Application of the W-shaped process for a Reinforcement Learning use case on drone navigation

Adding machine learning (ML) to safety‑critical avionics needs strict certification rules. The European Union Aviation Safety Agency (EASA) suggested a “W‑shaped” development process to help with ML integration, but the guidance mainly covers supervised learning. That leaves a big gap for reinforcement learning (RL), which works well for tasks that change over time, such as drone navigation, because RL learns from interactive, non‑static data. This research fills that gap by testing how the W‑shaped process works for an RL‑based drone navigation and collision‑avoidance system running on a real‑time embedded platform. The results show that the parts of the W‑shaped process dealing with implementation do not depend on the type of

ML used, but the learning phases must be changed to fit RL’s interactive nature. The main changes we suggest are: – Manage data by scenario rather than by static datasets. – Rethink traditional data‑quality rules to match RL’s needs. – Clearly define the agent‑environment interface before training starts. These adjustments give a clear way to match RL‑based aerospace systems with the new airworthiness standards.

Message: Hi FE, I need help with “Application of the W-shaped process for a Reinforcement Learning use case on drone navigation” in “Aerospace Engineering”