Pick by what you have
How to read a card. The line under each title is hardware · level · what drives the robot · code or no-code.
- Level. Beginner: you follow the steps, and there is little or no code. Intermediate: you are comfortable in a terminal, in Python, or in the Workflow editor. Advanced: you train models, write drivers, or run Docker and ROS.
- What drives the robot. A perception model, an LLM, a workflow, a VLA or an RL policy: see AI models and Ways to control a robot. Hardware setup means the tutorial connects hardware to a twin and uses no model.
- Early draft. The steps are complete enough to follow end to end. Screenshots and final code are still being added.
No hardware
Everything runs in simulation or in Docker on your computer.UR Sim to Cyberwave with a Python driver
No hardware (Docker) · Intermediate · Hardware setup · CodeRun Universal Robots’ simulator and ROS 2 in Docker, then stream the simulated arm’s joints into your twin with a ~75-line Python driver.
Train an RL policy in MuJoCo
No hardware (real arm optional) · Advanced · RL policy · CodeExport an environment to MuJoCo, train a policy locally with Gymnasium and Stable-Baselines3, replay it on the twin, then run the same script on the real robot.
RL task walkthrough
No hardware · Advanced · RL policy · CodeScript an OpenArm cube-pick RL task in Python: scene, actions, observations, and network. Then export it, train it, upload the policy, and deploy it as a controller.
Laptop camera only
Your computer and a camera are the whole edge device.Your first AI loop: see, pick, sort
Laptop webcam · Beginner · Perception model · CodeA webcam spots a cup or a bottle on your desk. An SO-101 twin in the free Playground aims, grasps and drops it in the right bin. Switch to
live for a real arm.Monitor a camera feed for safety hazards
Laptop or USB webcam · Beginner · Perception model · Code + workflowStream your webcam to a twin. A scheduled workflow then asks a VLM whether each person is wearing their safety gear and emails you when they aren’t.
Detect intruders in a zone
Camera on an Edge Core host · Intermediate · Perception model · No-codeGet an alert when a person stays inside a zone you draw. YOLO and pixelation run on the edge, so raw video never leaves the device.
SO-101 arm
You need an SO-101 leader and follower, a camera, and a computer to run the edge. Start with Get started with the SO-101 arm. The other tutorials in this group assume you have done it.Get started with the SO-101 arm
SO-101 set + camera · Beginner · Teleoperation + recording · No-codePair the arm, teleoperate it with the leader arm, record a dataset, train a model, and deploy it back to the same arm.
Train an AI model on the SO-101
SO-101 set + wrist camera · Intermediate · Recording + VLA · No-codeCollect demonstrations, build a dataset, train a VLA in the cloud, and control the arm with natural-language prompts.
Control an SO-101 arm with your voice
SO-101 follower + USB webcam + mic · Intermediate · LLM · CodeSpeak a command. Claude sees the workspace, writes a motion plan, and the SDK runs it. No training data. You need Anthropic and Mistral API keys.
Build a voice-controlled SO-101 with Workflows
SO-101 follower + mic · Intermediate · LLM · No-code · Early draftThe same voice agent, built on the Workflow canvas. A model picks from saved poses you taught the arm.
Voice-controlled pick and place
SO-101 set + wrist and top-down cameras · Advanced · VLA · Code · Early draftTrain your own VLA, then send it spoken pick-and-place instructions from a laptop voice layer.
Other arms and your own hardware
Get started with the AgileX Piper arm
AgileX Piper + USB-to-CAN adapter · Intermediate · Hardware setup · CodeBring the Piper up over CAN, pair it, jog it from the keyboard in the dashboard, then set its joints from the Python SDK.
Build a voice-controlled AgileX Piper with Workflows
AgileX Piper + mic · Intermediate · LLM · No-code · Early draftSpeak a command, and a model sequences the Piper’s saved poses. Built on the Workflow canvas.
Bring your own hardware
Any device + an edge machine · Advanced · Hardware setup · CodeCreate an asset from your URDF, write a Docker driver and its
cw-driver.yml, and get telemetry and commands flowing through a twin.Mobile robots and drones
Spot left-out equipment with a DJI Mini 3
DJI Mini 3 + RC + Android phone · Beginner · Perception model · No-codePair the drone, and a VLM checks its live feed for equipment left in the open. When it finds some, you get an alert and the drone climbs for a wider view.
Log DJI Mini 3 observations by voice
DJI Mini 3 + mic · Beginner · Workflow · No-code · Early draftWhile you fly by hand, spoken observations become structured alerts on the drone twin. The workflow never flies the drone.
Build a voice-controlled rover with Workflows
Waveshare UGV Beast + mic · Intermediate · LLM · No-code · Early draftSpeak a command, and a model plans a short drive. Each step is checked against the commands the rover accepts before it runs.
Control a UGV rover with your voice
Waveshare UGV Beast + mic · Intermediate · LLM · Code · Early draftThe Python version. Claude reads the rover’s camera frame and your command, then returns a driving plan that your validator clamps before the SDK sends it.
Built by the community
Projects from the Cyberwave Builder Program. Each one links to the author’s full source.Build a sandwich-making robot
SO-101 set + Raspberry Pi 4 + USB webcam · Advanced · VLA · CodeRecord teleoperated demonstrations, fine-tune SmolVLA on Colab (about 1.5 hours on a T4), and run it closed-loop on the arm.
Fold cloth with two SO-101 arms
2 × SO-101 sets + RealSense + Jetson · Advanced · Learned policy · CodeA bimanual setup that learns to fold cloth with SmolVLA, starting from about 50 human demonstrations.
Order-driven zero-shot picking with an SO-101
SO-101 + overhead USB camera · Advanced · Perception model · CodeAn HTTP order names an item. A hosted VLM finds it, the arm picks it and drops it in the bin, then a second look confirms it landed.
Demos (video)
Recorded walkthroughs with no step-by-step instructions. Watch them to see the end result before you build.Run an autonomous rover inspection
Simulation · Video · Perception model + workflowA rover follows inspection waypoints in a simulated environment while a workflow analyzes what its camera sees.
Take a Go2 from digital twin to real
Unitree Go2 · Video (about 3 min)A Go2 goes from the catalog to a paired robot. Along the way you see occupancy mapping, mission design, and a mission run in the real world.
Next steps
Connect AI to robots
How AI agents, AI models, controllers and training data fit together.
Supported hardware
Check what each robot supports before you buy or build.
Idea cookbook
More project ideas to build on these tutorials.
Add your robot
Get a device that isn’t in the catalog supported.