Community hub
0 subscribers8 pages, 0 posts
Recent from talks
All channels
Be the first to start a discussion here.
Be the first to start a discussion here.
Be the first to start a discussion here.
Be the first to start a discussion here.
Contribute something
Welcome to the community hub built to collect knowledge and have discussions related to Shadow Hand.
Nothing was collected or created yet.
Shadow Hand
View on Wikipediafrom Wikipedia
The Shadow Dexterous Hand is a humanoid robot hand system developed by The Shadow Robot Company in London. The hand is comparable to a human hand in size and shape, and reproduces all of its degrees of freedom. The Hand is commercially available in pneumatic- and electric-actuated models and currently used in a wide range of institutions including NASA, Bielefeld University and Carnegie Mellon University, and EU research projects such as HANDLE.[1]
The Shadow Dexterous Robot Hand is the first commercially available robot hand from the company, and follows a series of prototype humanoid hand and arm systems.
Design
[edit]The Shadow Dexterous Hand has been designed to be similar to the average hand of a human male. The forearm structure is slightly wider than a human forearm.
The Shadow Dexterous Hand has 24 joints. It has 20 degrees of freedom, greater than that of a human hand.[2] It has been designed to have a range of movement equivalent to that of a typical human being. The four fingers of the hand contain two one-axis joints connecting the distal phalanx, middle phalanx and proximal phalanx and one universal joint connecting the finger to the metacarpal. The little finger has an extra one-axis joint on the metacarpal to provide the Hand with a palm curl movement. The thumb contains one one-axis joint connecting the distal phalanx to the proximal phalanx, one universal joint connecting the thumb to the metacarpal and one one-axis joint on the bottom of the metacarpal to provide a palm curl movement. The wrist contains two joints, providing flex/extend and adduct/abduct.
The hand is available in both electric motor driven and pneumatic muscle driven models. The motor hand is driven by 20 DC motors in the forearm, whereas the muscle hand is powered by 20 antagonistic pairs of Air Muscles in the forearm.
All hands have Hall effect sensors integrated into every joint to provide precise positional feedback. The motor hand includes force sensors for each degree of freedom and the muscle hand includes pressure sensors for each muscle. There are also several options for tactile sensing on the hand from basic pressure sensors to the BioTac multimodal[3] tactile sensor from Syntouch Inc..
The Shadow Hand software system is based on Robot Operating System, through which configuration, calibration, simulation and control of the hand is implemented. A simulation of the Shadow hand can be downloaded and installed in ROS.[4]
See also
[edit]Further reading
[edit]- Transformation of Shadow Dextrous Hand and Shadow Finger Test Unit from Prototype to Product for Intelligent Manipulation and Grasping, Marco Reichel, The Shadow Robot Company, Intelligent Manipulation and Grasping, International Conference, July 1–2, 2004, Genova - Italy
- Learning Dexterity, OpenAI
References
[edit]- ^ "Handle Project Website". Handle-project.eu. Archived from the original on 2013-02-26. Retrieved 2013-07-25.
- ^ The Independent Online Article
- ^ "SynTouch". 6 October 2020.
- ^ "Robots/Shadow_Hand - ROS Wiki". Ros.org. 2013-01-18. Retrieved 2013-07-25.
