A Flexible Three Fingered Robotic Hand - Online Article


This paper addresses the design and implementation aspects of a flexible Three-Fingered Robotic Hand. The goal of designing 3-fingered robotic hand is to have minimal computation control approach for grasping and releasing an object of an arbitrary topology. The robot system interfaces to Graspar hand using grasp command and release command. The grasp command has one input parameter that specifies the type of grasp desired implementation of two types of grasps; a fingertip grasp and enclosing grasp are discussed. The hand design based on connected double revolute joint mechanism with release springs for an inexpensive, mechanically simple, and easy to control is outlined. Finally, some of the experimental results are discussed.


The field of robotics involves various topics such as control theory, operating systems, geometric transforms, motors, and signal processing. It can be studied as an application area for achieving multiple objectives such as improvements in volumes of production, improvement in quality of products, flexibility to change to different items of production, cutting in manpower and costs, applications in hazardous areas (nuclear environments, underwater situations, space application, certain defense requirements, handling of explosives, construction and maintenance of tall structures, fire fighting). Robots can be classified according to their method of control pathway, structural design, or level of technology. The two major types of control are servo and non-servo. The path way of the robot may be either point to point or continuous. The volume of the work space of a robot is either rectangular, cylindrical, spherical, or jointed spherical. Furthermore, a robot may be classified as either low-, medium- or high-tech, based on its number of axes, and effectors and grippers.

The shape of the gripper is determined by the task it has to perform. Two-pronged or general grippers are used to pick up cylindrical and cube objects, three-pronged grippers are used to pick up cylindrical and cube objects, three-pronged grippers are used on spherical objects, and specialized grippers can be designed for unique tasks. Some grippers are equipped with a sense of touch so that they can determine when an object has been grasped. Object grasping techniques which do not need the computation of three-dimensional surface information are important for robotics applications where computational power may be limited, for example on mobile delivery and retrieval robots. In the Grasper Robotic Hand (GRH) project at Andhra University, the aim is to determine a simple reflexive grasp that can be utilized for a wide variety of objects. The GRH is designed based on non-servo, point-to-point, and cylindrical robot structure with three-pronged gripper(three fingers). This approach is deviated from other researchers in that focusing is primarily on the task of grasping objects and not that of manipulating or assembling objects. This type of grasping device has a variety of applications in object retrieval systems for the handicapped, planetary and underwater exploration. To fulfill the objective, a general-purpose 3-fingered robotic hand is designed for grasping a wide variety of objects with possibly complex shapes, adjusting the grasp if unexpected external forces cause the object to slip during grasping, and a simple control scheme is used with two commands: grasp and release.

The methodology adopted is a simplified anthropomorphic design with three fingers and an opposing thumb. Each finger has three links, or phalanges, and three double revolute joints, almost similar in structure to the human finger. The thumb has two links and two double revolute joints. Each finger is actuated by three individual strings corresponding to three joints of the finger whereas the thumb finger is actuated by two individual strings corresponding to its two joints. One end of each string is connected to one joint of the finger and other end if each string is connected to one pulley placed over the shaft driven by a single motor. The three strings of each finger curls the finger towards the palm and the other two strings of opposing thumb finger extends the thumb finger away from the palm so as to grasp the object.

The paper deals with the design and implementation of three-fingered robotic hand. Currently developed hands are addressed first, and slow the motivation for the work. Next, the functional block diagram and description of total robot hand system is presented followed by mechanical structure of the hand and the antagonistic tendoning system. Then, how this structure moves and complies to the object surface is shown. Also, discussed how to choose the finger locking structure for tendoning system by maximizing the workspace of the fingers while mechanically insuring that the hand cannot collide with itself. The control algorithm which uses simple solenoids activation and deactivation is presented next. Finally, the results of the implementation and experimentation and present ideas for future work are outlined.

