Interbotix / WidowX
Home > Hardware > Arms > WidowX
Overview
Interbotix is a research robot arm brand developed by Trossen Robotics, providing a series of high-precision manipulators based on ROBOTIS Dynamixel servo motors. The WidowX and ViperX series are robot arm platforms widely used for VLA (Vision-Language-Action) model training, teleoperation data collection, and Embodied AI research. They have been adopted by major robot learning projects including ALOHA, BridgeData V2, and Open X-Embodiment.
| Item | Details | Source |
|---|---|---|
| Manufacturer | Trossen Robotics | Official site |
| Founded | 2005 (some sources cite 2004), by Matt Trossen | Tracxn |
| Headquarters | Downers Grove, Illinois, USA | CBInsights |
| Motors | ROBOTIS Dynamixel X-Series | Official docs |
| Price Range | $2,000 - $6,500 (as of 2024, subject to change) | Product page |
| Main Applications | Research, Education, ML/AI data collection | - |
Company Introduction: Trossen Robotics
Trossen Robotics is a robotics company founded circa 2004-2005 by Matt Trossen, supplying robot hardware to research labs and educational institutions for over 20 years1. They manufacture and distribute research manipulators, unmanned ground vehicles (UGV), and ML/AI integrated research kits, and are particularly famous for their Interbotix brand utilizing ROBOTIS Dynamixel servos.
By supplying hardware for the ALOHA project and Open X-Embodiment dataset, they have established a central position in the Embodied AI research community.
Product Lineup
X-Series Arms (Basic Lineup)
Specifications below are based on official Interbotix documentation, and prices are approximate reference prices as of 2024.
| Model | DoF | Reach | Payload | Servo Configuration | Price (ref) | Features |
|---|---|---|---|---|---|---|
| PincherX-100 | 4 | 335mm | 50g | XL430 | ~$500 | Entry-level, compact |
| PincherX-150 | 4 | 450mm | 50g | XL430 | ~$600 | Entry-level, extended reach |
| ReactorX-150 | 5 | 450mm | 100g | XM430/XL430 | ~$1,200 | Intermediate, wrist rotation |
| ReactorX-200 | 5 | 550mm | 150g | XM430/XL430 | ~$1,500 | Intermediate, extended reach |
| WidowX-200 | 5 | 550mm | 200g | XM430-W350, XL430-W250 | ~$2,500 | Research standard |
| WidowX-250 | 5 | 650mm | 250g | XM430-W350, XL430-W250 | ~$3,000 | Research extended |
| WidowX-250 6DoF | 6 | 650mm | 250g | XM430-W350, XL430-W250 | ~$3,550 | ALOHA Leader arm |
| ViperX-250 | 5 | 650mm | 450g | XM540-W270, XM430-W350 | ~$4,500 | High payload |
| ViperX-300 | 5 | 750mm | 750g | XM540-W270, XM430-W350 | ~$5,500 | High performance |
| ViperX-300 6DoF | 6 | 750mm | 750g | XM540-W270, XM430-W350 | ~$6,130 | ALOHA Follower arm |
Note: Prices may vary; check official site for latest pricing.
Detailed Spec Comparison
Specifications below are excerpted from official documentation (WidowX-200, WidowX-250, ViperX-300 6DoF).
| Item | WidowX-200 | WidowX-250 | ViperX-300 6DoF |
|---|---|---|---|
| Degrees of Freedom | 5 DoF | 5 DoF | 6 DoF |
| Max Reach | 550mm | 650mm | 750mm |
| Total Span | 1100mm | 1300mm | 1500mm |
| Payload | 200g | 250g | 750g |
| Repeatability | 1mm | 1mm | 1mm |
| Accuracy | 5-8mm | 5-8mm | 5-8mm |
| Gripper Opening | 30-74mm | 30-74mm | 42-116mm |
| Servo Count | 7 | 8 | 9 |
| Wrist Rotation | Supported | Supported | Supported |
Payload Note: Official documentation recommends 50% or less extension when using maximum payload.
AI Series (Released 2025)
Trossen Robotics announced a new AI hardware lineup specialized for ML/VLA research in 2025. Information below references WidowX AI official page and Trossen AI page.
| Model | Features | Main Applications |
|---|---|---|
| WidowX AI | 6DoF, 700mm reach, 1.5kg payload, 1mm accuracy, iNerve controller | ML/VLA research base platform |
| Solo AI | Leader-Follower configuration, teleoperation specialized | Optimized for data collection |
| Mobile AI | AgileX Tracer mobile base integration | Mobile Manipulation research |
| Stationary AI | 4-arm compound workstation | Large-scale multi-arm experiments |
WidowX AI is available in three configurations: Base, Leader, and Follower, with the Follower version equipped with Intel RealSense D405 depth camera. According to the official site, shipping started from mid-April 2025.
