Cloud computing, which is already invaluable to nearly every aspect of business infrastructure, has also become integral to the development and deployment of software-defined vehicles. Amazon Web Services (AWS), one of the top public cloud providers, has been staking out a leadership role in providing cloud services tailor-made to the needs of automotive, including advanced driver assistance systems (ADAS), autonomous driving, fleet management, AI/ML, and vehicle data management. They are providing services ranging from compute, data management, vehicle prototyping, and much more.
The cloud is also important to us: Sonatus delivers cloud software that takes advantage of these capabilities in our products. This blog will explain why the cloud is so critical to automotive, and in particular some of the ways AWS is leading.
Cloud data management
The amount of data produced by vehicles is skyrocketing, making it more and more complicated to capture, store, upload, and manage. While limited vehicle telematics and sensor data have been around a while, data signals are rapidly increasing as more vehicle systems are becoming digital and connected. Adding to that, the proliferation of rich data sources like video cameras, Radar, and LiDAR are expanding, delivering more advanced capabilities but generating significantly more data in the process. Leading-edge automotive OEMS are taking advantage of this data, creating integrated “data lakes” to form a single source of truth that can be queried in different ways from across their organization to add intelligence and ensure consistency in decision making.
Gathering data from vehicles produces a wide range of benefits. Here are a tiny fraction of the countless interesting use cases of vehicle data that can benefit from the cloud:
- Safety monitoring
- Fault diagnostics
- Ongoing component optimization
- Efficiency tuning
- Vehicle usage analysis
- Future vehicle planning
Advancing the ability to collect valuable, carefully selected data from vehicles is critical. Sonatus Collector is in production today and creating useful databases for our OEM customers that allow them to solve real problems and add customer value
Cloud Computing
One of the most fundamental benefits of cloud computing is the ability to flexibly expand compute capabilities based on variable demand, giving rise to AWS’s naming of this service EC2 for “Elastic Compute Cloud”. Based on the needs of the business, compute instances can be activated and put to work quickly – nearly instantaneously, if needed – while not consuming capital nor operating costs when not in use. This is an incredibly powerful tool for businesses and has changed the way businesses use compute.
A second equally-valuable benefit is the incredible diversity of compute instance types that provide a wide range of compute, storage, memory and specialized processing capabilities. Given the growing shift to Arm-based compute in vehicles, AWS offers cloud instances based on Arm-based AWS’s Graviton processor, enabling “environmental parity” between vehicles and the cloud. Other instances specialize in machine learning training and inference, which I will cover later in this blog.
Sonatus products across our Sonatus Vehicle Platform comprise in-vehicle software and corresponding cloud software that is built on and leverages the full capabilities of cloud computing, including Kubernetes and containerized workloads, based on Amazon EC2 and Amazon’s Elastic Kubernetes Service (EKS).
Vehicle Virtual Prototyping
Another important application of the cloud is rapid prototyping. Developing software for vehicles is challenging given its deeply-embedded environment with multiple interlocking systems. Moreover, time-to-market pressure is pervasive, so it is incredibly beneficial to be able to develop software in parallel with hardware. This is sometimes referred to as “shift left” for allowing software development to come earlier in the cycle.
Offering cloud instances with compatible architectures to vehicles can significantly improve developer efficiency and speed time to market by allowing development in the cloud and deployment later to vehicles. This design style can also promote collaboration across teams in multiple geographies, who may be addressing different regional requirements. In fact, it even allows virtual prototyping and software development while the hardware is still being designed.
The most capable teams use this approach to debug the hardware before it is finished, which has been proven to significantly speed time to market and reduce hardware iteration cycles to fix bugs. This prototyping approach also has additional benefits as OEMs are increasingly shifting their vehicle architectures away from dedicated ECU’s carrying out a single, fixed function to consolidated architectures with multiple cores side by side and managed through hypervisors and virtualization. That design style is more scalable, easier to verify, and matches the expertise of the cloud for prototyping and later for production deployment into vehicles.
Sonatus is using AWS to prototype our own work to increase the number of test cases we can run without being forced to replicate physical hardware or repeatedly power-cycle it to restart a test.
AI / Machine Learning and Data Analytics
Once we bring together a compelling data set and combine it with scalable, flexible compute, it unlocks the ability to do incredible data analytics that would never have been possible before. Add to that the rapid expansion in machine learning (ML) capabilities and even more powerful questions can be asked of data. ML algorithms are continuously evolving and improving, and cloud computing approaches allow rapid training of the models in the cloud. In particular, as the industry works to develop ML-based models for ADAS and Autonomous driving, the cloud model can be incrementally deployed to vehicles with feedback to ensure the new models do not cause regression.
Another of ML’s strengths is the ability to detect patterns and anomalies that would be difficult to code using conventional approaches. By studying data from a large number of vehicles, patterns can emerge that can signal early warning signs. For example, smart data collection coupled with ML can be used to proactively detect anomalies and flag them for corrective action before they become dangerous recalls. Sonatus OEM customers are already using Sonatus Collector to anticipate issues and better respond to service needs for their customers.
ADAS and Autonomous Driving
The advent of ADAS and its higher-capability cousin, autonomous driving, requires significant modeling and tuning, and the cloud is critical to these advances. The complexity of these tasks cannot be overstated, especially the higher levels of autonomy, and it is only through cloud computing and incremental model improvements that make the rapid improvements in these capabilities possible.
First, OEMS can capture and accumulate data from real vehicles or data capture vehicles to build a database that is representative of real driving tasks to stimulate the recognition and action models. The benefits of this approach do not stop there: most of the time, normal driving is “boring” with few difficult scenarios occurring per drive. In a simulated environment, we can artificially push through corner cases at an accelerated rate and make far faster training progress compared to only training based on vehicles in the field. The Sonatus Vehicle Platform can be used to capture rich data to assist in the important analysis needed for ongoing improvement of ADAS and autonomous driving, adding a new powerful tool to the engineers doing this important work.
If you found this topic interesting, you can learn a lot more: I recently had a long, two-part conversation with Stefano Marzani, the Worldwide Tech Leader for Software-Defined Vehicles at AWS for our podcast, The Garage. In that wide ranging discussion (links to Part 1 and Part 2), we touch on all of the topics above in incredible detail. If you’d like to learn more about this topic and hear from one of the leading voices in the industry on SDV, don’t miss those two episodes. You can find the full The Garage podcast on YouTube or on Spotify or Apple Podcasts. Be sure to subscribe to be notified of future episodes!