Design Structures for Robot hand

The parallel-jaw gripper is the simplest possible general-purpose robotic end-effector. This mechanism consists of two parallel structures and the robot can control the distance between the two jaws. Many researchers have outlined algorithms for grasping objects of various shapes with these grippers [Pertin-Troccaz, Mason]. Some other researchers have designed general purpose end-effectors by using the human hand as inspiration. In many of these configurations, the robot hand has the same kinematic structure as a human hand, and can independently control the joints of each finger. Examples for such robot hands are the Utah/MIT hand [Jacobson] and the Salisbury hand [Salisbury]. In most cases, the fingers are actuated through antagonistic tendons routing to a motor for each finger joints. The difficulty with these flexible mechanisms is that one must compute a desired position or torque for each joint of each finger. A four fingered hand with three joints on each finger would require twelve desired positions or torques to be computed for control. This flexibility is needed for manipulation by having independent control of finger joints. Rensselear Polytechnic Institute has also implemented a robotic grasping device where each finger ahs two controllable degrees of freedom, requires a total of 6 motors to drive a three-fingered hand [Zink]. This hand also uses an antagonistic tendoning system, but it was more controllable degrees of freedom that must be individually controlled.

A five-digit hand is designed at University of Belgrade which uses three motors; one to control the thumb and the other two to control two fingers each [Rakic]. The thumb can rotate from a position aligned with the four fingers to opposing positions with three of the four fingers. The other motors control two fingers which use a differential lever to rotate about the finger joints. A differential lever allows the motions of the two "connected" fingers to adapt to the differential forces acting on the finger pair. The design is particularly flexible in its ability to grasp an object in a variety of configurations.

An unique grasping device is designed by the University of Pennsylvania, in that it has a stationarythumb and two rotating fingers that can either be positioned in line with the thumb for hook grasps or on the opposite side of the thumb for enclosing grasps [Ulrich]. The hand also uses a clutch mechanism to control the fingers during grasping of the objects.

The graspar hand is a simplified anthromorphic grasping device which has one motor per finger. Graspar's fingers naturally comply to the surface of the object thereby minimizing the computation typically needed to compute surface geometry. The authors feel that this mechanism is simpler than others general purpose grasping devices and that it provides an alternative to the other available mechanisms. Table 1 compares several hands that have been designed and built. The second column describes the type of the actuators that are used to transmit the motion from the motors to the joints of the hand. The other columns give the weight of the objects that the hand can support, the number of fingers, the controllable degrees-of-freedom of each finger, and the total number of actuators in each hand.

Table 1. Practical Robot Hands

Hand Actuator Weight #Fingers #Degree of Freedom #Actuators
Belgrade DC Servo hooked to gears - 3 1 3
MIT/Utah Air valves via cable & tendons 4.5 Kg. 4 4 32
Stanford/JPL DC Motor via cables - 3 3 8
Anthrobot Tendon driven - 5 4 20
Upenn/Barett Clutch Mechanism - 3 2 6
Graspar Robot DC Stepper Motors 1.0 Kg. 3 3 3
Hand via Rollers & Solenoids

Robotic Hand Mechanism

The functional block diagram of a PC-based Three-Fingered Robotic Hand Systems is shown in fig. 1. The Robot system is interfaced to PC via General Purpose Prototype Card (GPPC) with I/O interface, Buffers, Base Stepper Motor Driver for moving arm up to maximum of 270 degrees, 3 fingers stepper motor drivers with solenoids placed on arm for motion control of joints of fingers with pitch of 45 degrees maximum, 2 wrist stepper motor driver with solenoids for motion control of wrist with roll and pitch of maximum 180 degrees, and A/D converter with inputs from pressure sensors placed on fingers for feedback of fingers position. The mechanical structure in this laboratory robot arm model consists of base structure, arm structure and wrist structure that is hooked to three-fingered graspar robotic hand. The operation of PC based robot arm is achieved through control software written in C. Two higher level operations namely Grasp and Release are incorporated into this.


Mechanical Structures

The design of mechanical structures for GRH (Graspar Robotic Hand) incorporates three digits: Two Fingers and one Thumb as shown in figure 2. Three digits are positioned at the corners of an inverted triangle since this geometry leads naturally to stable finger contact positions for an enclosing grasp [Mason]. Each finger consists of three rigid links (the proximal, intermediate and distal phalanges) constructed from cylindrical pipes. The phalanges are connected by three joints (the proximal, intermediate and distal phalanges) which have parallel axes of rotation and are responsible for curling the finger tip toward the palm. The thumb is similarly configured except that it consists of only two rigid links (the proximal and distal) connected by joints (proximal and distal). The dimension of GRH was selected to be approximately the size of an adult human male hand. The phalange lengths L1, L2 and L3 phalange width w, and the distance between the finger and the thumb, dft, were selected as shown in table 2.