Dynamixel Servo Technology
The core of Interbotix robot arms is ROBOTIS Dynamixel X-Series smart servo motors. Specifications below reference ROBOTIS e-Manual and Interbotix official documentation.
Key Features
| Feature | Description |
|---|---|
| Position Resolution | 4096 positions (approximately 0.088 degrees) |
| PID Control | User-definable PID parameters |
| Feedback | Real-time monitoring of position, velocity, current, temperature, voltage |
| Communication | TTL or RS-485 (varies by model), 1Mbps default baudrate |
| Compliance | Software-based compliance settings |
Servo Models Used
- XL430-W250: Small, lightweight, for gripper and wrist joints
- XM430-W350: Medium, for intermediate joints, high torque-to-weight ratio
- XM540-W270: Large, for base and shoulder joints, maximum torque
U2D2 Controller
All Interbotix arms connect to PC via ROBOTIS U2D2 interface. As a USB to TTL converter, it provides direct access to Dynamixel Wizard software and ROS/ROS2.
Software Ecosystem
ROS/ROS2 Support
Support status below is based on Interbotix official documentation. ROS distribution EOL (End of Life) status may change over time.
| Version | Status | Notes |
|---|---|---|
| ROS Melodic | Supported (Legacy) | Ubuntu 18.04, EOL 2023 |
| ROS Noetic | Supported | Ubuntu 20.04, final ROS1 LTS |
| ROS2 Galactic | Supported (Legacy) | EOL November 2022 |
| ROS2 Humble | Supported (Recommended) | Ubuntu 22.04 LTS, supported until 2027 |
| ROS2 Rolling | Supported | Development rolling release |
Recommended: ROS2 Humble recommended for new projects.
Provided Packages
- URDF/Meshes: Accurate inertia models included
- Driver Nodes: Physical robot control and joint state publishing
- MoveIt Integration: Motion planning support
- Gazebo Simulation: Simulation environment provided
- MuJoCo Models: Physics simulation (including ALOHA 2)
AI/ML Framework Integration (AI Series)
- Hugging Face LeRobot: Data pipelines and model training
- OpenPI (Physical Intelligence): Pi0, Pi0.5 policy training and inference
- NVIDIA Isaac: Simulation and deployment
- Pre-trained Model Support: ALOHA, BiACT, OCTO, Crossformers, etc.
Key Significance
1. Low-Cost High-Performance Research Platform
Interbotix robot arms provide research-quality precision (1mm repeatability) and reliability at the $2,000-$6,500 price range. This is significantly lower cost compared to traditional industrial robot arms, enabling academia and startups to conduct large-scale data collection and VLA research.
2. Open Source Ecosystem
All hardware designs, drivers, and URDF models are open-sourced. All code is accessible from the interbotix repositories on GitHub and continuously improved through community contributions.
3. Standardized Hardware
Major robot learning datasets including ALOHA, BridgeData V2, and Open X-Embodiment were all collected with Interbotix arms. This allows researchers to directly test and fine-tune pre-trained models on the same hardware.
4. Democratization of Embodied AI Research
From Mobile ALOHA systems (approximately $32,000, per ALOHA 2 paper) to single arms, various options are available for different budgets, enabling more researchers to participate in Embodied AI research.
VLA Research Applications
ALOHA / Mobile ALOHA
ALOHA (A Low-cost Open-source Hardware System for Bimanual Teleoperation) developed by Stanford’s Tony Z. Zhao, Zipeng Fu, and Chelsea Finn research team is built around Interbotix arms2.
| Component | Hardware | Role |
|---|---|---|
| Leader Arms | WidowX-250 6DoF x 2 | Human teleoperator input |
| Follower Arms | ViperX-300 6DoF x 2 | Actual task execution |
| Mobile Base | AgileX Tracer | Movement (Mobile ALOHA) |
| Cameras | 2 wrist + 1 top | Visual input |
ALOHA 2 provides improved performance, ergonomics, and robustness, with all hardware designs and MuJoCo models open-sourced.
