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

Article summary

The author built a custom octocopter from scratch with no prior hardware experience, designing it in Fusion 360 and assembling it by hand. The goal is to create a fault-tolerant drone that can sustain flight through single, dual, and quad motor failures using a reinforcement learning (RL) trained controller. The author has completed the initial phases of the project, including CAD design, assembly, and initial flight tests. The next step is to implement the RL policy and achieve sim-to-real transition.

Main themes

• Drone development
• Reinforcement learning
• Fault tolerance
• Hardware design
• Autonomous flight

What commenters say

• The term 'octocopter' is a widely accepted and commonly used term in the drone community, despite its etymological inaccuracies.
• Reinforcement learning is overkill for this project and other control methods like MPC might be more suitable.
• The use of RL in this project is justified as it allows for dynamic adaptation to changing environments and can potentially lead to more advanced autonomous flight capabilities.
• The author's decision to use RL is driven by their desire to work on an RL project, rather than a specific need for it in this particular application.
• The standard control method for drones is PID, but higher-end systems may use more advanced algorithms like MPC or RL.
• The project's use of a custom-built drone and RL controller is an interesting approach, but may not be the most practical or efficient solution.