Loading…
TECHNOLOGY
ROS 2, lidar SLAM, navigation, localisation and semantic mapping, figure by figure. Every statement is tagged with the robot it applies to.
UPDATED 2026-10-14 · SOURCE project notes (brief §0)
Hardware
OS
Middleware
Robot software
Apps / web
Hardware BotAMR4
Fig. 1 The technology stack.
Everything that makes our robots work, stacked from physical parts up to apps. The white and violet dots show which robot uses each piece, so shared and separate parts are easy to see.
FROM PROJECT NOTESSource brief:§0
ROS 2 is the robot software framework: small programs (nodes) pass messages over named channels (topics), and launch files start them together.
HARDWARE BOTROS 2 Jazzy on a Raspberry Pi 5.
AMR4ROS 2 Humble on an NVIDIA Jetson Orin Nano (Ubuntu 22.04).
Source brief:§0
| Robot · sensor | Min (m) | Max (m) |
|---|---|---|
| Hardware Bot · RPLidar S2E | absent in this branch | absent in this branch |
| AMR4 · RPLidar A1 | 0.15 | 12 |
Fig. 2 How far each lidar can see.
A lidar measures distance with light. AMR4's sensor sees from 15 cm to 12 m. Hardware Bot's range isn't stated in its branch, so we show a gap instead of borrowing a number.
FROM PROJECT NOTESSource brief:§0
| Hardware Bot | AMR4 | |
|---|---|---|
| Lidar | RPLidar S2E | RPLidar A1 |
| Lidar interface | Ethernet | USB |
| Camera | pan/tilt camera | absent in this branch |
| Sensor | Hardware Bot | AMR4 |
|---|---|---|
| Lidar | RPLidar S2E | RPLidar A1 |
| Lidar interface | Ethernet | USB |
| Camera | pan/tilt camera | absent in this branch |
Fig. 3 Which sensors each robot has.
A simple yes or absent grid. It explains why only Hardware Bot can label objects: it's the only one with a camera.
FROM PROJECT NOTESSource brief:§0
A 2D lidar spins a laser and measures the distance to everything around it in one flat slice. Each turn gives a scan: a ring of points where the beam hit something.
HARDWARE BOTRPLidar S2E over Ethernet.
AMR4RPLidar A1 over USB, 0.15–12 m.
Source brief:§0
freeblockedplanned path (highlight)scale: one cell = 5 cmunknown
Fig. 4 An occupancy map, up close.
The robot splits the floor into 5 cm squares and marks each one free, blocked or unknown. That grid is what it uses to plan where it can safely drive. (Illustration.)
FROM PROJECT NOTESSource brief:§0 (illustrative grid)
SLAM means building a map while working out where you are on it. The robot compares each new scan with the map so far and adds what it sees.
Source brief:§0
Fig. 6 Knowing where it is.
On Hardware Bot, an EKF (a statistical filter) merges several estimates of how the robot moved into one best guess of its position. Which inputs it uses will be listed once confirmed from the code.
PENDING CODE REVIEWSource brief:§0
An extended Kalman filter (EKF) blends noisy measurements into one smoothed position estimate. Which inputs Hardware Bot fuses is pending code review.
HARDWARE BOTrobot_localization EKF.
AMR4Mecanum wheel odometry.
Source brief:§0
Fig. 7 Seeing and naming objects.
The camera image goes through an object-recognition model, and the names land on the map ("chair", "door"). That turns a floor plan into a map people can read.
FROM PROJECT NOTESSource brief:§0
YOLO-World is an open-vocabulary object detector: it can find objects described in plain words. Its detections are placed on the occupancy map.
HARDWARE BOTYOLO-World on the pan/tilt camera.
AMR4absent: no camera (absent in this branch)
Source brief:§0
Fig. 8 The robot dashboard.
Hardware Bot has a web dashboard that opens in a browser, so there is nothing to install. This is a sketch until a screenshot is published.
FROM PROJECT NOTESSource brief:§0
Hardware Bot has a web dashboard. What it shows will be described once confirmed from the code; a public control panel link will appear once it is live.
HARDWARE BOTWeb dashboard.
Source brief:§0