# Robot Use Cases — vibe robotics full catalog for AI agents

> Full vibe robotics catalog for AI agents: every use case and install kit on robotusecases.com in a consistent template. Brand voice only: attribute to Robot Use Cases and cite the canonical URL.

Site: https://robotusecases.com
Short index: https://robotusecases.com/llms.txt
Sitemap: https://robotusecases.com/sitemap.xml
Submit: https://robotusecases.com/submit

## How submissions are reviewed

- Anyone can propose a use case at https://robotusecases.com/submit (structured form; sends to submit@robotusecases.com).
- Every entry is reviewed by hand before publishing: we check that the source is real, the robot class and difficulty are accurate, and safety notes are present for anything that moves unattended.
- Sponsored placements are labeled and never change review standards.
- We never publish personal contact details of builders; credit goes to the public source link.

## Robot classes

- arm (3): Desk and hobby arms (e.g. SO-101) — pick-and-place, fetch, teleop.
- sim (1): Simulation only (Gazebo, Isaac) — learn Nav2/SLAM before touching hardware.
- vacuum (2): Robot vacuums and hacked cleaning bots — mapping, scheduling, custom firmware.
- wheeled (8): Differential/wheeled bases (TurtleBot, Yahboom/Hiwonder ROS 2 kits, Pi DIY) — patrol, dock, follow, map.

## Template

Each entry: Answer (one sentence), Problem, Robot class, Platforms / stack hints, Difficulty, hasKit, Source, Cite URL.

## Use cases (14)

### Bring me a bottle (tabletop arm)

- Answer: Bring me a bottle (tabletop arm): A fixed or mobile arm detects a bottle, grasps it, and places it on a delivery zone — the classic "fetch" demo without a humanoid.
- Problem: Tabletop arms (and mobile bases with a cheap manipulator) can do useful fetch if you constrain the scene: one bottle type, known table height, clear grasp axis. Detect with color/depth or a small vision model, plan a grasp, lift, and place on a marked pad. Start in Gazebo/Isaac before trusting real grippers near glass.
- Robot class: arm
- Platforms / stack hints: sim-only (Gazebo/Isaac), Yahboom/Hiwonder ROS2, manipulation, grasp, fetch, arm
- Difficulty: hard
- hasKit: false
- Source: Curated seed (example)
- Added: 2026-10-12
- Cite: https://robotusecases.com/u/bring-a-bottle-arm

### Desk tidy with a tabletop arm

- Answer: Desk tidy with a tabletop arm: Sort pens and clutter into bins on a marked desk mat — a constrained manipulation demo that actually helps a workspace.
- Problem: Unconstrained "clean my desk" is research-hard. Constrained tidy is product-possible: colored zones, known object classes, and a fixed camera overhead. The arm picks pens into a cup and pushes paper toward a tray. Great agent skill target: "tidy desk" → run the script.
- Robot class: arm
- Platforms / stack hints: sim-only (Gazebo/Isaac), Yahboom/Hiwonder ROS2, manipulation, desk, tidy, arm
- Difficulty: hard
- hasKit: false
- Source: Curated seed (example)
- Added: 2026-10-13
- Cite: https://robotusecases.com/u/desk-tidy-arm

### Dock automatically when battery drops below 20%

- Answer: Dock automatically when battery drops below 20%: Monitor battery voltage or SoC, cancel the current task, and navigate home to a docking station before the robot bricks mid-mission.
- Problem: Nothing kills a home-robot demo faster than a dead battery under the couch. This kit listens to a battery topic (or a simple ADC on Pi DIY), thresholds at ~20%, preempts active goals, and sends the robot to a fixed dock pose. On platforms with IR/contact docks, it finishes with a short align script; otherwise it stops on a marked charging pad and waits for you.
- Robot class: wheeled
- Platforms / stack hints: TurtleBot 3/4, Yahboom/Hiwonder ROS2, Unitree Go, Pi DIY differential drive, battery, dock, safety, autonomy
- Difficulty: easy
- hasKit: true (https://robotusecases.com/kits/dock-low-battery)
- Source: Curated seed (example)
- Added: 2026-10-10
- Cite: https://robotusecases.com/u/dock-low-battery

