Frequently asked questions

Mission control for autonomous inspection robots. You write a mission script on the left and watch an ANYmal C quadruped execute it on a live 3D industrial site on the right — with mission telemetry, POI coverage, findings and alerts on the dashboard. Everything runs deterministically in your browser.

No. patrol-lab is an independent portfolio project and is not affiliated with ANYbotics AG. The ANYmal C robot model comes from MuJoCo Menagerie (anybotics_anymal_c, BSD-3-Clause), vendored with attribution.

No — and the UI says so truthfully: the status pill reads SIM KINEMATIC. The robot follows a kinematic trot gait with heading control, driven by a fixed-timestep (120 Hz) mission interpreter. No dynamics engine runs in the browser.

Not for the same seed. The simulation is deterministic: a seeded RNG produces the sensor findings, the interpreter runs on a fixed timestep, and nothing depends on wall-clock time. Pick the same seed and preset and you get the exact same mission, readings and alerts.

A small DSL: MISSION, GOTO, INSPECT, WAIT, SPEED and ALERT IF. The parser reports errors with line and column; the interpreter executes the program step by step, with pause/resume/step playback. The editor ships presets such as patrol_alpha to start from.

Never. The LLM is an optional plugin that can only propose mission DSL code from a plain-language request. The proposal is validated by the DSL parser before it can run — “AI proposes, the interpreter disposes”. Without an API key the compiler runs in mock mode, clearly labeled MOCK in the UI.

From real metric sources in the running simulation: mission clock, odometer, POI coverage, findings and alerts all read from interpreter state. No number on the dashboard is hard-coded or mocked. The inspection report cites mission event IDs for the same reason.

Thermal, visual and acoustic inspections, with seeded readings and threshold alerts such as ALERT IF TEMP>75. Findings carry a severity (info / warn / critical) and appear in the event log as the mission executes.

No. patrol-lab is a browser simulation of an inspection workflow — there is no connection to real hardware or a real fleet. It demonstrates the control-room experience, a deterministic mission interpreter and honest telemetry.

The layout adapts below 900 px (viewport on top, editor below), but patrol-lab is designed as a desktop control room — a wide screen gives the intended experience.

patrol-lab is an open portfolio project by Rico. The stack: Next.js (App Router), React Three Fiber, CodeMirror 6 and zustand, deployed on Vercel. The source is public on GitHub.