External links
[edit]- Shadow Dexterous Robot Hand home page
- Dungchuot79 Shadow Dexterous Robot Hand Technical Specification Archived 2013-05-10 at the Wayback Machine
- A to Z of Robotics Article
- The Shadow Hand on Robotnik.es
- The Future of Things Article
- BioTac multimodal tactile sensor from Syntouch LLC
Shadow Hand
View on Grokipediafrom Grokipedia
The Shadow Dexterous Hand is an advanced humanoid robotic hand developed by the Shadow Robot Company, a London-based firm specializing in dexterous robotics, featuring 24 degrees of freedom (20 actuated) to closely replicate the kinematics, size, and manipulation capabilities of a human hand.[1] Tendon-driven and equipped with over 100 sensors operating at 1 kHz, including tendon load sensors, an inertial measurement unit, and optional tactile fingertip sensors, it enables precise control for tasks such as tool handling and object manipulation in research environments.[1] Weighing approximately 4.3 kg in its standard configuration, the hand integrates fully with the Robot Operating System (ROS) and supports teleoperation via systems like the Shadow Glove for intuitive human-robot interaction.[1]
Founded in 1987 as a hobbyist group and formally established as a company in 1997, the Shadow Robot Company has invested over two decades in advancing robotic dexterity, with the Shadow Dexterous Hand emerging as its flagship product around 2000 as the first commercially available version following prototype development.[2][3] The hand's design draws from extensive research into anthropomorphic robotics, incorporating underactuated fingers for natural grasping and high-bandwidth torque/position control to achieve human-like precision, making it a benchmark for studies in artificial intelligence and machine learning.[1] Early applications included collaborations with organizations such as NASA for space exploration tasks and pharmaceutical firms like GSK for laboratory automation, highlighting its versatility beyond academia.[2]
In recent years, the Shadow Dexterous Hand series has evolved with variants like the lighter models (e.g., Super Lite at 1.8 kg) for energy efficiency and affordability, alongside the DEX-EE version developed in partnership with Google DeepMind since around 2019 to support reinforcement learning in real-world manipulation.[1][4] The DEX-EE emphasizes robustness, featuring high-speed tactile sensors with hundreds of taxels per finger, stereo camera-based 3D sensing, and resistance to impacts and aggressive use, enabling long-duration experiments without frequent maintenance.[4] A parallel 2024 release, the New Shadow Hand, shifts toward modularity with self-contained finger units and GelSight-inspired optical tactile sensors, prioritizing durability over strict anthropomorphism while maintaining force exertion up to 8 N and joint speeds of 180°/s.[5] In 2025, the company announced ARIA-funded projects (OGRES and UPWARD) to further advance dexterity and began exploring applications in care robotics.[6][7] These advancements have earned the company accolades, including the 2019 AIconics Award for Best Innovation in AI Hardware and the Queen's Award for Enterprise in 2019.[2]
These configurations collectively lower barriers to entry for users exploring humanoid robotics, building on the foundational design of the standard Dexterous Hand without compromising tendon-based precision.[1]
History
Founding of Shadow Robot Company
The Shadow Robot Company traces its origins to 1987, when it began as an informal hobbyist group founded by Richard Greenhill in the attic of his home in London, England. Greenhill, a photographer by trade with no formal background in engineering or robotics, gathered a group of about 12 enthusiasts who met weekly on Wednesdays to explore humanoid robotics. Driven by a passion for creating machines capable of human-like manipulation in everyday environments, the group focused on developing affordable components using scavenged materials, such as parts from skips and old printers, emphasizing open-source principles by freely sharing their knowledge and designs with the broader community.[8][9] Initial projects centered on rudimentary prototypes, including basic wooden models of robotic hands and arms constructed from materials like maple sourced from local waste. These efforts were motivated by the members' lack of professional technical expertise, relying instead on self-taught experimentation to mimic human dexterity without relying on expensive commercial hardware. The group's early work highlighted a commitment to accessible research tools, aiming to democratize humanoid robotics for academic and hobbyist applications.[8][10] By the mid-1990s, the hobbyist collective had evolved into a more structured operation, formally registering as the Shadow Robot Company in 1997 following a commission to build a robotic leg component. This milestone marked the transition to a professional entity, enabling the pursuit of advanced research and development through initial funding opportunities, such as a 1998 Smart award from the UK government that supported further innovation in dexterous manipulation systems. By the early 2000s, the company had secured additional resources to expand its R&D efforts, solidifying its role in the field of humanoid robotics.[8][9]Early Prototypes and Technological Evolution
