project // claudeDronestage // pre-prototypestack // ros2 · gazebo · esp32

No GPS.
Just a 3D map,
built mid-flight.

An autonomous mapping drone for GPS-denied environments — warehouses, tunnels, collapsed structures, anywhere a satellite signal can't reach. Built on ROS2, ArduPilot SITL, and a layered ToF sensor rig.

read the brief → view repository ↗

// signal_topology.png ClaudeDrone signal visualization

lat 38.5982 N · lon 8.7891 W
carrasqueira / pt

§ 01about

A flying scanner for places GPS can't see.

claudeDrone is an autonomous quadrotor that builds a multi-layer 3D map of its surroundings in real time, without relying on satellite positioning. Our target environments are the ones GPS gives up on — warehouse interiors, parking decks, mines, and post-disaster structures where every other navigation system fails first.

The platform combines a layered TF-Luna ToF sensor array, an ESP32 sensor bridge, and a ROS2 / Gazebo simulation pipeline running against ArduPilot SITL. The "Stop-Scan" behavior lets the drone hold position, rotate, and integrate a clean point cloud before continuing along the path.

Long-term: the same SLAM techniques developed for space rover navigation, ported back into industrial inspection and search-and-rescue work on Earth.

project codenameclaudeDrone

parent orgAeroSearch Solutions

primary domainGPS-denied SLAM

target trl04 → 06

simulationoperational

hardwarein assembly

based insetúbal · pt

licenseMIT (planned)

dev infrastructureclaude_team · ai-assisted

§ 02tech stack

The components that make it fly.

Open-source, simulation-first, hardware-honest. Everything below either runs today in our Gazebo bench or sits on the workbench waiting for a solder iron.

middleware

ROS 2 Jazzy

Node graph, message passing between flight controller, sensor bridge, and SLAM stack.

simulation

Gazebo Harmonic

Custom indoor world, sensor plugins, ROS2 ↔ Gazebo bridge for sim-to-real continuity.

flight stack

ArduPilot SITL

Software-in-the-loop autopilot, MAVLink command channel, mission planner integration.

sensor mcu

ESP32-S3

Reads the ToF array, timestamps frames, streams structured packets to the companion computer.

range sensor

TF-Luna ToF ×4

Layered time-of-flight modules sweeping at offset angles to build a 3D occupancy grid.

behavior

Stop-Scan loop

Hover-and-rotate primitive that captures a clean point cloud before continuing the path.

§ 03architecture

System topology — ESP32 at the center of the bus.

Real-time MCU brokers every sensor and actuator. Companion compute (Pi Zero 2W) is a thin WiFi bridge — flight-critical loops never leave the ESP32. The flight controller runs ArduCopter and stays low-level; mission logic sits one tier up.

ClaudeDrone system architecture: ESP32 MCU as the bus hub, with TF-Luna lidars on UART, VL53L0X array via I²C mux, PCA9685 PWM driver, optical flow over SPI, and a flight controller over SBUS. — // schema v1source: obsidian / drone_schema.md

mcuESP32 DevKitreal-time bus hub
lidar2× TF-Lunascanning + altimeter, UART
tof_array6× VL53L0X360° proximity, I²C mux
flowPMW3901optical flow, SPI
pwmPCA9685 + SG90lidar tilt servo
fcF4/F7 FClow-level stabilization, SBUS
bridgeRPi Zero 2WWiFi telemetry bridge

§ 04roadmap

ClaudeDrone Roadmap

Source of truth lives in Obsidian. CI parses it on every push and writes content/roadmap.json. The site renders the JSON — never edit it by hand. Format guide: docs/roadmap.example.md.

8// shipped

4// active

19// queued

1// blocked

25% complete·parsed may 01, 2026·docs/roadmap.example.md

Oct

Nov

Dec

Jan '25

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

trackSimulation4/11

Gazebo Harmonic + ROS2 Humble bring-up

ArduPilot SITL bridge

Custom indoor world with realistic clutter★

Stop-Scan-Move behavior in sim

SLAM evaluation against ground truth

trackHardware3/11

F450 frame + motors assembled

Pixhawk 6C wired and powered

Sensor bench rig★

Custom sensor PCB

Final airframe with all sensors mounted

trackFirmware1/5

ESP32-S3 dev environment

TF-Luna I²C driver

BNO055 fusion

MAVLink bridge ESP32 ↔ flight controller★

Failsafe behaviors

trackIntegration0/5

Hardware-in-the-loop with simulator★

First tethered hover★

First untethered indoor flight★

GPS-denied mapping demo for ESA BIC★

// backlog

✕Customer pilot site visit— Blocked on partner LOI. Reaching out in Q3.

§ 05gallery

Bench, simulator, schematics.

Real photos and screenshots replace these placeholders as each subsystem comes online.

§ 06video

Watch the simulation fly.

A walkthrough of the Gazebo world, the Stop-Scan behavior, and what the sensor stream actually looks like.

// simulation_walkthrough.mp4 · 04:12

// channel: @ClaudeDrone-f5b6q// 1080p · 30fps// recorded: 2026-04

§ 07documentation

What it's made of, what it runs on.

Open documentation: the bill of materials for the prototype airframe and the development environment we run simulations on. Everything below is the real working setup.

// bill_of_materialsbom.v0.1.csv · 12 items

qty	part	description	status
1	F450 frame	Quadrotor airframe, glass-fiber arms, 450 mm wheelbase.	on bench
1	Pixhawk 6C	Flight controller running ArduPilot Copter 4.5.	on bench
1	Raspberry Pi 5 (8 GB)	Companion compute for ROS2 nodes and SLAM stack.	on bench
1	ESP32-S3 DevKit	Sensor MCU, reads ToF array, streams over UART.	on bench
4	TF-Luna ToF	Time-of-flight modules, 8 m range, layered at offset angles.	on bench
1	BNO055 IMU	9-DoF orientation reference for sensor fusion.	ordered
4	EMAX MT2213 motors	935 kV brushless, with 30A SimonK ESCs.	on bench
2	LiPo 4S 5200 mAh	Flight pack, ~12 min hover endurance.	on bench
1	RFD900x telemetry	915 MHz long-range MAVLink link.	ordered
1	Holybro Sik V3	Backup short-range telemetry, bench debugging.	on bench
1	USB-C power module	Bench supply for ESP32 / sensor calibration rig.	on bench
1	Custom sensor PCB	In design — KiCad project, breaks out 4× TF-Luna + IMU.	in design