Shadow Dexterous Hand (DEX-EE)

Machine learning research breaks hardware. Policies flail, fingers slam into tables, and a delicate hand ends up in the repair queue for weeks. The Shadow Dexterous Hand (DEX-EE) was built for exactly that punishment: a robust three-fingered hand whose fingers can be swapped by the user, so the experiment keeps moving.

The product is available through this page, by quotation.

Background

A hand developed with Google DeepMind

The Shadow Robot Company developed the DEX-EE robot hands in collaboration with Google DeepMind, to meet the needs of their real-world machine-learning projects. That origin shows in the design. The hand is described as a robust 3 fingered robot built for demanding machine learning tasks, not as a delicate showpiece for a display case.

Shadow Robot is known for dexterous hands, and DEX-EE carries that lineage into a form aimed at repeated contact, repeated failure and repeated retries.

Tendons, motors and touch

The four joints of each finger are driven by five motors in an N+1 configuration, connected via tendons. Each motor unit executes a 10kHz force control loop, which gives active compliance to the finger. In practical terms, the finger can yield on contact instead of fighting the object.

Sensing is a major part of the design. Each finger contains 155 individual sensor channels, plus video from the distal tactile sensor. That sensor is a stereo camera pair viewing the inside surface of the silicone skin. The middle and proximal phalanges use user replaceable tactile modules with 22 (proximal phalanx) or 14 (middle phalanx) sensing cells.

Technical Specifications

Specifications as published by the manufacturer; confirmed against the datasheet at quotation.

The manufacturer has not published the ingress protection rating, the operating temperature range, the payload capacity, the cable lengths, the box contents, the control software requirements or the fingertip repeatability; they are confirmed at quotation.

What makes DEX-EE different from a human-scale hand?

DEX-EE is larger than a human hand, and it does not try to imitate one. It has three fingers, each a self contained unit that is user replaceable for quick repair. When a finger is damaged, it is exchanged rather than the whole hand being sent away.

The sensing also sets it apart. Video from all three distal sensors is available over a single USB 3.0 connection, each at 1280x480 and 50 FPS, alongside the 155 channels in every finger. For learning-based work, that is a rich stream of contact information.

Which version should I choose, DEX-EE or DEX-EE Chiral?

The fact sheet names two versions. DEX-EE is the hand described above. DEX-EE Chiral is available in Left, Right and Bi-Manual Pair, and is presented as opening up new possibilities in robust manipulation, whether in machine learning or in applications of robot hands. The sources do not give separate figures for the Chiral, so every figure in this article belongs to DEX-EE. Ask us at quotation how the two compare for your task. More hands are listed in Robot Hands, and the maker's range is under Shadow Robot.

Who is the Shadow Dexterous Hand (DEX-EE) for?

Research labs running reinforcement learning or imitation learning on real hardware are the clearest fit. They need a hand that tolerates hard contact and can be repaired quickly.

Universities and robotics groups building tactile perception work will value the camera-based fingertip sensor and the replaceable tactile modules. Teams that already work in ROS can use the hand directly, since it is fully ROS integrated as a research and development tool. Groups planning two-handed work should ask about the DEX-EE Chiral Bi-Manual Pair.

What is the Shadow Dexterous Hand (DEX-EE) used for?

1. Machine learning research: training and testing manipulation policies on a real, robust hand.
2. Tactile perception: studying contact with the stereo camera fingertip sensor and phalanx modules.
3. Compliant grasping: using the 10kHz force control loop for active compliance.
4. Arm-mounted manipulation: attaching to an arm through the 4 x M6, 50mm PCD pattern.
5. ROS development: building and testing control software over EtherCAT.
6. Rapid repair cycles: swapping a finger unit to keep experiments running.

Pricing and Availability

Pricing is provided by quotation. Delivery, customs and support are arranged with the quotation. Tell us which version and how many hands you need, and the Shadow Dexterous Hand (DEX-EE) is available on this page.

Frequently Asked Questions

How many fingers does the DEX-EE have?

It has three fingers. Each finger is a self contained unit that the user can replace.

How is the hand connected to the host?

Through EtherCAT, a 100 Mbps Ethernet-based field-bus fully integrated into ROS. Tactile video from all three sensors runs over a single USB 3.0 connection.

What power does it need?

The hand requires a 48v 200W power supply.

How does it mount to an arm?

The base connects with 4 x M6 screws in a 50mm PCD pattern, compatible with ISO 9409-1-50-4-M6.

How much does the hand weigh?

The 3-fingered hand weighs 4.1 kg, and a single finger weighs 1.2 kg.

Is a left or right version available?

DEX-EE Chiral is available in Left, Right and Bi-Manual Pair versions.

Where can I buy the Shadow Dexterous Hand (DEX-EE) and what does it cost?

You can request a quotation on this page. The cost is given in the quotation, together with delivery, customs and support arrangements.

Summary

Research hardware is judged by how it survives the second thousand attempts, not the first. The Shadow Dexterous Hand (DEX-EE) answers with replaceable fingers, tendon drive, fast force control and dense tactile sensing. For teams in the Caribbean and beyond who want to take learning-based manipulation onto real hardware, this page is the starting point.