Humanoid Platform for Robotics Research
A modular humanoid robot purpose-built for research. Train VLA policies, collect teleoperation data, validate in simulation, deploy on hardware — everything ready out of the box.
Built for Modern Robotics Research
Everything your lab needs to start collecting data, training policies, and running experiments — shipped with the robot.
VLA-Optimized Vision
Wrist cameras with customizable mounting angle — ideal for Pi0.5 and other VLA models. Plus head-mounted stereo for full-scene perception.
Pi0.5 · SmolVLA · ACTReady to be Trained
Pre-configured for modern ML pipelines. Start collecting demonstrations and training policies on day one — no integration work required.
Teleoperation Pipeline
Tele-op support out of the box for demonstration recording and remote control. Plug into your imitation learning workflow immediately.
URDF for Simulators
Full URDF robot model included. Drop-in compatibility with MuJoCo, Isaac Sim, PyBullet, Gazebo, and the rest of the standard sim stack.
Simulator Bundled
Simulator ships with every robot. Develop and validate in sim before touching hardware.
SDK
Complete SDK for control, perception, and integration. Python-first and well-documented.
Grippers Included
Manipulators ship with the platform. No need to source end-effectors separately.
Modular Hardware
Swap arms, grippers, and base modules. Adapt the platform to your research, not the other way around.
Sample Scenarios
Pre-recorded movements and example execution scenarios. Validate the stack on your bench in minutes.
Detailed Documentation
Comprehensive docs for hardware, software, calibration, and integration workflows.
Direct Engineering Support
Direct line to our team — from unboxing through deployment, experiments, and publication.
Why labs choose Prometheus
There's no off-the-shelf humanoid built for research. Teams burn months gluing together hardware, cameras, sim, and ML tooling before they can run a single experiment. Prometheus ships that whole stack ready to go.
Research-first, not a demo unit
Open architecture, full SDK, URDF, and bundled simulator — designed around how robotics research actually works.
Skip months of integration
Cameras, teleop, and ML pipelines are pre-configured. Start collecting demonstrations and training policies on day one.
Hackable & modular
Swap arms, grippers, and base modules, write your own controllers, extend the SDK. Adapt the platform to your work.
Custom builds
Bespoke sensor packages, alternative grippers, custom kinematics, and software integrations for specialized lab work.
Technical Specifications
Upper Body on tripod — suitable for bench-top research
| Maximum Joint Torque | 120Nm |
| Height | 160cm on tripod (172cm with future legs module) |
| Width | 50cm (tripod bottom 85cm) |
| Degrees of Freedom | 19 today (up to 53 with roadmap modules: legs, 5-finger hands) |
| Cameras | Head-mounted stereo + wrist cameras with customizable angle (ideal for Pi0.5 and other VLA models) |
| Compute | Raspberry Pi 5 / NVIDIA Jetson |
| Robot Model | URDF included (MuJoCo, Isaac Sim, PyBullet, Gazebo) |
| Software | SDK + bundled simulator |
Modules on Roadmap
Rail Module
Motorized linear axis — in development. The robot will travel along a rail to serve multiple stations on the line.
Legs Module
Bipedal locomotion — in development. Will extend the platform from stationary stations to mobile tasks.
Manipulator Options
Choose the right end-effector for your research
Type A
2-finger gripper
Simple pinch / precise closing. Ideal for delicate operations and precise manipulation tasks.
Type B
4-finger "claw" (paired fingers)
For robust gripping of crates and tote boxes. Designed for warehouse and logistics applications.
Type X
Advanced manipulators
Options include wrist cameras with customizable mounting angle (optimized for Pi0.5 and other VLA models); 5-finger hands are in development, along with other custom variants.
Frequently Asked Questions
Everything you need to know before getting started
Out of the box it supports modern vision-language-action and imitation-learning policies, including Pi0, Pi0.5, ACT, and SmolVLA. Collect teleoperation data, fine-tune on a consumer GPU, and deploy through the SDK.
Every unit ships with the full SDK and a simple REST API, a URDF model, a bundled simulator, head-mounted stereo plus wrist cameras, grippers, and direct engineering support. The teleoperation pipeline works on day one.
Yes. Teleoperate via VR with a Meta Quest 3S or a leader-follower controller, record demonstrations in the standard dataset format, and train ACT or fine-tune a vision-language-action model on a single consumer GPU.
Onboard compute is a Raspberry Pi 5 or an NVIDIA Jetson. Lightweight policies run on-device, while heavier vision-language-action models can run on a tethered workstation GPU and stream commands over the REST API.
Yes. You can swap grippers and end-effectors (2-finger, 4-finger, five-finger hands) to match your task. The base today is a fixed tripod; a motorized linear axis (rail) and legged locomotion are on the roadmap.
Prometheus is designed and manufactured in the European Union, which matters for European research labs and companies that care about supply-chain provenance, support, and EU data residency.
Ready to Put Prometheus to Work?
Whether you are collecting datasets, training policies, or teaching robotics — order a unit or book a call to discuss your configuration.