An Integrated Approach to the Design of Complex Robotic End-effectors

(1)

1

An Integrated Approach to the Design of

Complex Robotic End-eﬀectors

L. Biagiotti, F. Lotti, C. Melchiorri, G. Vassura

DEIS - DIEM, University of Bologna Via Risorgimento 2,

40136 Bologna, Italy

{lbiagiotti, cmelchiorri}@deis.unibo.it, {fabrizio.lotti, gabriele.vassura}@mail.ing.unibo.it Abstract— In this paper, a novel design approach for

the development of robot hands is presented. This ap-proach can be considered alternative to the “classical” one for several reasons. First, the overall project ad-dresses, besides the widespread speciﬁcations of dex-terity and anthropomorphism, issues that some time are neglected or not fully considered in the design of robotic end-eﬀectors, e.g. the structural complexity of the system, its reliability and costs.

In order to achieve these results the tools used are also different from the traditional ones: compliant struc-tures (e.g. elastic hinges) have been preferred to more standard robotic technologies, like pin bearing joints. Last, to cope with the problems due to the adopted mechanical technology (for instance undesired compli-ance effects) a mechatronic approach has been followed, which takes into account the all the possible interac-tions between mechanics, electronics and above all con-trol. In the paper, this research activity is presented and the preliminary experimental tests on a first pro-totype are reported and discussed.

Keywords: Articulated Hands, Mechatronic Design,

Compliant Mechanisms, Manipulation, Control

I. Introduction

In the last two decades, a number of advanced robotic end-eﬀectors has been developed in the labo-ratories of universities and institutions (for a complete overview see [1]). Main aim of these robotic devices is to reproduce the functional capabilities (and often the structure) of the human hand, and hence are usually named “hands”.

In the ‘80s, the seminal works of Salisbury and Jacob-sen paved the way for dexterous robotic hands, that is hands able to deal with unknown objects, in a possibly unstructured environment. Afterwards several devices have been designed with the main purpose of achieving dexterity. These end-eﬀectors are characterized by dif-ferent mechanical structures, sensory equipments and control strategies, but all of them share a common drawback, that is the complexity of both the hard-ware and the softhard-ware. In this paper a novel approach to the design of advanced robotic hands is proposed, which, besides dexterity and anthropomorphism, tries

to face the problem of the structural complexity of such robotic end-eﬀectors.

II. Issues in robotic end-effectors design

As mentioned in the introduction, the main draw-back and probably the main limiting factor for a wider diﬀusion of robotic hands outside research laboratories is their complexity.

Dexterous manipulation requires a high number of controlled degrees of freedom and consequently a large number of actuators, which must be hosted in an extremely small space (if a self-contained device is aimed). At the same time, in order to achieve dexter-ity, an adequate sensing apparatus is necessary. Be-sides the standard position sensors, other sensors able to detect forces/torques exchanged during the interac-tion with the environment are not only desirable but, in eﬀect, mandatory. In fact, all the robotic hands pre-sented in the last years are equipped with force/torque sensors, Intrinsic Tactile and/or tactile array sensors (see [1], [2]). If the richness of force and tactile infor-mation greatly contributes to the overall dexterity of a hand (as well demonstrated by the human beings), on the other side it requires a noticeable growth of complexity.

III. An integrated approach to the development of a new robotic hand

Considering that dexterous hands are too complex, and conversely commercial hands have low functional capabilities, how is it possible to find a trade-off be-tween these opposite trends? In the following, we try to give a possible answer to these basic questions. In our design of a new robotic hand, problems like complexity, reliability and costs have been faced since the beginning, together with functional specifications, adopting a mechatronic approach. That may sound obvious: as a matter of fact, all the examples of robotic hands reported in the literature show a deep

(2)

integra-Mechanics Electronics Control (information technology) oftware Hardware Control (information technology (information technology s Mechanics (a) (b)

Fig. 1. Traditional approach to robotic hand design (a) and

fully mechatronic design approach (b).

tion between mechanic and electronic parts. On the other hand, the third element commonly recognized in a mechatronic design, that is information technology and control theory [3], [4], is often added in a second stage, after the design of the hardware (see Fig. 1.a). In this way, control algorithms must be compliant to a speciﬁc structure, and cope with the drawbacks that such a structure may present. This approach adds complexity to complexity (as a matter of fact control strategies must already manage a large number of ac-tuators and an equally large number of sensors in or-der to get the desired capabilities). In particular, side eﬀects of the mechanical design (e.g. friction phenom-ena or backlashes) need sometimes heavy compensa-tions. In order to reduce the overall complexity of an articulated robotic hand, preserving its dexterity, a traditional mechatronic approach has been assumed, and, besides the mechanics and the electronics, also control issues have been taken into account from the beginning. In this way it has been possible to exploit the positive features of each component, and to opti-mally balance each part.

