Welcome to the Reinforcement Learning (RL) subgroup of AIDA3 for VIP This repository serves as ongoing documentation for learning progression, code implementations, and our current research for autonomous fixed-wing UAV control.
AIDA3 Research explores and develops machine learning models to solve real-world problems in autonomous fixed-wing aerial vehicles (UAVs). Our Purdue VIP subgroup focuses specifically on the application of Reinforcement Learning to create custom training environments and improve flight path planning.
We are currently developing and testing a path planning system for fixed-wing UAVs in multiple environments. Our research is broken down into three main phases:
- Phase 1: Build a static Pre-Planner for the race environment using Dynamic Programming (DP), Q-Learning, and SARSA agents.
- Phase 2: Integrate Signal Temporal Logic (STL) into the Pre-Planner to structurally define our flight mission requirements.
- Phase 3: Create a Dynamic Planner to work alongside the Pre-Planner, allowing the drone to adjust its flight path in real-time to handle physical uncertainties.