A New Continuous Development Toolchain for Autonomous Driving
The automotive industry is committed to building safer, smarter cars. This is a continuous, iterative process, with new Advanced Driver Assistance System (ADAS) features constantly moving the world toward an increasingly safer, more efficient autonomous future.
The Society of Automotive Engineers has outlined six levels of autonomy (SAE levels 0-5), spanning from zero autonomy to full autonomy. According to a new forecast from International Data Corporation (IDC), vehicles with some degree of autonomy will represent more than 50% of all vehicles produced by 2024. The most advanced cars on the road today are SAE Level 2 and some are beginning to enter the market with level 3 capabilities. Frost & Sullivan predicts that by 2030, 67% of vehicles sold globally will have at least level 2 and 3 autonomous driving capability.
Driving this massive shift are vehicle OEMs, Tier 1 suppliers and mobility providers and the data they collect and analyze from test fleets around the world. The leaders in this space are optimizing the way they ingest and manage this massive amount of data with infrastructure that meet current and future demands of ADAS and autonomous driving (AD) test and development.
Developing new ADAS (SAE level 3) features today typically requires tens of thousands of CPU cores and 50 to 100 petabytes of storage for the sensor data that was recorded. While that seems like a lot of data, the higher levels of autonomy will require an exponential growth in data. As ADAS features evolve from simple collision-avoidance systems to fully autonomous vehicles, these systems will combine complex sensing, processing, and algorithmic technologies – necessitating the need for even more recorded sensor data. This vehicle-generated data is a critical component to improving AD/ADAS systems, feeding into integrated test and development cycles (or development tool chains) for these systems.
Frost & Sullivan maintains that before connected and autonomous vehicle providers can leverage the data, they must first create a system that is flexible enough to handle challenges such as:
- Future-proofing ADAS simulation and architecture, to adapt to changes in vehicle sensors and other environmental data points
- Managing data storage to comply with regulatory and privacy requirements, while addressing security and accessibility needs
- Analyzing massive volumes of unstructured data sets, to support analytical modelling and querying of ADAS data. This requires costly and time-consuming data preparation steps, such as labeling data for analysis.
The challenge is to create future-proof architectures not only in the vehicles, but also in the datacenter. Automotive companies require infrastructure that allows them to leverage ever-growing sensor data to build, test and continually improve data models. Data and the infrastructure to manage it will become key differentiators that determine success in the automotive market.
To assure this success, it is important to reduce the complexity of the system for developing ADAS and AD functionalities, and to reduce complexity of development, for example to reduce the amount of needed Electronic Control Units (ECU) from 200+ to a fraction of that. Doing this will require end-to-end development solutions that can do everything from collecting data, to managing data, to developing algorithms that can be ported and deployed in an ECU inside the vehicle, to testing and validation, to hardware in the loop testing.
The problem many automakers run into when developing new AD/ADAS features is an incomplete toolchain that does not support continuous integration, continuous delivery, and continuous deployment. Very often, some small pieces of this tool chain are deployed while a real end-to-end solution is missing. Seeing the bigger picture is vital to ensuring infrastructure investments are contributing to a larger, long-term strategic vision.
The Solution: A complete autonomous driving data lake reference architecture
Our new Dell Autonomous Drive ecosystem supports the most important steps in the AD/ADAS data process. Developed in conjunction with leading industry and technology partners, Dell Autonomous Drive combines Dell Technologies and partner infrastructure, software and services to offer a complete end-to-end toolchain.
Figure 1: The Dell Autonomous Drive Ecosystem. This drawing does not represent all the connections or equipment required for a complete solution. It is provided as a high-level overview
The Dell Autonomous Drive end-to-end AD/ADAS development toolchain:
- speeds the development process, enabling a company to get new features to market faster.
- makes development easier for data scientists who otherwise spend 70% of their time on data management, including locating and preparing data.
- incorporates mechanisms to save costs by reducing the amount of data needed to train algorithms.
- enables automakers to focus on more advanced capabilities such as over-the-air updates.
- helps with managing standards and regulations.
- allows data scientists the option to include more open source software as a service and manage open source software throughout the development lifecycle.
The Dell Autonomous Drive ecosystem is an open alternative to a single public cloud data lake which can reside on-prem, in the cloud (public and/or private), or any combination of the two. The solution is a complete autonomous driving data lake reference architecture and workflow. This ecosystem helps avoid the unpredictable and unmanageable costs of utilizing a single public cloud solution.
This new ecosystem includes Dell hardware and software including a platform for analyzing streaming data; PowerScale unstructured data storage combined with PowerEdge CPU and GPU servers provide performance that scales and adapts easily; a data management system streamlines the process; and Dell Technologies Multi-cloud keeps options flexible, making data simultaneously available to multiple public cloud providers, eliminating vendor lock-in and reducing costs.