IV. A “nontraditional” robotic finger design

The starting point of this research activity has been the consideration about the appropriateness of the traditional mechanical solutions applied to dexterous robotic hands. It is in authors’ opinion that mechani-cal structures inspired to biologimechani-cal models (in partic-ular to the obvious model given by the human hand) may enhance the success of the overall design, con-cerning in particular its simplicity. Therefore, at the laboratories of DEIS and DIEM at the University of Bologna it is currently under evaluation a ﬁnger struc-ture based on the so called “compliant mechanisms”, i.e. chains of rigid links connected through elastic hinges allowing relative motion between them. The

Fig. 2. Prototype of the finger.

interest towards compliant mechanisms and the inves-tigation on their properties and design criteria have been rapidly growing in the last years, with significant applications in many fields, including MEMS (micro electro mechanical systems) and robotics [5], but ap-plication of compliant mechanism concepts to robotic end-effectors has been so far limited to small-scale ma-nipulation grippers [6].

The new structure tries to reproduce the biological model, whose frame is obtained by separate bones con-nected by ligaments. In particular the endoskeletal concept, which is the base of the new design, can eas-ily integrate distributed sensing capabilities as well as an external “skin” reproducing the tissues of the hu-man hand. As a matter of fact, it has been shown that the presence of an external soft layer can greatly im-prove the manipulation capabilities of robotic ﬁngers [7], in terms of stability of the interaction (because of dissipation introduced by the nonlinear visco-elastic behavior of the compliant layer) and of the grasp (due to the conformability to the object), and also enhances the reliability of the underlying (mechanical and elec-tronic) structure.

Moreover the whole articulated structure is particu-larly suitable to be realized by a single piece (e.g. a moulded plastic item) composed of rigid and elastic parts (hinges); this structure greatly simplifies man-ufacturing and assembly operations, reduces costs, improves reliability. Several morphological solutions have been defined and evaluated, according to a sys-tematic design approach [9], and a satisfactory design has been finally reached and implemented.

At present, only three parallel joints have been imple-mented: the adduction-abduction joint is not present. The proximal and the medial joints are independently actuated, while the distal joint is coupled to the move-ment of the medial joint. Joint actuation is provided by remote motors, and transmission is obtained with guided flexures, that in the present implementation are integrated with the finger structure in PTFE (see Fig. 2), but can be obtained in different material, e.g. high-strength steel. Optimization of hinges, as to topological definition, material choice, correct

(3)

siz-ing in order to obtain acceptable lifetime, is one of the major problems of structural design. In particu-lar, low bending stiﬀness is desirable but, at the same time, torque and compression loads should not deter-mine excessive strain. The present design is the evolu-tion of early sizing attempts, but could be changed by the structural optimization analysis still in progress.

V. Interactions between mechanics, electronics and control

The optimization process of the mechanical struc-ture, and in particular of the hinges, make use of theo-retical considerations and experimental tests [9]. But, as stated in Sec.III and shown in Fig. 3 this is only a part of the overall project. Side by side with the me-chanical implementation of the basic idea, the study of a suitable sensing apparatus and control strategy has been proceeding. Obviously, this research activity takes into account the main features of the proposed mechanical design, as well as the specifications tied to the desired target, that is a robotic hand able to interact with environment and performdexterous ma-nipulation. Conversely, the achieved results, in the both fields of electronics and control, have strong ef-fects on the mechanical structure, which, because of its simple manufacture process (see Fig. 2), is intrin-sically suitable for rapid prototyping and can be easily and rapidly modified.

A. From experimental test...

In order to validate the mechanical design of the ﬁn-ger and infer useful observations about its properties, a laboratory activity has been developed. A setup, shown in Fig. 4.a, has been built in order to actuate

Electronic design Mechanical design Control design Experimental tests Experimental tests Simulations Experimental tests + + +

Fig. 3. Approach followed in the development of the new

robotic hand. tendons displacements nib brushless motors + ball screws (a) x x Initial points Desired trajectory (b)

Fig. 4. Experimental setup for the endoskeleton (a) and trace of the desired motion (b).

the finger and to impose controlled linear displace-ments on flexures. Through a nib placed on the finger-tip, the trajectory in the workspace can be recorded. The main target of this early activities has been to verify kinematic properties of the finger, comparing the desired trajectory (computed on the basis of the “traditional” kinematic relation for a 3-dof planar ma-nipulator with revolute joints) with the real one. In particular, the same trajectory has been repeated several times, with different initial positions: although the tracking errors are not negligible, it is worth to no-tice that the behavior of the finger is perfectly repeti-tive.

