Considered to be the most advanced Dextrous Hand in the world!
The Shadow Hand is the closest robot Hand to the human Hand available. It provides 24 movements, allowing a direct mapping from a human to the robot.
The Shadow Hand has integrated sensing and position control, allowing precise control from off-board computers, or integration into your existing robot platform.
The Shadow Hand contains an integrated bank of 40 Air Muscles which make it move. The muscles are compliant, which allows the Hand to be used around soft or fragile objects.
The Shadow Hand can be fitted with touch sensing on the fingertips, offering sensitivity sufficient to detect a single small coin.
The Shadow Dexterous Hand is an advanced robot hand system that reproduces all the movements of the human hand and provides comparable force output and sensitivity. This means it can pick up or handle small to medium sized objects and perform precision tasks, so robots using it can have the versatility of humans.
The sensitivity and compliance of the Hand makes it possible for us to manipulate delicate objects such as fruit and eggs. For the same reason, the Hand is safe around human beings, since it is less strong than a person.
Being the same size as a human hand, the system is versatile and suitable for use for a variety of purposes:
At the moment the Hand is mainly sold for research purposes. The University of Bielefeld is using the Shadow Dexterous Hand in their research into situated learning. Carnegie Mellon is using it in their research into grasping. NASA's Robonaut group bought a Shadow Dexterous Hand "to inspire their engineers". Others are interested in using the Hand as a component in their neurological projects, rehabilitation projects, as part of humanoid robots, and many more applications.
Handling of delicate objects:
Handling fruits, eggs and other delicate objects is a task that, at the moment, can only be performed by human beings. In order not to break or bruise the objects a certain level of sensitivity and compliance is necessary, The Shadow Dexterous Hand excels in these areas.
A remote system using the Shadow Hand technology will allow an operator to work in an inaccessible area. This could be a harmful environment where radiation, toxic chemicals or biological hazards are present. The Shadow Dexterous Hand can also allow specialists to be present, whenever needed, anywhere across the world, even in places where humans can not reach. This way machine repair could be done for example on drilling platforms, inside wind turbines etc. Also medical examination, education and training by experts could be done on long distance. Even bomb disposal could be done this way!
Rehabilitation and assistive technology:
One has to be very careful with people in physiotherapy when doing exercises. The right pressure is required, the movements have to be smooth and the actuation needs to be safe. In order for an electronic or robotic device to help the therapist, it needs the same kind of care. Safety is paramount whenever a robot comes into direct contact with people. The muscle technology makes the hand soft, compliant and human-friendly. Therefore we see the Hand technology being useful in rehabilitation and assistive devices.
When doing ergonomic research, one is often in need of a model of a human hand. The Shadow Dexterous Hand is an almost perfect reproduction. It can be used to test all kinds of objects that are designed for human hands: gloves, tools, pens, handles, etc. We also see uses for the Hand in the testing of all kinds of human manipulation, for example the accessibility of objects.
Shadow's objective is to have the Shadow Dexterous Hand as part of a whole system that would help people in their daily environment. It would be a great help, not only for the disabled an the elderly but also for working parents who would rather spend time with their kids than doing the washing and ironing.
If you have a similar problem, or maybe something totally different, do not hesitate to contact us. Our engineers will be happy to look into it. Together we will find a solution.
The form factor of the hand is that of a typical human male. The forearm structure is comparable in length to the human forearm, although at the base it widens to 146mm:
|Finger length from tip of finger to middle of knuckle||100mm|
|Palm length from middle knuckle to wrist axis||99mm|
|Thumb base thickness||34mm|
|Forearm base to wrist axis||434mm|
The Hand, sensors, muscles and valve manifold have a combined weight of 3.9 kg. The centre of mass is approximately 160mm from the base.
There is some variation in movement speeds between the parts of the Hand. Also, different methods of movement produce different maximum speeds. However, the general movement is on average about half the speed of that of a human. For example, the time for transition from open to clenched is 0.2 seconds approx.
The entire system is built with a combination of metals and plastics.
|Palm:||Acetyl, aluminium, polycarbonate|
|Fingers:||Acetyl, aluminium, polycarbonate fingernails and polyurethane flesh|
|Base:||Acetyl, rubber, brass, cork|
Because the system is compliant, these are approximate measures of the maximum available output torques. The Hand is capable of holding its own weight. Measured force and torque maxima for the joints are given in the table below.
|Joint||Maximum Force(N)||Maximum Torque(Nm)|
The Shadow Dextrous Hand uses air muscle technology, and so the system requires both electric current and a source of compressed air.
|Electronics:||0.5 A @ 8 V.|
|Valves:||1 A max @ 28 V.|
|Compressed air (Filtered and oil free) @ 3.5 bar.||Consumption: each muscle has volume approximately 0.015 litres; worst case consumption for whole hand 24 litres/min.|
The Hand is driven by 40 Air Muscles mounted on the forearm. These provide compliant movements. Following the biologically-inspired design principle, tendons couple the air muscles to the joints. Integrated electronics at the base of the hand system drive the pneumatic valves for each muscle and also manage corresponding muscle pressure sensors. Three modes of actuation are used in the Hand system. An opposing pair of muscles permits full control and variable compliance of the movement for most joints. Conditionally-coupled drive is used for the Middle and Distal phalanges of the fingers to produce human movement characteristics.
