Three articulated fingers of the BarrettHand spread   and conform to various shapes.

Three articulated fingers of the BarrettHand spread and conform to various shapes.


The TorqueSwitch mechanism in cutaway

The TorqueSwitch mechanism in cutaway


BarrettHand's eight articulating joints can grasp objects   of various sizes in almost any orientation.

BarrettHand's eight articulating joints can grasp objects of various sizes in almost any orientation.


Repeatedly securing a bicycle seat dangling from a thread is nearly an impossible task for a robot gripper. Its fixedjaw geometry can't readily adapt to the seat's ever-changing orientation. No problem for a robotic grasper. Graspers with articulating, humanlike fingers can grasp objects in almost any orientation and securely hold them. And although the above task is exceptionally difficult, it underscores a major shortcoming of conventional grippers: lack of flexibility. To boost quick-change flexibility, the robotics industry took the automatic tool-exchange technique used for NC-mill cutting tools and applied it to robot grippers. But the approach is proving to be both costly and ineffective.

Unlike off-the-shelf cutting tools, grippers have customized jaws. The trial-and-error process used to make them is both labor intensive and difficult to scope in advance. Costs tend to balloon because each anticipated variation in part shape, orientation, and robot approach angle often requires another custom gripper, a place to store it, and an exchange mechanism. Furthermore, robot arms typically program from the base to the toolplate (holds custom gripper part) only, a few centimeters short of the workpiece itself. If robots are to fully realize their potential, all movements from base to target should be programmable.

Graspers allow just that. Graspers equipped with embedded software quickly and automatically conform to almost any part in any orientation. Graspers hold parts more securely, are quicker to install, and tend to cost less than banks of custom-machined grippers and associated hardware. Grippers also require bulky toolchangers at robotic-arm ends that limit useful payload and dynamic response. Graspers, in contrast, eliminate toolchangers and the added weight. They also eliminate custom hardware which simplifies workcell operations. Graspers are physically identical with all customization provided instantly in software.

Graspers and grippers
Despite all the advantages of graspers, most robotic part handling and assembly today takes place with grippers whose basic design has changed little over three decades. Vacuum cups and electro-magnets are also used for handling such items as automobile wind-shields and body panels. Payload sizes range from a few grams for tiny pneumatic grippers to over 100 kg for massive hydraulic models. Both types use simple on/off valves to switch between full-open and full-close states.

Grippers act as simple pincers with two or three nonarticulated fingers called jaws, which either pivot or remain parallel during open-close motions of about 1 cm. The jaw part that touches the target is removable and made of soft steel or aluminum. Tool designers shape these rectangular soft-jaw pieces for each job. It can take several iterations of shaping and adjustments to get proper function.

Compare this with the Barrett-Hand from Barrett Technology Inc., Cambridge, Mass. Here, four brushless-dc servomotors power eight joint axes through special mechanical couplings.

The device's three articulating fingers and palm act in concert to trap target objects firmly and securely within a grasp consisting of seven coordinated contact vectors — one from the palm plate and one from each link of each finger. The three fingers move independently, powered by one of three servomotors. Except for the spread action of fingers F1 and F2 (driven by the fourth servomotor), all fingers have inner and outer articulated links with identical mechanical structure. Each of the three "finger" motors drive two joint axes. Torque channels to these joints through a patented TorqueSwitch mechanism whose function is optimized for maximum grasp security.

It works like this: When a fingertip, not an inner link, touches an object, the system locks both joints, switches off the motors, and awaits further instructions from the hand-mounted microprocessors or external controller. When an inner link touches an object for a secure grasp, the TorqueSwitch reaches a preset threshold torque, locks that joint against the object with a shallow-pitch worm drive, then redirects all torque to the fingertip to make a second, enclosing grasp within milliseconds of the first.

After the grasper releases an object, the TorqueSwitch sets threshold torque for each finger in anticipation of the next grasp. This thresholding takes place by opening each finger against its mechanical stop with a controlled torque. The need for greater opening torque triggers higher subsequent threshold torque. In this way, graspers can accommodate a wide range of objects from delicate, to compliant, to heavy. The finger articulations allow each digit to conform to the object surface using two independent contact points per finger. Position, velocity, acceleration, and torque are all controllable over 17,500 encoder positions. Each finger can travel full range in either direction in less than one second and produce a 2-kg force measured at the fingertip. Once a grasp is secure, the links automatically lock in place which allows motors to switch off, conserving power until commanded to readjust or release their grasp.

