mention only a few of them):
fall prevention/detection: fall is a major problem for elderly
gait pattern monitoring: we believe that being able to monitor the trajectory of the walking aid will provide useful information on the gait pattern of the user
For reaching these objectives we have developed two walking aids:
ANG-light : a walking aid with encoders in the wheels, 3D accelerometer, gyrometer and GPS. These sensors allow to measure the trajectory of the walking aid and several features of the user's gait. This walking aid is currently being used at CHU Nice, see section 6.2.1.3.2 . A replica of ANG-light is currently under development at the Handibio laboratory of Toulon and will include force sensors in the handles to get measurement of muscular activities while walking.
ANG-II : this aid is an evolution of the motorized walker ANG , with a lower weight and better integration
Other devices
As seen in the previous sections we have focused our work on mobility as it has been identified as a major priority during our two years interview period. Another priority is fall management: it is adressed with the ANG 's) but requires that the patient uses a walking aid. To obtain a better coverage we have developed an instrumented vest that includes an Arduino Lilypad microcontroller to monitor a 3D accelerometer and verticality sensors in order to detect a fall. This system may communicate an alarm through wifi, zigbee or even infrared signals. Apart of its low cost the system is washable and has a very low power consumption (we are considering energy harvesting to further increase its energy autonomy).
Two other important issues for assistance robotics is activities monitoring and patient localization. In cooperation with the STARS project-team an activity monitoring system within the Dem@care project. Its purpose is to log specif events such as taking a pen, setting on a kettle, ...using only simple sensors such as proximity and distance sensors, switches, that are more reliable and less complex than using a vision system. We have provided to the CHU a system that allows to monitor up to 208 different events. In the same manner we are working on a localization system of elderly in a room based only on distance sensors, that will be less intrusive and/or may complement the measurements of a vision system.
Rehabilitation is also part of our activities. Apart of the MARIONET-REHAB system we plan to investigate the use of Sophia immersive room for that purpose. To complement the motion that will be provided by the MARIONET-VR robot we have bought a 6-dof motion base from Servos with a nominal load of 150kg, that we have modified to accommodate our needs. We will use also two lifting columns with a load of 100 kg each, that will allow to manage motion for rehabilitation apparatus such as treadmill.