Table 2: GRH Parameters

Description Variable VALUE (c.m.)
Distance from palm to proximal joint L0 4
Distance from proximal to intermediate joint L1 2.8
Distance from intermediate to distal joint L2 2.8
Distance from distal joint to digit tip L3 2.8
Distance from end to start of phalange L 0.7
Each phalange width w 1.2
Distance between fingers dff 3.3
Distance between thumb and fingers dft 9
Support thickness t 0.1
Each double revolute joint width rj 1.6

The distance between the two fingers, dff , was computed to be the distance between the fourth and index finger to give the largest human finger displacement. The thickness of the supports, t, was chosen for availability and strength. To reduce construction costs, two fingers are configured identically and the thumb to have identical link lengths as the proximal and intermediate links of the fingers. Each digit is controlled by three individual strings which are routed through phalanges as shown in Fig. 3.


The double revolute joints through strings/tendons act as routers to create angular displacement of the links, but do not transfer any torque to the joints. Each phalange to phalange is connected by a double revolute joint. The spring plates brazed on top of each phalange control the digit in opposite direction when the string is pulled in opposite direction. Each of the tendons through this system and each of the spring plates work as a differential mechanism. One end of each tendon is attached to the lower end of each phalange and the other is wound about a pulley attached to the shaft with two rollers (pulley roller and brake roller) attached at its near ends. The pulley roller gets motion via a intermediate roller located in between pulley roller and main roller fixed on a shaft that is attached to a single remotely-located DC stepper motor. The intermediate roller is attached through vertical plate to solenoid that controls motion transmission from main roller to pulley roller. Therefore, the finger joint moves when the pulley roller gets motion. The pulley roller moves when the intermediate roller upon solenoid's activation is intact with pulley roller and main roller which is driven by the stepper motor. The brake roller/drum is located between brake plates structure that is attached to solenoid. Applying brake plates to brake drum to lock finger joint motion is controlled by the solenoid. The strings in each finger are wound in both directions depending on the stepper motor direction. In this configuration, the motors control the relative length of the tendons when irregular object is selected. The complete mechanical structure of one finger's joint movement and lock control mechanism is shown in fig 5.

For practical purpose, aluminium is selected as the material for the pipes (fingers) since it is strong, rigid, lightweight, relatively inexpensive, and easy to machine. Aluminium is also used for the arm, wrist and palm, since it is not as susceptible to wear. Dial cords with small diameter plastic tubes as sleeves are used for the tendons of the digit since these are flexible.

Finger Movement

The Graspar robotic hand differs from many other hands [Crisman] in that the joints of each finger are independently controlled for motion. Since the joint axes of each digit are parallel, the motion of each finger lies on a plane. The digits are mounted such that their planar workspaces are parallel and therefore does not collide. The motion of a Graspar finger as the digit's driving motor is activated, is shown in fig 4.

The digit's tendons are wound in directions on the motor's direction so that the motor actively controls the relative length of the tendons. The tendons are of equal length when the digit extends straight from the palm, shown as the B position in the fig 4. When the grasp tendon is shortened, the finger will move sequentially from positions B to E, assuming no object is present in the workspace. Notice that the proximal joint will turn its maximum angle before the intermediate joint begins to rotate. The maximum angle of each joint is reached when the grasp tendon is tangential to the double revolute joint. Similarly, the intermediate joint will reach its maximal extent before the distal joint begins to move.


The sequential as well as independent motion of the joints of each finger is possible due to the independent control is incorporated in the design. When a tendon moves, the joint with the largest moment (in this case the proximal joint) will move first. In similar fashion, the graspar thumb moves to the fingers. If an object is within the workspace of a finger, the proximal joint will have the largest moment until it reaches it maximum angle or until it comes in contact with the object. The designed dimensions of robot hand are similar to that of human hand thereby specified many of the geometric design parameters and designed the double revolute joints so that the finger will not collide with the palm. Thus the work space volume from the zero configuration (shown as B in fig 4) to the palm (Position E) is maximized. From the two-dimensional kinematics of the finger, the perpendicular distance from the palm to the finger tip when the joints are curled to their maximum extents is computed. This distance is computed to be:

d = L3 cos(θ1max + θ2max + θ3max) + L2 cos(θ1max + θ2max) + L1 cos(θ1max) + L0

The maximum joint angles are determined from the configuration of finger's double revolute joints. Notice that the grasp is its shortest, and thereby the joint angles in their maximum position, when the string is tangent to the joint. The volume of the work space of the finger is a function of the maximum motion of the joint angles as can be seen from fig.4. As a finger moves from the zero configuration B to the maximum extent of the proximal joint C, it spans a volume of VBC1max(L1+L2+L3)2. Similarly as the finger continues moving through positions D and E, it will span the volumes VCD=θ2max(L2+L3)2 and VDE3maxL32 respectively. Therefore the total volume of the work space of the finger is sum of the partial volumes, i.e.:

