Build applications for any robot
Thousands of modules. One API.
The core services, in viam-server.
Motion planning, the frame system, data capture and sync, vision, navigation, and the fleet layer. Open source, runs on your hardware, works offline.
Someone already wrote the driver.
Arms, cameras, grippers, boards, sensors, and models from every major vendor and from other builders. Add one line of config and the component is on your machine.
The same API, from the other side.
Implement an interface in Python, Go, C++, or TypeScript and your hardware, service, or control logic is a first-class component: configured, versioned, deployed over the air, and under the same limits as everything else. Publish it privately or to everyone. Agents can write them too.
Built for agents.
Any vendor’s hardware. One API.
Arms, cameras, grippers, motors, and sensors expose the same interface whatever the vendor.
When the part you need isn't there, you extend Viam rather than work around it: implement the interface, and your driver, service, or control logic becomes a component like any other, versioned, deployed, and governed the same way.
Drive UR, xArm, Yaskawa, Dobot, UFactory and more through one arm API
Add a Registry module to a machine with one line of config
Write and publish your own module in Python, Go, C++, or TypeScript
Know where everything is, and move without hitting anything.
Viam provides motion planning with built-in safety guardrails, giving AI agents access to the robot's geometry, environment, and movement constraints through consistent APIs. Agents can plan around obstacles and adapt as the scene changes.
Declare frames, geometries, and obstacles in config; read them back from any client
Collision-free motion in one call, with constraints (keep level, straight line)
Rebuild obstacles between moves as the scene changes
SLAM safely
A mobile base and its mounted arm share one frame system, giving applications and AI agents a consistent view of the robot's position and reach. Build across drive types and environments without changing how you interact with the platform.
One base API across drive types; motors, encoders, and movement sensors under it
SLAM and mapping from the Registry (Cartographer and others), maps stored and shared per fleet
Navigation service: waypoints, obstacle avoidance, GPS and IMU fusion
Teleoperate any base from the app or your own UI
Built-in computer vision
Color detection, depth-based obstacle detection, and camera transforms run in viam-server and give the same answer every time.
When a decision has to be deterministic, you configure the rule rather than train a model: set the threshold, point it at a camera, and the result comes back in the same frame system your motion code uses, ready for the planner to act on.
Color and shape detection by threshold, with no dataset and no training
Obstacles from depth cameras and point clouds, turned into geometry the motion planner avoids
Crop, resize, and filter camera streams in config before the image reaches your code or a model
Test in Sim
The same machine config runs against Isaac Sim and against hardware. Build the cell, plan the motion, run the detector, and let an agent drive it, all before a robot is plugged in. When it works, point the config at the real arm.
xIsaac Sim as a Viam module: a simulated UR5e, cameras, and world in your org
Agents complete tasks in the sim over MCP with no hardware attached
Synthetic data and sim-to-real with NVIDIA Isaac and Cosmos