Currently, the ﬁnger is tested without any soft layer, but tests tending to choose (and characterize) the ma-terial for the external compliant shell have been also performed.

In particular a suitable material has been identiﬁed in Technogel, a polyurethane gel, which shows a soft-ness quite close to the human skin, a nonlinear vis-coelastic behavior (with a related dissipation that can contribute to stabilize of the grasp) and other promis-ing features (for a complete analysis see [8]).

B. ...to the control system

As stated in Sec. III, this project is an attempt to improve the integration of the structure and con-trol design. The target is not only the optimization of some parameters of the mechanical structure

(4)

ac-cording the features of the adopted control strategies

and vice-versa, but also to determine the most

suit-able mechanical solutions in order to realize a certain control (which is in turn inspired by the structure of the ﬁnger frame) [4].

In fact, a number of theoretical works have faced the problem of dexterous manipulation and control of robots interacting with the environment [10], [11], [12], but the results they show remain often unapplied (or with applications limited to simple test setups pur-posely developed, e.g. 1-dof manipulators), because of the gap between the assumed hypothesis and the real conditions. The target is to reduce this diﬀerence. In particular the presented mechanical design seems particularly suitable to be controlled by some “kind” of impedance controller based on passivity concept. This broad class of controllers can be represented by the well-known control law [13]:

Fa=−J(q)T[K(X(q)−X0)−J(q)TB(J(q)˙q)] (1) where Fa is the vector of forces exerted by the ac-tuators, q are the generalized coordinates of the sys-tem (in the speciﬁc case the length of the tendons),

X(·) and J(·) are the forward kinematic and the

Ja-cobian of the ﬁnger, K(·) and B(·) represent the

force/displacement and force/velocity relations,X0is the vector of equilibrium positions.

By exploiting the properties of passivity [11] of (1), or of other kinds of impedance controllers, e.g. the IPC [10] (which speciﬁcally addresses the issue of dexterous manipulation), it is possible to show that the system is robust towards two classes of crucial nonidealities and side-eﬀects which characterize the mechanical struc-ture introduced in the previous section, namely:

• large errors in ﬁnger kinematic equations;

• unmodeled interface dynamics between the ﬁnger and the environment (e.g. soft pads).

Moreover, as clearly shown by eq. (1), the controller does not need force feedbacks, and therefore, in a min-imalapproach, the use of force sensors, which are usu-ally cause of high complexity and low reliability, can be avoided.

On the other hand, the implementation of such a con-troller requires the back-drivability of the actuation chain, and in particular its performances are strictly depending on the friction. For this reason this con-trol strategies is often unapplicable in robotic hands. As a matter of fact the mechanisms that allow high reduction ratio (e.g. harmonic drive) and that are often necessary in order to provide suﬃcient forces by means of small (concerning sizes and therefore supplied forces/torques) actuators, likewise yield high

107 mm 23 mm Position sensor Stator windings Slider -160 2 4 6 8 10 12 14 16 18 20 -14 -12 -10 -8 -6 -4 -2 0 2 time (s) (N) Force setpoint Measured force (a) (b)

Fig. 5. Sketch of the linear motor used to actuate the finger

(a) and its behavior as force source (b).

friction levels.

From the last consideration, it follows the need of lim-iting the friction as much as possible, both in the kine-matic chain and in the actuation system. As the trans-formation of a rotational motion into a linear one nor-mally produces such eﬀects and often requires compli-cated and/or not-back-drivable mechanical solutions, linear motors have been chosen for the actuation. This kind of technology, not yet extensively adopted in robotics, allows a simple and direct connection with the ﬂexures. In this way, it is possible to have back-drivable transmission chain and with limited frictions. Moreover, in the motors used for our application (pro-duced by LinMot [15]) the position sensor is inte-grated in the structure, that is a simple tubular stator with a moveable slider, see Fig. 5.a. Since, in this approach, the motors are used as an ideal source of force, some experiments have been performed in or-der to test their capabilities (with the standard con-trol boards which implement the basic current/force controller) to supply desired linear forces. Therefore, some force set-points have been provided to the motor control board, and the real force measured, by means of a (strain-gauges based) load cell directly installed on the motor slider. The results are reported in Fig. 5.b, in the static case (application of a step set-point) as well as in dynamic conditions (sinusoidal set-point). In the worst case the absolute error, due to frictional phenomena, is about 2N. If better performances (in terms of precision of the applied forces) are required, the load cells can be used for control purposes, but this is not problematic because they can be easily in-tegrated in the forearm and they are co-located [14] (that is directly installed on the actuators).

