Smart car safety
Network advanced driver assistance systems (ADAS): Demonstrating next-generation automotive safety features with Renovo, Verizon and AWS
By Jason Stinson, CTO, and Chris McNamara, VP of Marketing, Renovo
Ever wondered how two driverless cars would know to brake before they run into each other? Here at Renovo, our team has been tackling precisely that question — among others.
Renovo is a Verizon Ventures portfolio company and creator of the first commercially available data-management platform enabling automakers to continuously learn from their production vehicles.
Recently, we joined Verizon at the University of Michigan test bed (MCity) to test our platform in a live environment of driverless, wirelessly connected vehicles with advanced driver assistance systems (ADASs). In this article, we’ll talk about our testing journey and how we integrated our platform — not only with the Verizon 5G network, but also with the Verizon 5G Edge mobile edge computing platform with Amazon Web Services ® (AWS®) Wavelength.
In the near future, we’ll see some cars with a mixed complement of advanced driver assistance features and others with nary a camera. How will such a cohort of cars react to the unexpected? Could a gaggle of ADAS-rich vehicles warn each other of hazards ahead?
The first step toward answering this question was to test autonomous vehicles in a proving ground where driverless cars run free.
Proving out network ADAS
At MCity, our teams went to work testing two self-driving car safety use cases via the facility’s 5G Edge connection.
Training sensor data transmission via 5G Edge with AWS Wavelength and the Renovo cloud. This use case demonstrated the benefits of connecting over 5G networks for near real-time edge processing of automotive data using artificial intelligence (AI). A road-legal vehicle streamed filtered sensor data over a 5G Ultra Wideband connection to a cloud-based application portal, through which software developers, data scientists and other third parties received critical, near real-time data.
Offloading sensor data is bandwidth sensitive: You want the biggest “pipe” possible, so all available data is uploaded quickly and reliably, without needing to store it on the vehicle. With the large amount of bandwidth from a 5G network connection, that pipe becomes huge. This enables data scientists to mine a vast amount of vehicle data and reduce feature development time.
Sending basic safety messages (BSMs) to 5G Edge during a high-deceleration (sudden braking) event.
Network ADAS in action: A vehicle with ADAS-generated BSM messages in response to specific combinations of signals or anomalous events, such as unexpected braking, sent the message to the cloud at the network edge for processing. The AI, in turn, alerted other connected vehicles in the area about potential hazards and traffic conditions. Those vehicles then reacted to out-of-sight events faster than a blink of an eye.
This safety-critical case, which wouldn’t be possible without ultralow-latency 5G and ultrafast processing at the network edge, has the potential to impact not just connected vehicles, but also nonconnected ones.
Real-time mapping. A road-legal vehicle streamed filtered data over the 5G network to AWS Wavelength at the network edge, where vehicle location was processed and time-sensitive data and insights transmitted to the vehicle.
In the future, drivers could potentially subscribe to a subset of insights — insights as a service — appropriate to the local area. With ultralow-latency 5G and edge computing, those insights would be relevant, timely and engaging. In a higher-latency environment, not so much.
Key performance indicators. As expected, the two smart car safety use cases tested at MCity showed a significant improvement in latency and throughput over the 5G network using 5G Edge.
Verizon continues to work with Renovo to test use cases on public roads over its 5G networks and on 5G Edge within AWS Wavelength. Future use cases under consideration range from broad (Amber Alert dissemination, vehicle-to-vehicle messaging) to specific (scenario-based data processing).
Renovo’s journey: Data management for advanced vehicles
A single vehicle can generate massive amounts of raw, unstructured and uncompressed data — on average, from two to six terabytes per hour. When collected and analyzed, this data could be used by automakers to:
- Train and fine-tune self-driving AI-enabled vehicles
- Investigate anomalies, model failures and lapses
- Highlight areas for product improvement
- Suggest new features
- Demonstrate regulatory compliance
Yet out of all the terabytes of data generated during a day’s driving, only a few gigabytes are actually useful to automakers for analysis.
This brings up a whole new set of questions, including:
- How could OEMs gather, upload and manage so much data, especially across a globally distributed fleet of vehicles?
- How could they do this quickly, securely and with minimal errors?
- How could relevant data be reliably delivered to data scientists and developer teams?
- Once software improvements are made, how could updates be quickly and securely deployed across entire fleets?
To meet these challenges, Renovo created the Renovo Platform, a scalable, edge-first data management platform that provides high-speed data transfer from advanced vehicles to speed and simplify tagging, searching and querying.
Data orchestration is the automation of data-driven processes from end to end. The Renovo Platform unlocks the orchestration of vehicle data to ingest trip data from one car or thousands. Once ingested, data is indexed, normalized and then enriched using machine learning (ML) to identify objects, scenarios and conditions of interest.
Mission-critical data can be identified and prioritized so it gets to developers quickly. And storage efficiency in the cloud is dramatically improved.
Renovo’s approach to data orchestration
Renovo’s approach to data orchestration via network ADAS is uniquely tailored to automobile OEMs. The Renovo Platform manages data from the point of generation (the vehicle sensor) and provides a variety of features to support all stakeholders whose lives could be improved by access to real-world, near real-time vehicle data.
Data transportation. Using the tools in the Renovo Platform console, customers can set filters that isolate the precise data of interest and a sampling rate for that data. Signals can be prioritized to take advantage of network conditions, whether the network is Ethernet, LTE or 5G.
Custom workloads. The Renovo Platform tools can be used to design and run custom workloads to dictate which data to store and transport. For example, a workload that isolates only hard braking events, or one that identifies ADAS disengagement events.
Data science Workbench. The Workbench provides a query tool that enables data scientists to query their fleet’s database directly; to launch Grafana® to chart and visualize fleet data; or to perform analyses and visualizations on fleet data using a Jupyter® hub.