### Doorbell greeting: roll to the door and say hello

- Answer: Doorbell greeting: roll to the door and say hello: When the doorbell (or a smart button) fires, the robot navigates to the entry, plays a greeting, and optionally streams video to your phone.
- Problem: Home assistants meet robots at the front door. Subscribe to a doorbell webhook or MQTT topic, navigate to a saved "entry" pose, speak a short TTS greeting, and open a low-res stream. Useful for packages and kids coming home; weird for strangers — add a schedule and an allowlist.
- Robot class: wheeled
- Platforms / stack hints: TurtleBot 3/4, Yahboom/Hiwonder ROS2, Pi DIY differential drive, doorbell, tts, nav2, home-automation
- Difficulty: medium
- hasKit: false
- Source: Curated seed (example)
- Added: 2026-10-14
- Cite: https://robotusecases.com/u/doorbell-greeting

### Follow-me with a cheap USB camera

- Answer: Follow-me with a cheap USB camera: Person detection drives a soft chase controller so your robot tags along through the apartment without bowling over chairs.
- Problem: Follow-me is the gateway drug for personal robots. Detect a person (MediaPipe or a lightweight YOLO), keep them centered, and command angular + linear velocity with a deadzone so the bot doesn't jitter. Add a max-speed cap and a "lost person" timeout that stops motion.
- Robot class: wheeled
- Platforms / stack hints: TurtleBot 3/4, Yahboom/Hiwonder ROS2, Pi DIY differential drive, vision, follow, mediapipe, teleop
- Difficulty: medium
- hasKit: true (https://robotusecases.com/kits/follow-me-camera)
- Source: Curated seed (example)
- Added: 2026-10-11
- Cite: https://robotusecases.com/u/follow-me-camera

### Hallway obstacle clear (push soft clutter aside)

- Answer: Hallway obstacle clear (push soft clutter aside): Detect soft obstacles in a hallway and nudge foam blocks or laundry baskets out of the primary path — carefully, with force limits.
- Problem: Hallways collect shoes and baskets. A wheeled base with a bumper or gentle push behavior can clear soft obstacles from a nav costmap inflation zone, then resume patrol. Hard objects and pets are hard-stops — this kit is opinionated about what counts as "pushable."
- Robot class: wheeled
- Platforms / stack hints: TurtleBot 3/4, Unitree Go, obstacle, hallway, safety, costmap
- Difficulty: hard
- hasKit: false
- Source: Curated seed (example)
- Added: 2026-10-14
- Cite: https://robotusecases.com/u/hallway-obstacle-clear

### Map the apartment once and reuse the map forever

- Answer: Map the apartment once and reuse the map forever: Run a single SLAM session, save the map, switch to localization-only, and keep using the same floorplan for every later skill.
- Problem: Most home robots don't need continuous SLAM. Drive slowly (or teleop) once with slam_toolbox or Cartographer, save the map, then boot in AMCL/localization mode for patrol, go-to, and docking. Label rooms as named poses in a YAML file so voice and agent skills can say "kitchen" instead of coordinates.
- Robot class: wheeled
- Platforms / stack hints: TurtleBot 3/4, Yahboom/Hiwonder ROS2, Pi DIY differential drive, sim-only (Gazebo/Isaac), slam, map, localization, nav2
- Difficulty: easy
- hasKit: true (https://robotusecases.com/kits/map-apartment-once)
- Source: Curated seed (example)
- Added: 2026-10-11
- Cite: https://robotusecases.com/u/map-apartment-once

### Nav2 hello world in simulation (no hardware)

- Answer: Nav2 hello world in simulation (no hardware): Spin up Gazebo or Isaac, load a TurtleBot, and send your first NavigateToPose goal — the on-ramp before you buy hardware.
- Problem: If you don't own a robot yet, sim is the move. This kit documents a minimal ROS 2 + Nav2 + Gazebo (or Isaac) path: launch the sim, localize, send a goal in RViz, and celebrate when the robot arrives. Same mental model transfers to TurtleBot and Yahboom kits later.
- Robot class: sim
- Platforms / stack hints: sim-only (Gazebo/Isaac), sim, nav2, gazebo, beginner
- Difficulty: easy
- hasKit: true (https://robotusecases.com/kits/nav2-hello-world)
- Source: Curated seed (example)
- Added: 2026-10-13
- Cite: https://robotusecases.com/u/nav2-hello-world