The development of the Shadow Hand began in 1987 with the creation of the first prototype, a wooden model designed to mimic the anthropomorphic form of the human hand, constructed by a group of hobbyist robotics enthusiasts.[11] This initial design focused on replicating the basic structure and proportions of an adult human hand to explore foundational concepts in dexterous manipulation. Over the following years, the team iterated through a series of humanoid hand and arm prototypes, testing for enhanced dexterity and integrating early control systems to enable coordinated movements. These prototypes progressively addressed the integration of mechanical components with rudimentary software for basic task execution, laying the groundwork for more advanced systems.[12] A significant advancement occurred in 2004 with the introduction of pneumatic "Air Muscle" actuators in the prototype design, which provided human-like force generation and compliance without relying on rigid electric motors.[13] These actuators, inspired by McKibben-style muscles, allowed for softer, more adaptable movements that better simulated natural hand dynamics, marking a shift toward bio-inspired actuation in the prototypes.[14] The 2004 Bielefeld Shadow Hand prototype, featuring 20 degrees of freedom (DoF), exemplified this evolution and became available for research testing.[12] Throughout this period, key challenges included achieving over 20 DoF within a compact, human-scale form factor comparable to an adult hand's dimensions, while ensuring reliability in dexterous tasks.[1] Engineers addressed issues such as actuator compliance, joint precision, and overall system integration to prevent bulkiness or loss of agility.[13] A notable milestone came in 2007, when the prototype demonstrated basic grasping capabilities, such as securely holding fragile objects through anthropomorphic manipulation, validating the design's potential for real-world dexterity.[12]Commercial Launch and Key Milestones
The Shadow Dexterous Hand was first commercially released in 2005 as the flagship product of the Shadow Robot Company, marking the transition from research prototypes to a market-available robotic manipulator designed for advanced dexterity studies.[15] This launch enabled broader access for academic and institutional users, with the hand offered in pneumatic and electric variants to support diverse manipulation tasks.[1] Early adoption included integration into the European Union's HANDLE project (2009–2013), a FP7-funded initiative focused on replicating human-like grasping and in-hand manipulation, where the Shadow Hand served as the primary end-effector alongside a biomorphic arm for human-robot interaction research.[16] In 2013, NASA acquired a tactile-equipped Shadow Hand for its Robonaut program at the Johnson Space Center, utilizing it for space manipulation experiments to enhance robotic dexterity in extraterrestrial environments.[17] Key milestones in accessibility and demonstration included the 2010 integration with the Robot Operating System (ROS), which standardized control interfaces and expanded its use in open-source robotics research by enabling seamless software reusability for dexterous tasks.[18] This was followed in 2012 by public YouTube demonstrations showcasing human-like gestures, such as precise object handling, which highlighted the hand's anthropomorphic capabilities and garnered attention from the global robotics community.[19] By the mid-2010s, the company expanded its offerings to include integrated arm-hand systems, with the Shadow Dexterous Arm introduced as a complementary platform emphasizing the hand's role in full upper-limb teleoperation. Over the subsequent decade, iterative improvements—spanning more than 20 years since the initial commercial release—have positioned the Shadow Hand as a staple in dexterity research, adopted by numerous leading institutions worldwide, including OpenAI for AI-driven manipulation studies.[17][4]Design and Specifications
Mechanical Structure and Degrees of Freedom