C. ...to the sensing apparatus

In a minimalism approach, impedance controller could perform interaction tasks and manipulation of objects without any force information, but based only on motor position sensors. Nevertheless, in order to

(5)

Tactile sensors

Position joint sensors

Fig. 6. Sensory equipment of the finger.

enhance the capability of the robotic end-eﬀector the adoption of a suitable set of sensors (including posi-tion sensors on the joints and distributed tactile sen-sors, as shown in Fig. 6) seems necessary [2]. In par-ticular, additional sensors must be used to eliminate position errors, mainly due to bending of tendons and undesired deformations of hinges, to compensate for inevitable friction phenomena and to allow ﬁne ma-nipulation tasks, which involve small forces. Moreover position, and above all force and tactile sensors can be used to properly plan the manipulation operations. C.1 Position sensors

In order to know the relative positions between the links of the finger, a measure based only on tendon lengths is not sufficient. On the other hand, the pe-culiar structure of the joints (without a fixed rota-tion center) makes it difficult to find a suitable posi-tion sensor. In fact, the standard robotic technologies (e.g. potentiometers or hall-effect based sensors) re-quire well defined paths.

A possible solution could be a special purpose sen-sor (see Fig. 7.a), built directly on the hinges of the endoskeleton. This solution is based on strain gauges, glued on the deformable structure. Nevertheless, in this manner only a partial compensation of kinematic errors is possible, such as those produced by tendons bending, but not of errors directly imputable to the hinges.

An alternative solution, currently tested, is based on ﬂex-sensors, Fig. 7.b, usually used in data gloves to

Hinge Strain gauge

(a) (b)

Fig. 7. Joint position sensors: special purpose sensor (a), com-mercial flex sensor (b).

measure the bending of human fingers. These sensors are based on piezoresistive effect, and provide a vari-ation of resistance proportional to the bending angle. In this case, the position information can compensate for all the kinematic errors, because the sensor is phys-ically separated from the finger structure.

Both solutions appear suitable according the the cri-teria of simplicity and reliability that characterize the project.

Alternatively, in order to keep the ﬁnger structure (and hence the overall end-eﬀector) as simple as pos-sible, a control based on visual servoing can be im-plemented (and also this possibility is currently eval-uated).

C.2 Force sensors

In order to exert very small forces on the environ-ment, an implicit force control based on motor cur-rent(/force) control appears not to be an optimal solu-tion. Additional force/tactile sensors seems necessary, also in the task planning phase, in order to have a suf-ﬁciently precise estimate of exchanged forces and to known the exact location of the contact points. Two are the possible solutions:

• tactile array sensors [2];

• intrinsic tactile sensors [2].

The former set includes a number of different devices, able to detect the amplitude of the normal forces ex-erted on their surface (usually planar) and the con-tact shape. They can have different spatial resolu-tions and different sensitivities to the applied forces according their degree of complexity. For the present application, inspired by criteria of simplicity and re-liability, the use of a commercial single element pres-sure sensors has been investigated. In particular, the FlexiForcemanufactured by Tekscan [16], has been tested in order to verify the suitability of its integra-tion in the finger design. Therefore, the focus has been the study of the interactions between the sensor and the other elements of the robotic finger, in particular the polyurethane gel for the pads. As the sensors must be placed under this soft cover, a setup to measure how the force is transmitted to the sensor through the gel has been designed. By means of this setup (which exploit a load cell to measure the applied forces) the response of the force sensor to a known force has been recorded, with and without a layer of gel.

As shown in Fig. 8, where the achieved results are reported, the presence of the gel layer (3 mm thick) reduces the sensitivity of the sensor, and above makes the dynamic of the sensor very slow. The reason of

(6)

0 1 2 3 4 5 6 7 8 9 -5 0 5 10 Time (s)

Load cell response (N)

0 1 2 3 4 5 6 7 8 9 -5 0 5 Time (s) Sensor response (V) without gel with gel

Fig. 8. Response of the pressure sensor to the application of a force of 10 N with and without soft cover.

this behavior is due to the polyurethane gel which, af-ter the initial application of the exaf-ternal force, changes its configuration, distributing the external pressure on the overall underlying surface (also outside the sensi-tive area of the sensor). This filtering effect of the Technogel makes the use of this sensor for control purposes very difficult.

