Amazon Web Services has been selected as Aurora’s preferred cloud provider for machine learning training and cloud-based simulation workloads. Aurora uses AWS’s proven infrastructure and capabilities to safely accelerate the development of Aurora Driver, its scalable autonomous vehicle technology.

The Aurora Driver system consists of several sensors, Aurora’s proprietary software, and a computer that integrates the company’s hardware and software with a variety of vehicle platforms. Aurora has said that it will leverage AWS’s capabilities to optimize and scale its virtual testing efforts to expand the capabilities of it’s Aurora Driver system.

For its machine learning training and cloud-based simulation workloads, Aurora will leverage the cloud capabilities of AWS to process trillions of data points daily. It will also use these capabilities to scale its training workloads in the cloud to complete up to 12 million physics-based driving simulations per day by the end of the year – building on the data collected during its real-world road tests.

Aurora’s AWS-powered Virtual Testing Suite works similarly to aid the development of Aurora Driver. The company can use data from its real-world tests to inspire virtual permutations in the suite. Virtual testing helps train Aurora Driver to quickly, and safely, manage complex driving situations, such as road construction and jaywalkers.

The offline components of the Aurora Driver software stack run on AWS, including the Virtual Testing Suite, high-definition road maps, machine learning models and software development tools. Aurora has adopted Amazon SageMaker to create, run, and continuously refine the machine learning models that enable its driving simulations. This service, offered by AWS, helps developers and data scientists to accelerate the development, training, and deployment of machine learning models. It also provides Aurora with access to Amazon Elastic Compute Cloud instance types that enable high performance for cloud-based machine learning training.

Before developing simulations, Aurora uses AWS to securely store and process the data it logs during real-world testing, and then trains its machine learning models on that data. The pre-processing workloads run on Amazon EKS and Amazon EMR, AWS’s service for processing data in the cloud with open-source tools. It’s machine learning training workloads then rely on deep learning frameworks optimized by AWS, such as TensorFlow and PyTorch. Finally, Aurora orchestrates and auto-scales its simulation workflows over concurrent vCPUs and GPUs with Amazon EKS and Amazon EC2, providing accelerated computing instance types like G4dn.