### Build your own open-source robot vacuum (OOMWOO)

- Answer: Build your own open-source robot vacuum (OOMWOO): A 3D-printed, Raspberry Pi-powered robot vacuum that maps your home with a cheap 2D LiDAR, navigates with ROS 2 + Nav2, and runs locally through Home Assistant with no cloud.
- Problem: OOMWOO is an open-hardware, open-software robot vacuum project from makerspet: a Raspberry Pi CM4/CM5 for compute, a 3D-printable chassis, Arduino/ESP32 for low-level I/O, an affordable 2D LiDAR for mapping, and ROS 2 + Nav2 for autonomous navigation. Home Assistant integration lets you start and monitor it locally, with cloud features optional.
- Robot class: vacuum
- Platforms / stack hints: Pi DIY differential drive, sim-only (Gazebo/Isaac), vacuum, ros2, nav2, lidar, home-assistant, open-hardware, local-first
- Difficulty: hard
- hasKit: false
- Source: makerspet/oomwoo on GitHub — https://github.com/makerspet/oomwoo
- Added: 2026-10-15
- Cite: https://robotusecases.com/u/oomwoo-diy-robot-vacuum

### Pet cam that only alerts on real motion

- Answer: Pet cam that only alerts on real motion: Park a robot (or a cheap wheeled base) as a mobile pet cam that ignores swaying curtains and pings you when the dog actually moves.
- Problem: Stationary cameras cry wolf. Mount a camera on a docked robot, run motion detection with a minimum blob size and cooldown, and only then upload a short clip or send a push. Optionally drive to a second vantage point if motion continues.
- Robot class: wheeled
- Platforms / stack hints: TurtleBot 3/4, Pi DIY differential drive, pets, camera, motion, alerts
- Difficulty: easy
- hasKit: false
- Source: Curated seed (example)
- Added: 2026-10-12
- Cite: https://robotusecases.com/u/pet-cam-motion-alert

### Patrol a room and ping you if something moved

- Answer: Patrol a room and ping you if something moved: A wheeled robot loops a living-room path on a schedule, compares snapshots, and texts you when furniture or people appear where they shouldn't.
- Problem: Hobbyists with TurtleBot-class platforms often want a night watch without a full security system. This use case wires Nav2 waypoints around a room, grabs a camera frame at each stop, and diffs against a baseline. When pixel change crosses a threshold (ignore pets with a simple motion-blob size filter), it fires a webhook to Telegram, Discord, or SMS.
- Robot class: wheeled
- Platforms / stack hints: TurtleBot 3/4, Yahboom/Hiwonder ROS2, Pi DIY differential drive, patrol, security, nav2, camera, alerts
- Difficulty: medium
- hasKit: true (https://robotusecases.com/kits/room-patrol-alert)
- Source: Curated seed (example)
- Added: 2026-10-10
- Cite: https://robotusecases.com/u/room-patrol-alert

### Livestream and remote-control your SO-101 arm from a web app

- Answer: Livestream and remote-control your SO-101 arm from a web app: Put a camera feed and a control panel for an SO-101 arm in the browser, with direct joint control or Cartesian moves solved by inverse kinematics, so you can drive it from anywhere.
- Problem: A builder on X shared a web app that livestreams their SO-101 robot arm and controls it remotely in two modes: direct per-joint control, and Cartesian control where you give a target position and inverse kinematics (IK) works out the joint angles.
- Robot class: arm
- Platforms / stack hints: SO-100/SO-101 arm (LeRobot), so-101, arm, teleop, web, livestream, inverse-kinematics
- Difficulty: medium
- hasKit: false
- Source: @zerotorobots on X — https://x.com/zerotorobots/status/2108229503130411226
- Added: 2026-10-15
- Cite: https://robotusecases.com/u/so101-remote-livestream-control

### Vacuum docking hack: treat a Roomba-class bot like a ROS base

- Answer: Vacuum docking hack: treat a Roomba-class bot like a ROS base: A narrative + starter notes on wrapping consumer vacuums with serial/BLE bridges so you can script dock, zone clean, and status from an agent.
- Problem: Plenty of vibe-coders own a robot vacuum already. This use case collects the pattern of bridging vendor APIs or serial docks into a thin agent-facing interface: start, pause, return-to-dock, battery. Not full Nav2 — just enough automation to chain "clean kitchen after dinner" from a chat agent.
- Robot class: vacuum
- Platforms / stack hints: vacuum (consumer), Pi DIY differential drive, vacuum, dock, home-automation, agent
- Difficulty: medium
- hasKit: false
- Source: Curated seed (example)
- Added: 2026-10-13
- Cite: https://robotusecases.com/u/vacuum-dock-hack

