Aerial informatics and robotics platform

Machine Learning technologies are increasingly becoming an important tool in building autonomous systems. Learning-based techniques have been useful but typically need a large amount of training data, which is an expensive and time consuming process. Often collecting such data is non-trivial and introduces safety concerns. Consequently, it is becoming increasingly important to be able to accurately simulate the physical environment that autonomous vehicles/robots would operate in. We present a new, easy-to-use simulator that aims to enable designers and developers of a robotic system to seamlessly generate lots of training data. The biggest advantage of this simulator is that it uses recent advances in computation and graphics to simulate the physics and perception such that the environment realistically reflects the actual world. Such realism can enable efficient training and testing of machine learned models by generating vast quantity of ground truth data. One of the key aspects of the fast physics engine is that it enables high frequency simulations with support for hardwarein-the-loop (HIL) as well as software-in-the-loop (SIL) with widely supported protocols (eg MavLink). Our cross-platform (Linux and Windows), open-source architecture focuses on being easily extensible to accommodate diverse new types of autonomous vehicles, hardware platforms and software protocols. We use quadrotors as our first autonomous vehicle showcase.