Section: New Results

Robotics

Calibration and identification

Calibration of a cable-driven robot

Participants : David Daney, Julien Alexandre dit Sandretto, Jean-Pierre Merlet, Gilles Trombettoni.

To improve the accuracy of a cable manipulator, it is necessary to identify the uncertainties of its model. The robots, studied in Cogiro, an ANR National initiative, are redundantly actuated: the number of powered wires is larger than the number of degrees of freedom of the manipulator. Under some cable properties hypothesis, this over-constraint mechanism allows to perform a self-calibration - i.e. the identification of the parameters does not need additional external measurement. A first experimentation, done in Montpellier, validated a novel approach which consists in a simultaneous identification of parameters and robot position (unknown in self-calibration process).

Cable properties

Participants : Julien Alexandre dit Sandretto, Gilles Trombettoni, David Daney.

The majority of researches done on cable-driven robot modeling need to take into account that the mass and the elasticity of wires are neglectable. However, they can not prove that these hypotheses are acceptable regarding these objectives. We have proposed an algorithm based on interval analysis to judge the validity of these assumptions for a cable-robot in a specific workspace. This method have been tested on the Tecnalia/LIRMM's prototype and used for the construction of the Cogiro robot.

Optimal calibration poses of a 3-RPR planar parallel robot

Participant : David Daney.

The choice of the measurement configuration is crucial to improve the robustness of the calibration for measurement uncertainties. This year, a geometrical approach has been used to determine formally the set of the optimal poses for the identification of the kinematic parameter of a 3-RPR planar parallel robot. This result is important because it explains the influence of the location of some particular robot poses in their workspaces during the model identification process. A generalization is explored to construct automatically an optimal set for robot calibration and moreover, to improve experimental design algorithms. The aim is now to obtain similar results for cable driven robots calibration.

Geometric calibration of a space telescope

Participants : Thibault Gayral, David Daney, Jean-Pierre Merlet.

In October 2010 begun a collaborative work with Thales Alenia Space on the calibration of the mechanical structure of a space telescope. Its architecture is based on a parallel manipulator (type active wrist 6-PUS) used to correct the relative position of two mirrors. The aim is to identify the parameters of this robot, to improve its accuracy and then increase the quality of the images provided by the telescope. Thus, a geometric calibration procedure was considered and a campaign of photogrammetry was performed on the telescope. Using a kinematic description, a final accuracy of at worst 10 $\mu m$ was reached on the position of the platform of the telescope. The aim of these measures were also to valid or not the model of comportment of the flexible parts of the device. This campaign brought to light the necessity to consider forces and torques acting on the structure in the deformation of the flexible parts in order to reach a submicrometric accuracy.

Modelization of flexible articulations of the telescope

Participants : Thibault Gayral, David Daney, Jean-Pierre Merlet.

In order to improve the final accuracy of the above-mentioned space telescope, a novel model including the statics equations in order to calculate deformations of the flexible articulations is currently under study. The main difficulty is to identify parameters (stiffness matrix and geometric parameters) that have different units and are not of the same order of magnitude. To solve this issue, we are focusing our effort to write the problem in a better robust form.

Interval Identification

Participants : David Daney, Julien Alexandre dit Sandretto, Gilles Trombettoni.

There are many approaches to identify the parameters of a model. In most cases, it consists in providing a particular solution of an over-constraint set of equations which must be robust to measurement to errors: in least square sense, with some statistical properties.... However, the interpretation and the validity of the result can be difficult and prone error. We propose to investigate some interval approaches in order to associate to the result some information and a certification of solutions.

Rehabilitation and biomechanics

Participants : Sami Bennour, David Daney, Mandar Harshe, Jean-Pierre Merlet [correspondant] .

The focus of the work is on analyzing knee joint motion during a walking activity. The measurement system is based on the wire actuated parallel robot architecture. 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.

The main principle of the system is to observe relative motions of the collars attached to tibia and femur. These are connected to the base by wires and also hold the other sensors. Measurements in the global frame and collar specific local frames give precise data to reconstruct collar (and thus, knee joint) motion.

Over the past year we have finalized the experimental setup, by calibrating the collars and the sensor systems, and adapting the existing wire robot system (MARIONET-REHAB) to work along with the other sensors. The software developed uses a single unified input file to specify all sensor configurations, streamlining experiments. We performed our preliminary experimental trials for walking motion on three subjects using the wire sensors, accelerometers and optical motion capture system.

We began work on processing the data obtained from these trials. Post-processing functions have also been developed to calculate additional collar properties, perform sensor data processing (filtering, noise removal and estimation) and access files in the C3D file format, which is used a binary file format used by the motion capture system.

The main challenge we are working on is to perform sensor data fusion and increase reliability of results. For this we must identify parameters that correlate the different sensor measurements and perform error analysis. Possible solutions include using interval analysis methods to address the uncertainties.

Kinematics of wire parallel robots

Participant : Jean-Pierre Merlet.

The kinematics of wire robot is a complex problem because a solution is possible only if the tension in the wire is positive. Hence the static equilibrium has to be taken into account. This problem is not well addressed in the literature. Curiously the forward kinematics of robot (i.e. finding the possible poses of the platform for given wire lengths) with at least 6 wires is straightforward: the distance equations allows to determine all poses and then we use the static equations to calculate the wire tensions and discard the one having at least one negative tension. For robot having less than 6 wires we have to consider simultaneously the distance equations and the static equations in order to get a square system (of $n+6$ equations for a $n$-wires robot). We have investigated the case of a 3-wires robot with all wires attached at different points on the platform and have shown that all solutions can be computed provided the solving of an univariate polynomial of degree 158 [17] . Although we are not able to guarantee that the degree of this polynomial cannot be decreased, we believe that nevertheless the order will be too high for robust determination of the solutions and can only be used to determine an upper bound for the maximal number of solutions. We have also investigated theoretically and experimentally the kinematics of a $n$-wires ($n\ge 4$) robot with all wires attached at the same point (i.e. only the position of the center of the platform can be controlled). Although this robot is apparently redundant, we have shown that in any pose at most 3 wires will be simultaneously under tension and therefore that the redundant wires cannot be used to control the wire tensions.

