Air valves are integrated into the Hand design. The muscles are below the hand, and the valves are in the manifold at the bottom.

The valve control boards do PID control, taking data from the sensors on the Hand and via CANbus. So you can easily send set-points to the system. The GPL software provided on the host PC makes this easy, and provides various interfaces you can use.

The whole Hand system weighs about 3.5kg, and the center-of-mass is in the forearm. The valves we are using are really nice solenoid valves from Lee (LHDA231111 or similar) which run very fast and provide excellent response - they're plumbed as close as possible to the muscle, to prevent significant delay.

We normally run PID controllers on the valve control boards, two-per-muscle (and hence 4-per-joint) which means we can get reasonable tracking of position or pressure setpoints.

At present, the Hand requires:

The Host PC is used primarily for configuration of the Hand, and could be embedded in the Hand if necessary.

Once the Hand has been configured, all sensor data is transmitted on the CANBUS and can be read by other system components.

Control of the Hand can be done in three ways:

Each muscle can be switched between these three modes independently.

The control is operated by sending messages on the CANBUS.

The normal setup for the Hand is to have 28V DC supplied for the valves (needs less than 1 Amp), 9V DC supplied for the electronics (again, less than 1 Amp), and compressed air supplied at 3.5 Bar. We supply power supplies with the Hand, but an air compressor must be sourced locally.

On the Hand, all necessary hardware (except the air compressor) is included. Basically, two air valves are used for each muscle, one to fill it and one to empty it. These have to be controlled from some electronics. Position sensing and air pressure sensing is normally necessary as well, in order to allow more precise control.

At present, the programming system allows posture control of the hand, and also (some degree of) control based on tactile response. Other controls can (fairly) easily be programmed in - the Hand hardware takes commands directly via CAN bus, and we provide a set of open-source software with a C and scripting API for use from Linux.

Setpoint tracking on the Hand is better than 1 degree accuracy, although because it is an antagonistic system there are several tuning "sweet spots", depending on whether speed of movement or precision of results are your goals.

Simplest possible method:

	#!/bin/bash
	sendupdate FFJ2_Target 45 FFJ3_Target 45 
	sleep 1
	sendupdate MFJ2_Target 0 MFJ3_Target 0 
	sleep 1

There is one _Target for each joint. FF, MF, RF, LF are the four fingers, with 4 joints tip to palm. LFJ5 is the extra palm joint. Thumb is THJ1...THJ5. Wrist is WRJ1 and WRJ2.

In all cases, the zero angle should correspond to a flat, relaxed human hand. Positive angles (typically) cause the hand to close.

When the Hand leaves us, we calibrate the joints at 5 points and the software interpolates linearly in-between.

We've designed the Shadow Hand for use in robotics research, as a means of giving robots the same capabilities as humans. Applying this hardware to humans wouldn't really work; we suggest you talk to their medical professionals about options.

That said, we're interested in applying the Hand "off" the human - on a mobile base under high-level control - to do tasks for a person without them needing to be in the same place as the task.

Yes. In 2002, we borrowed a cyberglove for a robotic exhibition in Japan. Here is the exact interface script we used to read data from the glove and feed it into the Hand system:

	#!/bin/bash
	# Drive a CyberGlove attached to ttyS0
	SERIAL=/dev/ttyS0
	echo "glove 18" > /proc/robot/shm_alloc
	/usr/lib/robot/bin/loadcal cyberglove.cal
	stty 115200 -parenb cs8 <> $SERIAL
	while true ; do
	echo -n g > $SERIAL
	read char values < $SERIAL
	/usr/lib/robot/bin/sendupdate glove 0 $values
	sleep 0.1
	done
      

We then had a C program running that read the "glove" sensor values, and mapped them into setpoints for the Hand. You can find it in our CVS.

We found at the time that the cyberglove output data was not a perfect match for the Hand: it was "similar" to the hand position but not exactly the same - if you wanted to do fine manipulation you would probably need to process the data in a much complex way than the program we wrote.

Currently it is designed as a research tool - so we have not gone through any approvals procedure to make it appropriate for this purpose! Please get in touch with hand@shadowrobot.com to discuss this further.

Certainly, there are people who purchase muscles and build systems from them using standard pneumatic parts. The Hand is designed as an integrated unit, and so wouldn't be the right starting point for that kind of re-assembly, but we supply necessary components. Look at the SPCU for the control of the system.

Note that we do build other types of robot hardware - ask us for a quote!

The Hand system itself weighs under 4kg. You will need a pneumatic supply, but that can be a long way away, just running a thin air-line to the Hand. The PC does not need to be on-board; we currently use a little Mini-ITX system, but a smaller embedded system could be used. The compressor weight depends on your noise budget; a "silent" (i.e. <38dBA) compressor will weigh 15kg, but a very noisy compressor might weigh 1-2 kg.

So the complete bundle will fit comfortably through a half-inch hole.

30kg is outside the operational range for the Hand. However, we currently regard the Hand system as a "lab prototype for research" rather than an actual industrial component, because every time we build one, it is as close to the state of the art as possible at the time of the build.

If you have a specific requirement for which we can agree on a standardised version of the Hand, then we would endeavour to meet your operational requirements for that.

We are producing a 4-DOF arm in muscle technology to match the Hand at present. If you're interested in getting one of these, then contact us.

One of our customers used a Mitsubishi PA-10 for their Hand - we provided a standard mountpoint on the end the Arm. Other customers have used other arm hardware.

The requirement for the arm is that it be able to support a payload of 4kg (Hand) + whatever your desired payload is - normally, that means you need a total payload of 7-8kg. If you supply us with the relevant information, we can mount the Hand to your hardware.

We have been in conversation with the developers of GraspIT, and they have added the Hand to the system. There is also a Blender model available from our website.

At present, no. We have developed the Hand as a component; most of our interested customers have their own vision systems under development. However, we do have links to other organisations who have developed various components of vision systems; if you are looking at teleoperation use I think this device may be of interest.

It's been designed so the maximum strength is less than that of a typical human, to avoid safety worries. When we do exhibitions, we shake lots of people's hands, and we've never hurt one. Though there were a few times we were tempted.

Pretty much - it has all the movements of the human Hand, and is the same size. We've even implemented comparable touch sensing to the Human hand, as well as position sensing. If we made it any more complex, it would probably be asking us for a job.

Our customers run the Hand from their intelligent software, so, for example, the University of Bielefeld in Germany have had theirs picking up objects under speech direction. We don't do the "application" stuff much yet, preferring to concentrate on making the best Hand we can.

"Many people are scared of opportunity, because it arrives demanding overalls, looking like hard work"

As part of a quotation for a Hand system, we normally specify installation and training. This covers two days on-site time for one of our engineers to take your staff through the setup and use of the Hand system.

Our support package includes email support for most issues that need resolving in the use of the Hand; if you need additional training, we'd be pleased to quote for it.

There are press-quality pictures available; please contact us for details.

Unless an image says otherwise, it is (C) Shadow Robot Company. (If it says otherwise, you probably can't reprint it!)

If you use pictures of our stuff to illustrate the book, then we need to see the context of use beforehand to verify that they are "correct in context" - no fair using our Hand to illustrate someone else's research!

Also, when the book is printed, you must send us a copy so we know where and how it has been used.

No, it really does go back a long way. (We just got -90 out of a wrist to check - but it didn't have tendons.)

You're right: the greater range is for the little finger side.

This is correct.

This is correct

Yes, it does.