The Shadow Dexterous Hand features an anthropomorphic five-fingered structure designed to closely mimic the size and shape of a typical human male hand, measuring approximately 20 cm in length with fingers of equal length and staggered knuckles for realistic fingertip positioning.[20] This design enables precise manipulation tasks by replicating human-like proportions and joint arrangements. The hand incorporates 24 joints in total, providing 24 degrees of freedom with 20 actuated. The thumb has 5 joints and 5 actuated degrees of freedom (DoF): the interphalangeal (IP), metacarpophalangeal (MCP) flexion/extension, MCP abduction/adduction, and two carpometacarpal (CMC) joints for opposition and flexion. Each of the four fingers has 4 joints: metacarpophalangeal (MCP) flexion/extension, proximal interphalangeal (PIP), distal interphalangeal (DIP; coupled and underactuated to PIP), and MCP abduction/adduction, yielding 3 actuated DoF per finger (MCP flexion/extension, PIP flexion driving the coupled DIP, and MCP abduction/adduction). The little finger includes an additional actuated palm abduction joint for enhanced thumb opposition, giving it 4 actuated DoF. The integrated wrist contributes 2 actuated DoF (flexion/extension and abduction/adduction).[20][1] This configuration supports 24 distinct movements overall, closely approximating human hand dexterity, with the 4 underactuated DoF being the DIP joints of the fingers. The mechanical transmission employs a tendon-driven system, which contributes to the hand's compact and lightweight profile, with a total weight of 4.3 kg for the hand and forearm assembly.[20] The structure utilizes lightweight materials such as aluminum and brass alloys for the frame, acetyl and polycarbonate for joints and components, and polyurethane for the synthetic flesh covering, ensuring durability during repetitive operations.[20]Actuation Mechanisms
The Shadow Dexterous Hand employs dual actuation paradigms—electric and pneumatic—to drive its 20 actuated degrees of freedom, allowing researchers and engineers to select based on application needs for precision or compliance. The electric variant utilizes 20 high-precision DC motors, specifically Maxon EC flat series motors integrated into proprietary "Smart Motor" nodes, each equipped with gear reduction to amplify torque for fine joint control. These motors are mounted in the forearm, providing backlash-free operation and enabling precise positioning suitable for tasks requiring high accuracy, such as delicate manipulation in teleoperation.[21] In contrast, the pneumatic variant relies on 20 pairs of contracting air muscles (40 total actuators) mounted in the forearm, which contract linearly when pressurized to mimic biological muscle action and produce compliant motion for adaptive grasping. These air muscles, often McKibben-style or similar linear pneumatic actuators, offer inherent softness and shock absorption, making the hand ideal for interacting with fragile or irregularly shaped objects without rigid force application. The system's design ensures variable stiffness through antagonistic pairing, where opposing muscles balance tension for controlled compliance.[22] Both variants transmit force via a tendon routing system, where braided steel cables run from the forearm-mounted actuators through the palm to the finger and thumb joints, facilitating natural curling and spreading motions that replicate human kinematics. This underactuated tendon drive enables efficient force distribution across 24 total movements, with four underactuated degrees of freedom for passive adaptation. Control is achieved at up to 1 kHz update rates using the EtherCAT protocol for real-time responsiveness, supporting closed-loop position or torque commands. The electric model requires a 48 V DC power supply at 2.5 A, while the pneumatic system needs compressed air at approximately 3.5–6 bar, with a maximum consumption of around 24 liters per minute under full load. These mechanisms highlight the hand's versatility: pneumatics excel in soft, adaptive scenarios, whereas electrics provide superior precision and repeatability without the need for air infrastructure.[21][22][23]Sensors and Tactile Feedback Systems
The Shadow Hand utilizes Hall effect sensors to provide joint position feedback for all 20 actuated degrees of freedom, enabling absolute positioning with a typical resolution of 0.2 degrees and 12-bit ADC sampling.[20] These sensors measure magnetic field rotations along each joint axis, ensuring precise tracking of hand configuration during manipulation tasks.[24] Force and torque sensing in the Shadow Hand is implemented via strain gauge-based load cells on tendon pairs, with approximately 40 such sensors in the full model configuration.[20] These allow detection of fingertip forces up to 10 N, with a resolution of about 30 mN, supporting torque control and compliant grasping.[25][20] Sampling occurs at 500 Hz for force data, contributing to responsive feedback in dynamic interactions.[20] Tactile feedback systems in the Shadow Hand incorporate over 100 sensors overall, with options for advanced fingertip technologies to capture contact details.[1] The BioTac sensors, developed by SynTouch and mountable on all five fingertips, deliver multimodal data including pressure distribution (via 19 force outputs), vibration, and temperature, facilitating nuanced touch perception akin to human skin.