Key Papers:
- Zhao et al., “Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware” (RSS 2023)
- Fu et al., “Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation” (2024)
BridgeData V2
A large-scale robot manipulation dataset collected by UC Berkeley RAIL Lab3.
| Item | Details |
|---|---|
| Robot | WidowX-250 6DoF |
| Trajectories | 60,096 |
| Environments | 24 |
| Skills | 13 |
| Control Frequency | 5Hz |
| Average Trajectory Length | 38 timesteps |
Includes various basic manipulation skills such as pick-and-place, pushing, sweeping, drawer/door manipulation, block stacking, and clothes folding. Collected via VR controller teleoperation and is a core component of the Open X-Embodiment dataset.
Key Papers:
- Walke et al., “BridgeData V2: A Dataset for Robot Learning at Scale” (CoRL 2023)
Open X-Embodiment
The world’s largest open-source robot dataset led by Google DeepMind, collected from 34 research labs4.
| Item | Details |
|---|---|
| Total Trajectories | 1M+ |
| Robot Types | 22 embodiments |
| Skills | 500+ |
| Tasks | 150,000+ |
| Data Format | RLDS (TFRecord) |
The Bridge dataset collected with WidowX is a core component of Open X-Embodiment and plays an important role in cross-robot transfer for RT-X model training. Research has confirmed that skills learned from WidowX data transfer to Google Robot.
Key Papers:
- Open X-Embodiment Collaboration, “Open X-Embodiment: Robotic Learning Datasets and RT-X Models” (2023)
OpenVLA
Open-source VLA model developed by Stanford and UC Berkeley research teams5.
| Item | Details |
|---|---|
| Parameters | 7B |
| Training Data | Open X-Embodiment 970k trajectories |
| Base Models | Llama 2 + DINOv2 + SigLIP |
| Training Infrastructure | 64x A100 GPU, 15 days |
Achieved 16.5% higher success rate compared to RT-2-X (55B) across 29 evaluation tasks on WidowX and Google Robot embodiments. Shows particularly strong performance on BridgeData V2 WidowX tasks.
Supports efficient fine-tuning through LoRA and lightweight deployment through quantization, enabling operation on consumer-grade GPUs.
Key Papers:
- Kim et al., “OpenVLA: An Open-Source Vision-Language-Action Model” (2024)
Pi0 (Physical Intelligence)
VLA flow model for general robot control developed by Physical Intelligence6.
| Item | Details |
|---|---|
| Parameters | 3.3B (PaliGemma 3B + Action Expert 300M) |
| Base Model | PaliGemma VLM (SigLIP + Gemma) |
| Training Data | 7 robot platforms, 68 tasks, including Open X-Embodiment |
| Control Frequency | Up to 50Hz |
| Action Generation | Flow Matching (Diffusion variant) |
Demonstrated zero-shot and fine-tuning performance on complex real-world tasks such as laundry folding, table cleaning, grocery bagging, and box assembly. Supports cross-embodiment learning across various robot types including single arms, dual arms, and mobile manipulators.
Released OpenPI framework as open-source in February 2025, fully integrated with Trossen AI hardware.
Key Papers:
- Black et al., “Pi0: A Vision-Language-Action Flow Model for General Robot Control” (2024)
See Also
References
Official Documentation
- Trossen Robotics Official Site
- Interbotix X-Series Arms Documentation
- Interbotix GitHub
- WidowX AI Product Page
- ROBOTIS Dynamixel e-Manual
Datasets
Papers
- ALOHA 2: An Enhanced Low-Cost Hardware for Bimanual Teleoperation
- Mobile ALOHA: Learning Bimanual Mobile Manipulation
- OpenVLA: An Open-Source Vision-Language-Action Model
- Open X-Embodiment: Robotic Learning Datasets and RT-X Models
- Pi0: A Vision-Language-Action Flow Model
- BridgeData V2: A Dataset for Robot Learning at Scale
Software
Footnotes
-
Trossen Robotics company profile referenced from business databases including Tracxn, CBInsights, Crunchbase. Detailed metrics like employee count and revenue may vary by database. ↩
-
Fu et al., “ALOHA 2: An Enhanced Low-Cost Hardware for Bimanual Teleoperation”, 2024. https://aloha-2.github.io/ ↩
-
Walke et al., “BridgeData V2: A Dataset for Robot Learning at Scale”, CoRL 2023. https://rail-berkeley.github.io/bridgedata/ ↩
-
Open X-Embodiment Collaboration, “Open X-Embodiment: Robotic Learning Datasets and RT-X Models”, arXiv:2310.08864, 2023. ↩
-
Kim et al., “OpenVLA: An Open-Source Vision-Language-Action Model”, arXiv:2406.09246, 2024. ↩
-
Black et al., “Pi0: A Vision-Language-Action Flow Model for General Robot Control”, arXiv:2410.24164, 2024. ↩