AlphaGo

AlphGo is the first computer program to defeat a professional human Go player, the first program to defeat a Go world champion, and arguably the strongest Go player in history. Can we achieve the same breakthrough in autonomous flying?

FPV Drone Racing

FPV Drone Racing is simply drone racing and FPV stands for First Person View where the pilots use virtual-reality glasses and use wireless controllers to fly the drones at high speeds above 80 MPH and they fly through a 3D course with checkpoints that pilots have to fly through as quickly as possible. It is a new emerging sport that has its own league.

AlphaPilot

Recently, DRL launched the Artificial Intelligence Robotic Racing (AIRR). It is the Drone Racing League’s new autonomous racing class. AlphaPilot is the first challenge 2018 - 2020. The challenge is to use AI that runs on Nvidia Jetson TX2 (and possible Jetson Xavier for 2020), so all processing will be 100% edge computing (on-drone) without offboarding data to be processed in a computer.

The Drone Racing League Simulator

I found DRL simulator in STEAM and I think it’s a good start to get the feel of it and also it could be a possible way to train machine learning models in the way that DeepMind or open OpenAI use in training models to play Atari games like this. Those FPV racing drones simulators could be a little more difficult in integration with machine learning models if we compared with other simulators that were built originally to support research like AirSim. I will explore the possibility of using the racing drone simulators and write about it later.

Research Work

A good starting point is this paper: AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles. It uses a convolutional neural network (CNN) with a path-planning and control system to build a vision-based drone racing in dynamic environments. This video explains how it works.