Allegro Hand
Home > Hardware > Hands > Allegro Hand
Overview
Allegro Hand is a multi-fingered robot hand for research developed by Wonik Robotics in South Korea. It was commercialized under license based on technology developed by the Humanoid Robot Hand Research Group at the Korea Institute of Industrial Technology (KITECH).
Technical Specifications
| Item | Spec |
|---|---|
| Manufacturer | Wonik Robotics |
| Headquarters | Bundang-gu, Seongnam-si, Gyeonggi Province, South Korea |
| DoF | 16 (4 fingers x 4 joints) |
| Fingers | 4 (including thumb) |
| Weight | 1.20 kg (finger 0.17 kg x 3 + thumb 0.19 kg) |
| Grasp Capacity | Up to 1.5 kg |
| Max Torque | 0.17 Nm (overdrive: 0.70 Nm) |
| Max Joint Speed | 0.11 s/60° |
| Joint Sensor | Potentiometer (resolution: 0.002°) |
| Communication | CAN bus (1 Mbps) |
| Control Frequency | 333 Hz |
| Power | 12-24V, 120W |
| Price Range | Approx. $24,500 (V4, as of 2024) |
Key Significance
Allegro Hand has established itself as a standard platform for dexterous manipulation research:
- Accessible Pricing: At approximately 1/5 the price of Shadow Hand (€110K+, ~$120K), suitable for research lab deployment
- Research-Friendly Design: 16 DoF independent torque control enables complex grasping/manipulation research
- Rich Ecosystem: ROS support, simulation environment integration (MuJoCo, Isaac Gym), various research papers
- Anthropomorphic Structure: 4-finger structure including thumb suitable for human hand movement research
Version History
| Version | Release | Key Features | Notes |
|---|---|---|---|
| v1.0 | Early | Basic functionality | Reliability issues reported |
| v2.0 | - | Improved design | Some hardware issues |
| v3.0 | - | Enhanced stability | Continuous improvements |
| v4.0 | Early 2020s | Major technical/design improvements | Current main model |
| v5 (3F) | Latest | 9 DoF, 360° tactile sensors | 3-finger version |
| v5 Plus (4F) | Latest | 16 DoF, 360° tactile sensors | 4-finger premium |
v4.0 Key Improvements
- Significantly Enhanced Reliability: Resolved hardware issues from previous versions
- Improved Usability: Simplified setup and operation
- Design Improvements: Enhanced appearance and structural completeness
- Expanded Sensor Options: Various fingertip sensors and attitude/orientation sensors (AHRS) available
- Software Support: Improved ROS packages (Python library, 32/64-bit support)
VLA Research Applications
Allegro Hand is widely used in Vision-Language-Action (VLA) and learning-based manipulation research:
Major Research Cases
| Research Area | Application | Notes |
|---|---|---|
| Imitation Learning | Data collection via teleoperation, then BC/DAPG training | CMU LEAP Hand comparison study |
| Reinforcement Learning (RL) | MuJoCo/Isaac Gym simulation → Sim2Real transfer | Standard platform for policy learning |
| Visuomotor Diffusion | Complex in-hand manipulation with visual-motor diffusion policy | Lid turning, etc. |
| MPC with Forward Models | In-hand manipulation with forward model-based MPC | Luo et al., 2023 |
| Grasp Planning | Grasp planning considering uncertainty | Comparison with Barrett Hand |
Learning-Based Research Advantages
- DAPG (Demo Augmented Policy Gradient): Effective policy learning in sparse reward environments
- Multimodal Feedback: Integration of vision, tactile, and proprioception possible
- Sim2Real Transfer: Easy application to actual hardware after simulation training
Limitations
- Limited finger count compared to Shadow Hand (5 → 4)
- Pre-v5 models lack built-in tactile sensors (separate purchase required)
- No wrist joint (separate robot arm required)
Hardware Connection
Communication Interface
- CAN Bus: 1 Mbps speed, multiple hands can be connected simultaneously
- Supported CAN Interfaces: NI USB-8473s, PEAK CAN, etc.
- Connector: JST PHD series 20-pin (power and communication)
Mounting
- Mounting block: Uses 6 M3 flat head screws (3 on each side)
- Compatible with standard robot arms (UR, Franka, KUKA LWR, etc.)
Software Support
| Platform | Support |
|---|---|
| ROS/ROS2 | Official support (v1, v2, v4) |
| Windows | Supported |
| Linux | Supported (32/64-bit) |
| MuJoCo | Community models |
| Isaac Gym | Community models |
Purchase Information
- Price Inquiry: info@wonikrobotics.com
- Warranty: 90-day hardware warranty (1-year warranty available with certain options)
- Options: Fingertip sensors (XELA, BioTac, etc.), AHRS sensors available separately
- Availability: Currently on backorder
References
- Allegro Hand Official Site
- Wonik Robotics Official Site
- Allegro Hand Wiki
- ROS Package (v4)
- Dexterous Manipulation for Multi-Fingered Robotic Hands With Reinforcement Learning: A Review
- Survey of Learning-Based Approaches for Robotic In-Hand Manipulation