A solution which seems more suitable for the proposed finger are intrinsic tactile sensors, directly built on the finger frame. As a matter of fact this kind of sensor, able to detect the magnitude of the applied force/torque and the position of the contact centroid, provides a measure of the forces/torques resultant, and therefore it is insensitive to the above mentioned effects of gel; nevertheless a deeper analysis seems nec-essary, in order to test the interactions between the gel layer and an IT sensors.

VI. Conclusion and future work

In this paper the design of a innovative robotic fin-ger has been reported. The basic idea is the adoption of so the called “compliant mechanisms” in the field of robotic hands, which appears still too tied to tra-ditional criteria of robot design. Because of the lack of suitable technological solutions (e.g. artificial mus-cles), these criteria lead to complex, costly and not enough reliable structures. Conversely, the proposed design shows attractive characteristics, such as a small size, which allows the adoption of a visco-elastic cover (similar to the human hands) and an easy integration with the sensory system, and above all a simplicity, also in the production process, that could lead to a wider diffusion of robot hands.

The ﬁnal target is the development of an anthropo-morphic robot hand with 4 upper ﬁngers and an op-posable thumb (whose sketch is depicted in Fig. 9).

Fig. 9. Design of a robotic hand based on compliant

mecha-nisms (UB-Hand III).

References

[1] L. Biagiotti, F. Lotti, C. Melchiorri, G. Vassura, “Design

Aspects for Advanced Robot Hands”, in IROS02 Tuto-rial: Towards Intelligent Robotic Manipulation, Lausanne, September, 2002.

[2] C. Melchiorri, “Tactile Sensing for Robotic Manipulation”,

inArticulated and Mobile Robotics for SErvices and TEch-nologies (RAMSETE), A. Bicchi, S. Nicosia, B. Siciliano, P. Valigi (Eds.), Springer Verlag, 2001.

[3] J. van Amerongen, “The Role of Control in Mechatronics”,

Engineering Science and Education Journal , Vol. 9, No. 3, June 2000.

[4] Q.Li, W.J. Zhang and L. Chen, “Design for Control

-A Concurrent Engineering -Approach fo Mechatronic Sys-tem Design”, IEEE/ASME Trans. on Mechatronics, Vol 6, No.2, June 2001.

[5] L.L. Howell, “Compliant mechanisms”, John Wiley &

Sons, 2001.

[6] J. M. Goldfarb and N. Celanovic , “A flexure-based gripper

for small-scale manipulation”, Robotica (1999), Volume 17, pp.181-187.

[7] K.B. Shimoga and A.A. Goldenberg, “Soft Materials for

Robotic Fingers”, in Proc. IEEE International Conference on Robotics and Automation, ICRA92, Nice, France, 1992.

[8] P. Tiezzi, F. Lotti, G. Vassura , “Polyurethane Gel Pulps

for Robotic Fingers”, submitted to International Confer-ence on Advanced Robotics, ICAR03, 2003.

[9] F. Lotti and G. Vassura , “A Novel Approach to

Mechan-ical Design of Articulated Fingers for Robotic Hands”, in Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and Sys-tems ’02, IROS 02, Lausanne, Switzerland, 2002.

[10] S. Stramigioli, C. Melchiorri, S. Andreotti, “A passivity-based control scheme for robotic grasping and manipula-tion”, in Proc. of the 38th IEEE Conference on Decision and Control, 1999.

[11] N. Hogan, “On the Stability of Manipulators Performing Contact Tasks”, in IEEE Journal of Robotics and Automa-tion, Vol. 4, No. 6, December 1988.

[12] A. Bicchi and V. Kumar, “Robotic Grasping and

Manip-ulation”, inArticulated and Mobile Robotics for SErvices

and TEchnologies (RAMSETE), A. Bicchi, S. Nicosia, B. Siciliano, P. Valigi (Eds.), Springer Verlag, 2001.

[13] N. Hogan, “Impedance Control: An approach to manipu-lation, Parts I-III”, in ASME Journal of Dynamic Systems, Measurement and Control, Vol. 107, pp. 1-24, 1985. [14] S.D. Eppinger and W.P. Seering, “Three Dynamic

Prob-lems in Robot Force Control”, in IEEE Transactions on Robotics and Automation , Vol. 8, No. 6, December 1992.

[15] url: www.linmot.com.