Trained policies. Edge hardware. Open-source stack. Any robot.
Pretrained RL policies, edge hardware boxes, and an open-source brain stack — drop the whole pipeline into any robot platform without rebuilding from scratch.
No credit card required · Any robot platform · Open-source SDK + hardware
Sound familiar?
Your demo is in 3 days.
The robot still can't do the new task. You're staring at weeks of training work you don't have time for.
You just won the contract.
It requires a capability your robot doesn't have. The clock is ticking and building from scratch isn't an option.
Your competitor shipped first.
They deployed a new skill in days. You're still setting up your simulation environment.
These aren't edge cases. Every robot team hits them. The difference is how fast you recover.
See how we solve it →The problem
Here's what gets in the way.
So you can't get from "I need this behavior" to "robot does it" in hours. Here's what gets in the way.
Generic behaviors fall short in real environments.
When you need robots that match real-world environments, generic behaviors fall short—so the outcome you need keeps slipping.
You rebuild from scratch instead of reusing what works.
So you end up rebuilding from scratch instead of reusing what already works—and time-to-robot-does-it stretches out.
Training stays slow and expensive across teams.
Training RL for real tasks stays slow and expensive when it's scattered across teams—so you rarely get to "robot does it" fast.
The promise
From environment scan to deployed robot skill
Out-of-the-box robot skills
Supported training environments
Powered by cap-x — open-source robotics AI framework, peer-reviewed and published
How it works
Three steps. No ML PhD required. Just scan, train, ship.
Use our LiDAR scanner app to capture your real workspace. Point, scan, done — you get a simulation-ready 3D scene that matches your actual environment.
iPhone LiDAR app · 3D mapping · Instant sim export
Our AI agent writes and iterates Python control code in cloud simulation. No ML expertise needed — the system auto-synthesizes skills and runs reinforcement learning on your captured environment.
Code-as-Policy agents · Cloud GPU · RL fine-tuning
One click pushes the trained policy to your robot — running on your own compute, our Kinematics Mini ($1,499 open-source box), or Kinematics Max (industrial Jetson T4000/T5000 or AGX Orin). Works across humanoids, quadrupeds, and manipulators.
Mini / Max / BYO compute · One-click deploy · Cross-embodiment
Works with any robot
Unitree
Humanoid / Quadruped
Franka
Manipulator
Universal Robots
Manipulator
Cross-embodiment transfer — train once, deploy to multiple robot types · More platforms in active development
Platform
A vertically integrated platform — all in one place.
From environment capture to real-world deployment: pretrained RL model hub, real-to-sim, cloud simulation, one-click deploy, and a growing marketplace of robot skills.
Foundation models for any robot platform.
A versioned hub of foundation models for humanoids, quadrupeds, and manipulators. Reuse what others have trained—no rebuilding from scratch.
Scan your environment, train in simulation.
Capture real environments with our LiDAR scanner app or 3D mapping—so you get simulation-ready scenes that match your actual world.
Cloud simulation—iterate before you deploy.
Fast simulation-based training on your captured environment. Iterate in hours, not months, before deploying to real hardware.
New skill on real hardware in under 60 minutes.
Get trained models onto real hardware fast. A new robot skill in under 60 minutes, with support for major platforms.
Subscribe to skills—locomotion, manipulation, tasks.
Subscribe to pretrained policies for locomotion, manipulation, and task automation. Every new skill becomes available to all subscribers.
Hardware
Open-source Mini or industrial Max — both run the same brain stack.
The same brain stack on two NVIDIA Jetson form factors. Mini is open and 3D-printable for makers and labs. Max is industrial-grade for production fleets — with modular payloads.
Open-source edge brain
or build it yourself — STL files included
NVIDIA Jetson Orin Nano + 2× Intel RealSense D435i in a 3D-printable case. Ships with the OpenBrain stack — teleop, SLAM, Nav2, RL deployment. Drop it on any ROS2 robot.
Industrial brain — fully configurable
Build-to-order configurator · everything optional · volume pricing for fleets
Choose Jetson T4000 (64 GB), T5000 (128 GB), or AGX Orin 64 GB in a machined aluminum enclosure. Configure cameras (RealSense D435i / D456 / ZED X), LiDAR (Livox / Hesai / Ouster), industrial IMU (Xsens / VectorNav), storage, payloads (thermal, mmWave, RTK, mics, F/T, telephoto, 5G), and software tier — every component optional.
Or run the brain stack on your own NVIDIA Jetson — OpenBrain is MIT-licensed.
Prices are starting figures in USD, exclude shipping, duties, and taxes, and are subject to change without notice based on component availability and supplier pricing. Specifications may evolve. Third-party brand names (NVIDIA, Intel RealSense, Livox, Hesai, Ouster, Xsens, VectorNav, Stereolabs ZED, etc.) are trademarks of their respective owners — see our Disclaimer.
Why OpenKinematics
Every robot team faces the same wall. We built the ladder over it.
Research-backed
OpenKinematics is powered by cap-x — an open-source robotics AI framework developed in our labs — and OpenBrain, the MIT-licensed edge stack that runs on Kinematics Mini, Max, or any NVIDIA Jetson. We publish our research so the field advances. Our platform packages it into a production-ready product teams can actually ship with.
CaP-Gym
39 tasks across 3 simulators. Agents control robots by composing Python code.
CaP-Agent0
Training-free agents with visual differencing and auto-synthesized skill libraries.
CaP-RL
Post-training with GRPO reinforcement learning. Language models learn from environment rewards.
CaP-Bench
8-tier evaluation framework testing abstraction levels and interaction modes.
OpenBrain
MIT-licensed edge stack: ROS2 Humble, teleop, SLAM, Nav2, RL deployment. Runs on Kinematics Mini, Max, or any Jetson.
Kinematics Lab
Open demos hub — runnable apps spanning manipulation, locomotion, VLA, and teleop. Filter by hardware tier.
Kinematics Lab
Reference apps for the Mini and Max boxes — manipulation, locomotion, VLA, teleop.
Curated reference apps that run on Kinematics Mini, Max, or any NVIDIA Jetson. Each demo ships with source code, hardware requirements, and a one-line install.
Predefined teleop dashboard with dual cameras, 3D model, map, and controls.
Drag-and-drop dashboard — every CockPit widget, your layout.
Multi-waypoint autonomous patrols with a state-machine orchestrator.
Pricing
Integrate in under 1 hour. No ML expertise needed. Upgrade as your robot fleet grows.
Locomotion and balance skills that premium vendors charge thousands for — free.
OpenBrain Pro — fleet, OTA, 5G relay
Task-specific skills for teams shipping real-world deployments fast.
Optional add-ons
Custom training, dedicated env · min. 3-robot commitment
Industrial workflows, custom training, dedicated success manager. Need a motion we don't have? We train it.
Optional add-ons
+ Hardware
Kinematics Mini from $1,499 — open-source 3D-printable Jetson Orin Nano box. ·
Kinematics Max from $4,999 — fully configurable: industrial LiDAR, IMU, and modular payloads.
All prices are in USD, exclude applicable taxes (VAT/GST), shipping, and duties, and are subject to change without notice until a written order is confirmed by us. SLA targets, savings figures, and integration-time claims are illustrative — see our Terms of Service, Disclaimer, and applicable Service Level Agreement.
The brain for any robot
Subscribe to skills, drop in a Mini or Max box, run open demos.
One stack. Any robot.
No credit card required · Open-source SDK + hardware · Any robot platform