Implementing an Autonomous FPV Racing Module

Goal

1. Getting an idea of classical path planning algorithms used in navigating an unknown environment We implemented these 2D algorithms using gui libraries in C++ and Python. This is organized under /simulations_2d
2. Implementing a naive approach of hard-coding positions of the gate, and commanding the UAV to the centre coordinates; which are in turn obtained by classical detection methods. The source code and simulation demo for this can be found under /naive_fpv_simul
3. Implementing an intelligent system to navigate at efficient trajectories. We are re-implementing the approach demonstrated recenetly by UZH - "Champion-level drone racing using deep reinforcement learning". We are aiming to mould the approach in order to make it more robust, at the expense of time or efficiency.

Other References

• AlphaPilot: Autonomous Drone Racing - UZH
• Learning High-Speed Flight in the Wild

Learning Resources

• Google DeepMind - RL/DL Lecture Series