As the wire length measurements are not sufficient to determine the current pose of the platform (which is necessary for control purposes) we are investigating the use of additional sensors. Our prototypes MARIONET-ASSIST and MARIONET-VR are instrumented to measure wire directions, but with a large uncertainties. We have started a theoretical investigation to determine under which conditions these uncertainties may lead to a non-unique solution and we will validate the results on the two prototypes.

Rehabilitation robots for the immersive space

Participants : Michael Burman, Jean-Pierre Merlet.

The on-site immersive room provides 3D visualization but is lacking of haptic feedback and motion capabilities. We plan to implement in this room a movable system, constituted of:

• a 6 degrees-of-freedom motion base: the motion system 710-6-500-220 by Servos Simulation Inc. has been selected and is now operational and fully calibrated. If necessary the user may stand on this motion base

• the MARIONET-VR wire-driven parallel robot: this robot uses the same actuation principle than the MARIONET-REHAB robot (linear actuator with a pulley system for coiling and uncoiling of the wires), but is able to lift a person. The prototype is basically functionnal but its installation in the immersive room has been delayed because of lack of appropriate fixing elements

The full system will be installed in the immersive room at the beginning of 2012.

Assistance robotics

Participants : Michael Burman, David Daney, Jean-Pierre Merlet.

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 communicant devices that may help disabled, elderly and handicapped people in their personal life, with the credo that they 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) [18] , [14] , [21] . We have started last year the development of a simulated 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. We describe in the following sections several devices that have been developed/improved during this year(pictures of this assistive flat are available at http://www-sop.inria.fr/coprin/developpements/main.html ). Note that our demonstration in assistance are highlighted during the visit of Sophia (275 visitors have attended our demonstration during 14 visits) and have received serious press coverage (5 papers, 2 TV interviews).

Walking aids

Wheeled walking aids are usually the first tools that are used when motricity problems occur. We are developing the family of robotized Assistive Navigation Guide (ANG), which are 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 (it is estimated that fall is the main cause of 10 000 elderly deaths per year in France).

• mobility help: provide an on-demand mobility help

• 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. These sensors allow to measure the trajectory of the walking aid and several features of the user's gait (step pattern, gait assymetry,...). ANG-light has been tested by the CHU of Nice-Cimiez that was willing to perform an in-depth investigation of its use. For that purpose we have asked in September 2009 for the necessary formal authorization to the local CPP, which has been granted only in December 2011. To prepare this study we have organized a large scale experiment at INRIA, where 24 users were asked to perform the trajectories of the protocol twice, with and without the aid. When not using the aid the users were equipped with 3D accelerometer on the wrists and knees and were using specific shoes with force sensors in the sole. Initial analysis of the records shows that indeed we are able to obtain significant information on the gait pattern, that are not available using the existing tools, and detect differences in the gait pattern for user having even a light pathology in the lower limb. The experiment with elderly patients at CHU will take place in January 2012.

• ANG-II : this aid is an evolution of the motorized walker ANG , with a lower weight and better integration

MARIONET-ASSIST

This wire-driven parallel robot is installed in the ceiling of the flat. It has been used this year in the 4-1 configuration (4 wires attached at the same point), which allows for controlling the position of the platform, but not its orientation. Several platforms have been developed, all of them incorporating a webcam and allowing for a free rotation around the vertical axis, while an accelerometer measure the tilt angle of the platform (which is used to determine in which direction the end-user is willing to move). One of the platform incorporate a 4 d.o.f. robot that may grasp light object (one of our objective is to use also the robot as a manipulator for bringing object back to the user in a more or less autonomous way, which is the subject of the PhD thesis of R. Ramadour).

We have shown that the the robot can be used for sit-to-stand transfer and for lifting handicapped people. A specific attention has been devoted to propose very simple control interface: joystick, remote TV set, control box whose tilt determine the motion axis.

Other flat equipments

Our scenario includes the management of emergency situations such as the fall. Fall detection can be performed by the ANG walkers but we have also started investigating the inclusion of fall detection system in the clothes of the end-user either through a GEO-300 devices or by incorporating an Arduino Lilypad processor. When a fall is detected indoor an alert is transmitted to a coordinator (a Nabaztag) which will order the walker and the MARIONET-ASSIST robot to move close to the user to provide a support. At the same time two mobile robots will converge to the same location: a remote-controlled, webcam equipped ROVIO (which can provide images of the end-user to a rescue center) and a Pekee II, that we will equip to provide first aid.

An important point in assistance is to be able to have at all time a rough idea of the localization of the patient. Although we plan to use a Kinect for that purpose, we will also investigate the use of non-vision sensors (which are much less intrusive and therefore can be more easily accepted) such as RFID tag (ANG-II has a RIFID tag reader), directive distance sensors and light barriers.

Another axis for assistance is to reduce the risks of fall by using the principle that the objects has to come to the hand of the user (or of the robot), not the opposite. This implies instrumenting the environment with drawer openers and doors manipulation and we have started implementing them on drawers and on the fridge of the flat.