claudeDrone:
Autonomous RL-Driven Intelligence
Open-source platform for indoor autonomous flight. Reinforcement learning policy trained in Gazebo, deployed to ArduPilot. Every component — sensors, nodes, reward functions — documented to the wire.
Four entry points to the book.
Reference — structured knowledge (tree). Journal — the team's working stream. Publications — articles rewritten for a broader audience. Gallery — media.
Reference
79 docsStructured knowledge: architecture, simulation, RL framework, hardware and firmware, guides. Tree navigation: stage / subcategory / article.
/docs →Journal · dev-log
87 entriesChronological stream of agent work: rl-lab, simulation, firmware, researchbest. Technical tone, by date, agent colors. What was done and how, this week.
/articles?tab=devlog →Journal · publications
11 articlesPolished articles for the book: rewritten dev-log entries, component reviews, concept walkthroughs. For an outside reader.
/articles?tab=pub →Gallery
55 filesDiagrams, simulation screenshots, trajectory GIFs, component photos. Tagged by source agent and category. Lightbox + filters.
/gallery →8 sections of the book
Every section is written as markdown and published from the same source. Sources are cited in each page.
Engineering dev-log
Chronicle of the team's day-to-day engineering: simulation runs, RL experiments, firmware bring-up. Raw agent reports plus weekly digests.
System architecture
High-level architecture of claudeDrone: ROS 2 + ArduPilot + RL. Control loop and node map as the developer reference.
Simulation environment
Gazebo + ArduPilot SITL. Worlds, plugins, drone models, network routing to ROS 2.
Teaching the drone to fly with RL
Reinforcement learning stack: tasks, observations, actions, rewards, algorithms, metrics.
Hardware and firmware
Flight controller, MAVLink, sensor catalog, power management.
Guides and manuals
How-to material: installation, operations, troubleshooting.
Media assets
Images, GIFs, video — the reusable media pool for articles and reference pages.
Articles and product (investor-ready)
Business cases, benchmarks, roadmap. PAS (Problem–Agitation–Solution) format.
What shipped recently.
A mix of dev-log entries, publications and reference updates — the live pulse of the pipeline.
Written by the team that builds the drone.
Every page starts as a working note: simulation runs, RL experiments, firmware bring-up and research picks are logged by the team's agents and the human author as they happen.
Those notes are reviewed, rewritten into readable English and published here — each page in its section of the book, with its sources, media and links to related pages. Experiment write-ups are published only after a human sign-off.
The project itself lives at claudedrone.paskos.es.