AutoSPADA - A platform for distributed data analytics

Project owned by Fraunhofer-Chalmers Centre English
324d ago update
New methods and workflows for processing large-scale distributed data are needed to address the challenges and opportunities of big data at the edge. Within the AutoSPADA project, we at Fraunhofer-Chalmers Centre have developed an edge computing platform that enables such distributed data analytics. We built on the ideas and experiences from the preceding project (OODIDA) to create a platform reaching TRL-7 (System prototype demonstration in operational environment).

The AutoSPADA platform puts the data and computational resources of the edge directly in the hands of its users. Users write Python tasks that execute on any number of connected vehicles. With our provided Python library, users can easily access vehicle signals and publish results in their tasks. No assumption is made as to how the Python task is written. This means that you have full flexibility to implement anything you want, from simple statistics to complex use cases such as federated learning. Concrete applications include A/B testing, fuel economy analysis, and road condition monitoring.

As a general edge computing platform, you can still use AutoSPADA for traditional data collection. However, its true potential lies in its interactive nature that lets you perform analytics on live data and immediately see the results. This interactivity facilitates rapid prototyping, allowing users to quickly gain data-driven insights that support decision-making, and adapt to changes in customer needs or behaviour.

Partners: Alkit Communications, Chalmers, Volvo Cars

Founded on the ideas and experience from the OODIDA project, we designed and implemented a new architecture with a strong focus on scalability, privacy, and maintainability. Although nearly all technologies and infrastructure have changed, the design remains centred around three kinds of nodes.

The client nodes are the edge component of the AutoSPADA platform and are responsible for spawning assigned tasks and reporting the results of these tasks.

The user nodes send tasks to be run on the clients and retrieve task results. Tasks are expressed in Python code and are executed in containers on the host, which makes the platform agnostic to the task implementation.

The server nodes are the bridge between the user and client nodes. Server nodes follow the stateless paradigm, meaning that servers read and write the necessary state to a shared database for each request. Paired with a partition-tolerant database, the backend achieves a high degree of horizontal scalability.

All inter-node communication is encrypted with TLS and authenticated by OIDC using JSON Web Tokens (JWTs) or by mutual TLS (mTLS) using X.509 certificates. By using OICD for user authentication, organisations can easily integrate their existing user databases.

We deployed the platform in simulated, lab, and industrial environments. AI Sweden supplied us with several edge nodes through their Edge Learning Lab. This allowed us to validate the platform using the same hardware as our targeted in-vehicle computer.  In the final months of the project, we successfully deployed the platform to two Volvo C40 cars via Alkit Communications’ in-vehicle WICE platform. To demonstrate the platform, our collaborators at Volvo Cars designed two use cases – durability analysis and battery usage in electrical vehicles. As the very first users, they implemented the use cases as AutoSPADA tasks on the Volvo C40 cars.

A preprint that discusses the system in-depth is available on arXiv:  arXiv:2311.17621

pages

Resources

2023-12-04 11:48 Weblink Research & Reports
Contemporary connected vehicles host numerous applications, such as diagnostics and navigation, and new software is continuously being developed. However, the development process typically requires offline batch processing of large data volumes. In an edge computing approach, data analysts and developers can instead process sensor data directly on computational resources inside vehicles. This enables rapid prototyping to shorten development cycles and reduce the time to create new business values or insights. This paper presents the design, implementation, and operation of the AutoSPADA edge computing platform for distributed data analytics. The platform's design follows scalability, reliability, resource efficiency, privacy, and security principles promoted through mature and industrially proven technologies. In AutoSPADA, computational tasks are general Python scripts, and we provide a library to, for example, read signals from the vehicle and publish results to the cloud. Hence, users only need Python knowledge to use the platform. Moreover, the platform is designed to be extended to support additional programming languages.

Attributes

Technology
Automotive, Information Technology
IT & Software, Research & Development
More Efficient, Saving Cost
Federated, Machine Learning