AyroX Labs builds the robots. The software ships as separate platforms, each named for what it does — and each new one will be too.
Write real ROS2, run it against Gazebo, and watch it in RViz2 — in a browser tab. Nothing to install.
LiveTrain a policy in simulation, then deploy it to a physical robot over an encrypted link, with safety checks and one-click rollback.
In developmentBuild your environment, define your task, and train your policy entirely in simulation. No hardware required to get started.
Run your trained policy against structured benchmarks in our cloud-hosted robot fleet. Real hardware, real physics — no lab needed.
Catch the sim-to-real gap before it costs you. AyroX surfaces the delta between simulated and physical behavior, so you know exactly what to fix.
Push the same policy directly to physical hardware. One codebase, one pipeline, from first training run to production robot.
01 Simulate 04 Deploy
Our quadrotor flies a warehouse aisle with no GPS. Its lidar maps the racks and flags each box it has scanned. Once the run holds up in simulation, the same policy flies the same airframe for real.
Your policy runs against physical robots you never have to buy, maintain, or stand next to. Queue an evaluation on a platform, then read the sim-to-real delta against the same benchmark you ran in Gazebo.
The reference ROS 2 learning platform. Small, predictable, and well documented — the fastest way to get a navigation or SLAM stack running against real wheels.
// SLAM, navigation tuning, multi-robot formation
Built on the iRobot Create 3 base with a depth camera and onboard docking. Handles longer autonomy runs and perception work that the TurtleBot 3 cannot carry.
// Depth perception, autonomous docking, long-horizon runs
A 4-DOF arm with a parallel gripper, driven by the same DYNAMIXEL bus as the bases. Pairs with a TurtleBot for mobile manipulation, or runs bench-mounted for pick-and-place.
// Grasping, pick-and-place, mobile manipulation
Our own quadrotor, built around a 360° lidar for indoor mapping. It is where a SLAM and navigation stack written in simulation meets real flight.
// Indoor mapping, SLAM, waypoint navigation
AyroX connects simulation, cloud evaluation, and hardware deployment into a single, hardware-agnostic platform — so your team ships faster without rebuilding the pipeline for every new robot.
Write once, train anywhere. AyroX's simulation environment is physics-accurate and hardware-agnostic — your policy doesn't know or care which physical robot it will eventually run on.
4
Available
1
Running
0
Offline
Access a managed fleet of physical robots in the cloud. Run your policy on real hardware without owning, maintaining, or physically accessing a single robot.
The same policy you trained and validated deploys directly to your physical hardware — no re-implementation, no porting, no surprise behavior changes at the last step.
Most robotics infrastructure is optimized for either academic research or scaled production — not the critical middle ground where policies actually get validated and shipped. AyroX is built specifically for that gap: rigorous enough for real-world deployment, accessible enough for a two-person team.
Validated on real hardware, not just a simulator score.
No lab, no fleet to own, no infrastructure team to hire.
One policy, many robots. The pipeline does not change.
Sim, cloud eval, and deployment - never rebuilt per robot.
0
Robots to own
Validate on real hardware you never buy or maintain
1
Policy, sim to real
One codebase from first training run to production
100%
Hardware-agnostic
Your policy does not care which robot it lands on
Structured tasks run in both simulation and on physical robots, so the delta is measured, not guessed.
Reproducible evaluation runs on real hardware, with the same protocol every time you change a policy.
Where simulated behavior diverges from physical behavior, and what actually makes a policy transfer.
RoboLab is AyroX's developer environment — a browser-based platform where developers learn to write real ROS2 code, run it against a live Gazebo simulation, and see results in RViz2, all in a single tab. No local setup. No Docker installs. No dependency management. Just open a browser and start building.
Try RoboLabWhether you want RoboLab access, a cloud fleet evaluation, or a conversation about deploying to your own hardware — send the details and we will route it to the engineer who owns that area.