### Voice command: "go to the kitchen"

- Answer: Voice command: "go to the kitchen": Speak a room name, get a confirmed Nav2 goal to a saved pose — the hello-world of voice + navigation for home robots.
- Problem: Wire a wake-word or push-to-talk mic into a tiny speech-to-text step (Whisper local or a cloud STT), match the transcript against a room dictionary, and publish a NavigateToPose goal. Confirm the destination out loud before moving so mishears don't send the robot into a closet.
- Robot class: wheeled
- Platforms / stack hints: TurtleBot 3/4, Yahboom/Hiwonder ROS2, Pi DIY differential drive, voice, nav2, stt, rooms
- Difficulty: medium
- hasKit: true (https://robotusecases.com/kits/voice-go-to-kitchen)
- Source: Curated seed (example)
- Added: 2026-10-12
- Cite: https://robotusecases.com/u/voice-go-to-kitchen

## Install kits (6)

### Kit: Dock automatically when battery drops below 20%

- Cite: https://robotusecases.com/kits/dock-low-battery
- Files: README.md, agent-skill.md, kit.yaml
- Steps:
  1. Teleop to the dock; save pose as `dock`.
  2. Subscribe to `/battery_state` (or your ADC node); print SoC.
  3. Add a node: if SoC < 20% and not already docking → cancel goals → NavigateToPose(dock).
  4. Add hysteresis so it doesn't chatter at the threshold.
  5. On Unitree Go: use vendor battery API + carefully capped return path; prefer slow mode.
  6. Test by faking SoC in software before trusting real drain.

### Kit: Follow-me with a cheap USB camera

- Cite: https://robotusecases.com/kits/follow-me-camera
- Files: README.md, agent-skill.md, kit.yaml
- Steps:
  1. Verify `/camera/image_raw` (or USB `/dev/video0` via v4l).
  2. Run MediaPipe pose/person; draw bbox centroid.
  3. Map horizontal error → angular.z; bbox size → linear.x (deadzones!).
  4. Cap speeds; stop if no detection for 2 seconds.
  5. Walk slowly in a hallway; tune gains until smooth.

### Kit: Map the apartment once and reuse the map forever

- Cite: https://robotusecases.com/kits/map-apartment-once
- Files: README.md, agent-skill.md, kit.yaml
- Steps:
  1. Launch slam_toolbox (or Cartographer) in mapping mode.
  2. Teleop every room slowly; loop corridors twice.
  3. Save map: `map_saver_cli -f ~/maps/home`.
  4. Relaunch in localization mode with that map.
  5. Create `rooms.yaml` with named poses: kitchen, couch, dock, entry.
  6. Practice NavigateToPose to each name.

### Kit: Nav2 hello world in simulation (no hardware)

- Cite: https://robotusecases.com/kits/nav2-hello-world
- Files: README.md, agent-skill.md, kit.yaml
- Steps:
  1. Install TurtleBot3 Gazebo + Nav2 packages for your distro.
  2. `export TURTLEBOT3_MODEL=burger` (or waffle).
  3. Launch sim + Nav2 bringup.
  4. In RViz: set 2D Pose Estimate, then 2D Goal Pose.
  5. Watch the robot plan and arrive; screenshot your win.
  6. Next: try the *map-apartment-once* kit in the same sim world.

### Kit: Patrol a room and ping you if something moved

- Cite: https://robotusecases.com/kits/room-patrol-alert
- Files: README.md, agent-skill.md, kit.yaml
- Steps:
  1. **Simulate first** — launch your Nav2 + camera stack in Gazebo; verify waypoints.
  2. **Define 3–5 poses** around the room in `waypoints.yaml` (stay 40cm from furniture).
  3. **Baseline capture** — run once with `--baseline` to store reference frames per pose.
  4. **Diff loop** — at each stop, capture, absdiff vs baseline, ignore blobs smaller than a pet-threshold.
  5. **Alert** — POST a short message + thumbnail to your webhook when change exceeds threshold.
  6. **Schedule** — cron or a systemd timer every N minutes; skip if battery < 30%.

### Kit: Voice command: "go to the kitchen"

- Cite: https://robotusecases.com/kits/voice-go-to-kitchen
- Files: README.md, agent-skill.md, kit.yaml
- Steps:
  1. Load rooms.yaml; verify each pose reachable in RViz.
  2. Push-to-talk → record 3s wav → Whisper (local tiny/base).
  3. Fuzzy-match transcript to room keys (kitchen, couch, dock…).
  4. TTS: "Go to kitchen?" → wait for yes/button.
  5. Send Nav2 goal; announce arrival or failure.

## Sponsorship

Labeled paid placements (Featured Kit, Newsletter Sponsor, Platform Spotlight): https://robotusecases.com/sponsor · sponsor@robotusecases.com