Allivesim: An Extensible Virtual Environment For Training Autonomous Vehicles

Virtualization technologies have become commonplace both in software development as well as engineering in a more general sense. Using virtualization offers other benefits than simulation and testing as a virtual environment can often be more liberally configured than the corresponding physical environment. This, in turn, introduces new possibilities for education and training, including both for humans and artificial intelligence (AI). To this end, we are developing a simulation platform AlLiveSim. The platform is built on top of the Unreal Engine(1) game development system, and it is dedicated to training and testing autonomous systems, their sensors and their algorithms in a simulated environment. In this paper, we describe the elements that we have built on top of the engine to realize a Virtual Environment (VE) useful for the design, implementation, application and analysis of autonomous systems. We present the architecture that we have put in place to transform our simulation platform from automotive specific to be domain agnostic and support two new domains of applications: autonomous ships and autonomous mining machines. We describe the important specificity of each domain in regard to simulation. In addition, we also report the challenges encountered when simulating those applications, and the decisions taken to overcome these challenges.