V = θ1max(L1+L2+L3)2 + θ2max(L2+L3)2 + θ3max L32

Considering the size of designed double revolute joints in this project that were inexpensively fabricated, the work space volume is maximized while insuring the finger tip will not touch the palm (i.e. d>0).

Control Algorithm

A three fingered hand has been designed built and tested. The hand design is based on connected differential mechanisms and has been designed to be inexpensive, mechanically simple and easy to control. The tendoning system of the differential mechanism provides the hand with the ability to confirm to object topology and therefore providing the advantage of using a simple control algorithm. The achievement of this hand is to demonstrate that reliable grasping can be achieved with inexpensive mechanisms and sensors. The authors have demonstrated that this hand can grasp a variety of objects with different surface characteristics and shapes without having to reconstruct a surface description of the object. This grasp is successfully completed as long as the object is within the workspace of the hand.

The control algorithm used by Graspar is extremely simple as was the goal of the design. The controller takes a high- level command from the user (grasp or release) and products a successful grasp of the object that is within the workspace of the hand. Unlike most robot hands which use joint position measurements for control feedback, it uses inexpensive simple contact sensors. These sensors are being used primarily due to cost considerations, but have certain limitations. The advantage of designed robot hand is its simplicity and inexpensive cost. However, this hand does have some limitations. It cannot be controlled to perform fine manipulation like writing. The Graspar robotic hand was designed as a first step in a future project that will integrate vision and grasping to control a robot to grasp objects and the future work will use a simple stereo vision system to determine the topology of the desired objects and to visually move the robot arm so that the desired object is aligned within the work space of graspar. The simple control algorithm developed in this project will be applied to the future project in order to grasp an arbitrary object. As the robot hand system moves the contact switches/pressure sensors will be monitored to determine the successful grasping. This method is relatively simpler than the hugging algorithm presented for a robot hand configured as graspar which relies on dense contact sensing. The total robot hand system has independent control of finger joints of each finger leading to flexible control of finger motion as compared to three fingered joints of each finger leading to flexible control of finger motion as compared to the three fingered graspar.

Feedback to the motors are provided by miniaturized pressure sensors/simple contact switches mounted on the grasping surface of each phalange. The GRH is currently using slightly modified contact switches mounted in parallel along the length of each phalange and added rubber sheathing under the lever of a generic limit switch and over the surface of a small push button switch to increase the amount of pressure needed to close the switch, to dampen bouncing often associated with switchers, to increase the area in which the contact can be sensed, and to provide additional friction between the fingertip and the object.

Similarly, contact switches/sensors are used for grasping. Each phalange has contact switches mounted on the grasping surface of the finger. When a grasp command is issued to Graspar, it shortens the grasp tendon of each finger until a certain contact pattern is noticed on the digits. To execute a fingertip grasp, Graspar shortens the grasp tendon until the contact is made on all of the phalanges. If at anytime the contact positions are lost, graspar then continues to shorten the grasp tendon of the digit until contact is reestablished. Each finger has a mechanical limit at the fully retracted position (A in fig 4.) and contact switches to determine when the hand is in this configuration. Therefore, when a release command is issued to the Graspar hand, it simply moves simultaneously the motors to shorten the release tendons until contact is made on all phalange mechanical stops.


The goal in designing Graspar is to have a simple, minimal computation control strategy. The robot system interfaces to graspar using two commands, grasp and release. It is assumed that when the grasp command is issued by the robot system, the manipulator has already been moved so that the object to be grasped is in the workspace of the robot hand. The grasp command has a single input parameter which specifies the type of grasp desired. At the current time, the authors implemented two types of grasps: a fingertip grasp and enclosing grasp.

About the Author:

No further information.


No comment yet. Be the first to post a comment.