The inner and outer joints close in a 3:4 ratio with respect to the robot toolplate until the inner link strikes an obstacle and activates the TorqueSwitch. Although the inner and outer finger-link motions curl anthropomorphically, the spread motion more closely resembles a primate's opposable (thumb) finger. But instead of one opposable finger, the BarrettHand has twin, symmetrically opposable fingers centered on parallel joint axes that rotate 180° about the entire palm. The arrangement forms a nearly limitless variety of grip-per shapes and fixture functions. Spread is controllable to any of 3,000 positions in either direction within a half second. Unlike the mechanically lockable finger-curl motions, spread motion is fully reversible, so its servos actively control stiffness as well as position, velocity, acceleration, and torque. This compliant spread motion lets fingers both close around an object for maximum stability and find their lowest respective energy states, even on highly concave surface features.

Dexterity defined
Truly dexterous robots would require independent, intelligent motor control over each and every articulated joint axis. In other words, at least n independent servomotors, and sometimes as many as n + 1 or 2n motors, drive n joint axes. Although this strict definition may be mathematically elegant, it leads to impractical designs. A dexterous BarretHand would require between 8 and 16 motors, making it far too bulky, complex, and unreliable to be practical. But, the use of four intelligent, joint-coupling mechanisms approaches dexterous action with only four servomotors. Interestingly, most human hands aren't dexterous either. For example, try moving the outer joint of your index finger without moving the adjacent joint on the same finger. Most people can't move these joints independently because human hands are optimized for grasping.

Control electronics
The 1.18-kg grasper is completely self-contained with only an 8-mm-diameter umbilical supplying dc power and serial communications to an external controller. Communications electronics, five microprocessors, sensors, signal processing electronics, electronic commutation, current amplifiers, and servomotors, all reside in the grasper palm body. A central supervisory microprocessor coordinates four dedicated motion-control microprocessors. The control electronics are built on a parallel 70-pin backplane bus. Associated with each motion-control microprocessor are the related sensor electronics, motor commutation electronics, and motor-power current-amplifier electronics for that finger or spread action. The supervisory microprocessor directs I/O communication via a high-speed, industry-standard RS-232 serial port to the workcell PC or controller.

An open source grasper communications language called GSL optimizes communications speed by exploiting the difference between bandwidth and time-of-flight latency. Graspers generally remain inactive during most of a cycle — as the arm performs gross motions and operate only for short bursts at the end of a trajectory. Therefore grasper on-board electronics are able to receive large amounts of setup data during an approach. Upon arrival, the workcell controller issues a command for the grasper to close, which it does within a couple milliseconds.

Grasper control language (GCL)
The Grasper Control Language (GCL) lets the grasper communicate and accept commands through the serial port from a robot-work-cell controller, a PC, Mac, or Unix-based computer, or a Palmpilot. Executing and acknowledging a command from the workcell controller to the grasper and then back again takes only a few milliseconds.

There are two control modes: supervisory and real time. The supervisory mode has a simple prefix-command structure. The prefix refers to motors 1 through 4 corresponding to the three fingers and the spread motion. Any number of prefixes may be used in any order. For example, the command "12C" closes fingers Fl and F2. Commands with no prefix activate all available axes. Similar commands open and close fingers incrementally to user-defined values and interpret signals from optional finger-mounted strain gages.

The real-time mode is accessed through a Windows-based GUI and reserved for complex tasks such as real-time teleoperation. Users specify a packet structure in supervisory mode and the software outputs a histogram of 20 successive time-of-flight tests. The information helps users balance information content with latency. The GUI also includes a pictorial of the grasper with radio buttons for position and rate control. A "Generate C++ Code" button saves commands in C++ for later recall without programming.

Target markets
The BarrettHand BH8-250 was introduced commercially in 1999. Thirty units have been sold thus far, mostly to Japanese automotive manufacturers and suppliers, including Honda, Yamaha Motorcycles, and NGK. These companies are now exploring the capabilities of the device, while others, such as Fanuc Robotics and the U.S. and Japanese space programs, have become repeat customers.

Information for this article was provided by Clive Loughlin and William Townsend of Barrret Technologies Inc. For more information, visit these Web sites: www.barrettrobotics.com and www.emeraldlibrary.com, publishers of Industrial Robot: An International Journal