[26][27] Alternatively, the proprietary Shadow Tactile Fingertips (STFs) employ 17 taxels per unit, each with three-axis Hall effect sensing for 3D force vectors (normal and tangential), enabling detailed surface profiling and contact localization; standard setups include STFs on the thumb and index finger, with expandability to others.[20] STF data is captured at 1000 Hz with 12-bit resolution, supporting high-fidelity tactile mapping for manipulation.[20] Additional sensing includes a single inertial measurement unit (IMU) per hand for orientation and motion tracking, integrated into the sensor suite for enhanced spatial awareness.[1] All primary sensors, including position, force, and tactile arrays, operate at a 1 kHz sampling rate via the EtherCAT bus, ensuring low-latency feedback for real-time control.[1] This data can be processed within the Robot Operating System (ROS) for higher-level integration.[20]Software and Control
Integration with Robot Operating System (ROS)
The Shadow Dexterous Hand has been fully compatible with the Robot Operating System (ROS) since 2010, supporting ROS 1 through dedicated packages that enable joint control, trajectory planning, and hardware abstraction.[28][20] As of 2025, support remains primarily for ROS 1 (Noetic). The integration is facilitated by the open-sourceshadow_robot_ethercat stack, available on GitHub, which provides drivers for seamless hardware-software interfacing and allows users to extend functionality for custom applications.[29][30]
Communication between the hand and ROS occurs via an EtherCAT bus, an Ethernet-based protocol operating at 100 Mbps, enabling real-time command execution with low latency of approximately 1 ms for control loops running at 1 kHz on the host PC.[20] This setup supports position control via PID on the host and torque control closed at 5 kHz within individual motor units, ensuring precise and responsive operation.[20]
For teleoperation, ROS APIs map human inputs from the Shadow Glove to the hand's joints, streaming glove data at up to 960 Hz with end-to-end latency as low as 1 ms, allowing intuitive replication of natural movements.[31] Sensor data streams, such as position and tactile feedback at 1000 Hz, are published directly to ROS topics for integration into control pipelines.[20]
Safety features are embedded in the ROS nodes, including built-in torque limits, adjustable operational boundaries for force and temperature, and emergency stop mechanisms that monitor current and thermal states in the smart motor units to prevent overload or damage.[20] These protections integrate with ROS's hardware abstraction layer, enabling safe trajectory execution and rapid motor resets when limits are approached.[20]
Simulation Tools and Programming Interfaces
The Shadow Dexterous Hand integrates with Gazebo, an open-source robotics simulator, to enable physics-based modeling of hand dynamics and interactions in virtual environments. Gazebo supports accurate simulation of the hand's 20 actuated degrees of freedom, tendon-driven mechanisms, and sensor feedback, allowing researchers to test grasping, manipulation, and contact scenarios without physical hardware. Installation typically involves a Docker-based setup on Ubuntu with ROS Noetic, using commands likeroslaunch sr_robot_launch srhand.launch sim:=true to launch unimanual or bimanual configurations, including optional arm integrations such as UR10. This setup facilitates custom scene loading for complex interactions, such as object manipulation in cluttered spaces, and incorporates NVIDIA GPU acceleration for enhanced performance.[32]
URDF models for the Shadow Hand are provided as modular xacro files, which generate Unified Robot Description Format descriptions for kinematics visualization and integration with the MoveIt! framework. These models define joint limits, link geometries, and collision properties, enabling path planning, inverse kinematics solving, and collision avoidance in simulated environments. MoveIt! compatibility allows for motion planning pipelines tailored to the hand's anthropomorphic structure, supporting tasks like trajectory optimization for dexterous reaching and grasping. Researchers can load these URDFs into the MoveIt Setup Assistant to configure semantic grasp planning and execute plans via ROS topics.[33][26]
Programming interfaces for the Shadow Hand include Python and C++ APIs accessible through the ROS ecosystem, facilitating scripting of complex manipulations such as multi-finger coordination and reinforcement learning setups. The sr_hand ROS package provides nodes for joint control, trajectory following, and sensor data streaming, allowing developers to write custom controllers in Python for rapid prototyping or C++ for performance-critical applications. These APIs support reinforcement learning frameworks by exposing observation spaces (e.g., joint positions, velocities) and action spaces (e.g., torque commands), enabling training of policies for tasks like in-hand object reorientation.[34]