The hand system presents a Controller Area Network (CAN) bus interface to the outside world. The CAN interface has been tested with standard controller cards as well as the supplied interface cards. All sensor data, components, configuration and controller set points can be accessed over this bus. A simple protocol is used for the communication. Example code for protocol interface is supplied as part of the GPL code base only; alternate licensing is also available as an option. An embedded Ethernet interface option permits direct access to robot data and configuration by TCP/IP communication.
The protocol used allows a variety of system-specific configuration to take place. This includes:
- enable and disable a component of the robot,
- set sensor transmission rates,
- enable and disable valve PID controllers individually,
- change PID controller sensor and target, as well as P,I,D gain values,
- change the CAN addresses used by a component,
- reset components.
The off-board PC provides access to all these functions over CANBUS via shell script, device, file system and program code.
A Hall effect sensor measured with typical resolution 0.2 degrees senses the rotation of each joint. This data is sampled locally by 12-bit ADC s and transmitted ’ on the CANBUS. The sampling rate is configurable up to 180Hz.
The pressure in each muscle is sensed by a solid-state pressure sensor mounted directly on the valve manifold, and measured with 12-bit resolution across the range of 0 - 4 bar.
|First, Middle, Ring finger|
|1||Distal - Middle||-20 – +90||Coupled pair|
|2||Middle - Proximal||0 – +90|
|3||Proximal - Knuckle||-20 – +90||Pair|
|4||Knuckle - Palm||-25 – +25||Single with Spring|
|1||Distal - Middle||-20 – +90||Coupled pair|
|2||Middle - Proximal||0 – +90|
|3||Proximal - Knuckle||-20 – +90||Pair|
|4||Knuckle - Metacarpal||-25 – +25||Single with Spring|
|5||Metacarpal - Palm||0 – +40||Pair|
|1||Distal - Middle||-20 – +90||Pair|
|2||Middle - Proximal 1||-40 – +40||Pair|
|3||Middle - Proximal 2||-15 – +15||Pair|
|4||Proximal - Palm 1||0 – +80||Pair|
|5||Proximal - Palm 2||-60 – +60||Pair|
|1||Palm - Wrist||-55 – +45||Pair|
|2||Wrist - Forearm||-30 – +10||Pair|
The thumb has 5 degrees of freedom and 5 joints.
Each finger has 3 degrees of freedom and 4 joints.
|The Muscle Types|
|Single with Spring||One muscle with an opposing return spring.|
|Coupled pair||The two joints are coupled such that the angle of Joint 2 is less than the angle of Joint 1. Two muscles drive these joints.|
|Pair||Two antagonistic muscles drive this joint.
The distal and middle joints are coupled in a manner similar to a human finger, such that the angle of the middle joint is always greater than or equal to the angle of the distal joint. This allows the middle phalange to bend while the distal phalange is straight. The movement from 0 to -20 of the distal joint is a purely passive movement.
The little finger has an extra joint in the palm.
All joints except the finger distal joints are controllable to +/- 1° across the full range of movement.
A rendering of the kinematic structure of the Hand. Each large cylinder represents one joint of the Hand, aligned approximately at the "at rest" angle.
- Bus: Controller Area Network (CAN) bus interface to on-board electronics. Optional Ethernet on-board.
- Palm Sensor: 7 ADCs distributed across the palm provide 26 active 12-bit sensing channels.
- Valves: Valve driver nodes at base of forearm incorporating per-muscle pressure sensing and providing timed and PID control.
The valve driver boards implement PID control of individual valves. This control can be flexibly configured to take set point and target data from a variety of sources. These controllers can be configured via the standard robot interface and appropriate programmes, scripts and graphical examples of this are provided.
A standard x86-compatible PC (VIA Mini-ITX: others by arrangement) running Debian GNU/Linux with the RTAI real-time system and Shadow s GPLrobot code ’ is supplied. This can be used for initial set up, evaluation and operation, as well as serving as a template for your own control system. The PC is fitted with an external CANBUS interface.
Software in the host PC provides sensor calibration and scaling, mappings from sensor names to hardware and permits easy access to all robot facilities from C code, shell scripts, or GUI.
PIC18F4580 micros are used for embedded control throughout the robot system. The firmware is provided as source on the host PC. All micro-controllers are connected to the robot CANBUS.
Valve control nodes:
- Drives a set of valves at 0.25mS resolution.
The PID controllers can be configured to operate from sensor data and from user-supplied values, permitting control of joint position, muscle pressure, or user-supplied parameters.
Hand sensor node:
Other sensors can be attached to the Hand sensor node by request and arrangement.
- All source code for the micro-controllers and schematics for the electronics subsystems are provided on the host PC.
- Example RTAI real-time code along with full documentation is provided, along with access to e-mail support from Shadow.
- Solid models (VRML) and kinematic data supplied for use in 3D modelling packages.
- Software layer supports easy interfacing between this and other systems, as well as quick prototyping of algorithms and tools.
The Left Hand is functionally identical to the standard Hand, but mirrored for use in a bimanual system.
Tactile sensing can be provided on the finger and thumb tips. This consists of 34 tactels distributed approximately evenly across the tip.
Each tactel will produce a measurable output change for an applied weight of 10g, and provides an output range up to 1000g.
The sensing electronics introduces +/-1g of noise in the 1-10g range, and less at higher ranges. Data from the tactile sensor is transmitted across the CANBUS and made available at the host PC as with other sensor data.
The tactile sensing performance is guaranteed for one year, and the sensors can be easily detached for maintenance if required.