Building a custom octocopter from scratch with no prior hardware experience (karolina.mgdubiel.com)

A tech enthusiast has successfully built a custom octocopter from scratch in just 2.5 weeks, despite having no prior hardware experience. Utilizing CAD software Fusion 360 and CNC milling techniques on materials like G10 fiberglass and carbon fiber, the individual designed and assembled a drone that aims to feature a reinforcement learning (RL) controller capable of maintaining flight even after multiple motor failures. This project demonstrates the viability of integrating RL into drone technology, showcasing phases including CAD design, electronics wiring, and eventual flight testing. The project's significance lies in its innovative approach to fault tolerance in drones. While octocopters typically handle single motor failures well due to thrust overcapacity, the introduction of an RL-trained controller is expected to enhance resilience against more complex failure scenarios, such as losing two motors in proximity. Key technical elements involve direct motor control via a serial link, allowing the RL policy to allocate thrust dynamically during failures. The simulator, built in MuJoCo, incorporates multiple training variables to address issues like motor lag and loop latency, ultimately aiming for effective real-time responses in hardware deployment. This effort could broaden the scope of autonomous drone operations in challenging environments, fundamentally advancing the field of AI and robotics.