Dedicated simulation tools, such as the Shadow Hand Simulator based on Gazebo, MuJoCo, and NVIDIA Isaac Sim, allow offline testing of grasping algorithms by replicating real-world physics and sensor noise. MuJoCo models from DeepMind's Menagerie provide high-fidelity rigid-body dynamics for rapid iteration, while Isaac Sim offers GPU-accelerated environments for large-scale simulations. These tools generate synthetic datasets for algorithm validation, reducing hardware wear and accelerating development cycles.[35][36]
The Shadow Hand's simulation ecosystem supports integration with machine learning libraries like TensorFlow, enabling training on simulated data for tasks such as policy optimization via reinforcement learning. Simulated environments produce diverse interaction data, which can be processed in TensorFlow for neural network training, bridging the sim-to-real gap through domain randomization techniques. This compatibility has been demonstrated in seminal works on dexterous manipulation, where policies trained in simulation transfer to physical hardware with minimal fine-tuning.[34][24]
Variants and Models
Standard Dexterous Hand
The Standard Dexterous Hand represents the flagship model in Shadow Robot's lineup, designed for high-precision manipulation tasks in research environments. It features a baseline configuration with 20 actuated degrees of freedom (DoF) driven by tendon mechanisms, enabling 24 joints across five fingers to closely replicate human hand kinematics. The system weighs 4.3 kg, including the integrated forearm, and employs 20 Maxon DC motors housed in smart motor units for precise control via PWM and PID algorithms. This tendon-driven architecture allows for underactuated movements, such as coupled distal interphalangeal joints in the fingers, supporting a wide range of joint angles (e.g., 0–90° for proximal finger flexion).[20] Performance-wise, the hand excels in executing 24 distinct movements, including precision pinch grasps, power grasps for larger objects, and complex tool manipulations like pen handling or key insertion, thanks to its over 100 sensors providing position feedback at 0.2° resolution and 1000 Hz sampling. These capabilities make it suitable for advanced dexterity studies, such as in-hand object reorientation or bimanual coordination. Accessories enhance its versatility; it is compatible with a full robotic arm offering 7 DoF for extended reach and singularity avoidance during mounting on mobile bases or fixed setups.[20][1] Pricing and availability are tailored for institutional users, with custom orders placed directly through Shadow Robot Company, starting at approximately €110,000 as of late 2022 estimates (including shipping, installation, training, and initial support). However, the model's limitations include elevated power consumption at 48 V and 2.5 A, alongside increased mechanical complexity from its full actuation, which demand more robust control systems and maintenance compared to lighter variants.[37][20]Lite Series Configurations
The Lite series of the Shadow Dexterous Hand represents scaled-down variants designed to enhance accessibility for research, education, and prototyping applications by reducing complexity and cost while preserving essential dexterity. Introduced in 2021, these configurations address budget constraints in settings where the full 20 degrees of freedom (DoF) of the standard model may be excessive, offering a tendon-driven architecture that maintains compatibility with core control systems.[38][1] The Lite configuration features 13 DoF across 16 joints, with three fingers and one thumb, enabling independent control for tasks requiring moderate precision; it weighs 2.4 kg and incorporates 13 DC motors to retain core finger functionality while simplifying wrist and palm mechanisms for improved energy efficiency.[1][38] In contrast, the Extra Lite variant prioritizes essential grasping with 10 DoF across 12 joints, utilizing two fingers and one thumb powered by 10 DC motors, at a weight of 2.1 kg, by omitting advanced thumb opposition capabilities to further streamline design.[1][39] The Super Lite model offers the most basic setup with 7 DoF across 8 joints, comprising one finger and one thumb driven by 7 DC motors, weighing 1.8 kg, and suited for simple manipulation tasks that demand minimal anatomical replication.[1][40] Across all Lite series variants, the tendon-driven actuation system ensures reliable force transmission, and full integration with the Robot Operating System (ROS) supports seamless programming and simulation for educational and developmental workflows.[1][38]| Variant | DoF (Joints) | Weight (kg) | Fingers (+ Thumb) | Actuators |
|---|---|---|---|---|
| Lite | 13 (16) | 2.4 | 3 + 1 | 13 DC motors |
| Extra Lite | 10 (12) | 2.1 | 2 + 1 | 10 DC motors |
| Super Lite | 7 (8) | 1.8 | 1 + 1 | 7 DC motors |