by Jon Soon July 09, 2018
As Vice President of Engineering, Annie Cheng is among an elite group of women engineers leading a transportation tech team. Cheng brings a background of computer science, information networking, engineering, and artificial intelligence (AI) expertise to Nauto, balanced with practical experience. Cheng’s parents groomed their first born to take over the family-owned, small business. At an early age, Cheng’s curiosity about science and technology led her away from the business to working on technologies with real-world impact.
One focus for your team is to detect distraction and assess situational risk in real-time. Can you explain how Nauto uses artificial intelligence (AI) to do this?
Cheng: Our approach to AI is unique in that we’re applying AI at two points: at the edge and in the cloud. The combination gives us an intelligent, closed-loop driver safety platform.
Our challenge is different from many IoT environments that are connected over stable networking environment. For stationary devices, which operate in predictable networking environments, AI can be run in the cloud. We don’t have that luxury because vehicles are on the move, so our devices need to operate in variable networking conditions. This creates an additional challenge, as our devices need to function regardless of the network connection, which means we have to intelligently choose what runs on the device versus the cloud.
See Nauto's intelligent driver safety system in action
When we detect distraction or other high-risk events, we need to be able to alert drivers in real-time to help coach and stop risky behaviors, and simultaneously, upload these high-risk events to the cloud to ensure that fleet and safety managers get the right data at the right time — even when we aren’t guaranteed a well-connected environment. We’re able to do this by leveraging AI to assess and understand the driver in the cabin and road ahead. For example, inside the vehicle, AI is integrated into proprietary algorithms on the device that we use to assess driver behavior. One example is our distraction algorithm, which processes incoming images in real-time to detect whether or not driver distraction occurred.
Outside the vehicle camera, we use machine learning and deep learning technologies to not only detect objects and understand overall context. By using AI on the edge, we’re able to operate reliably in any environment, regardless of networking connection conditions.
What else is happening on the device that makes Nauto unique?
Cheng: We also use proprietary algorithm on-device to detect hard acceleration, braking and cornering (also known as ABCs), and collisions. Because we have visual modality, for near misses or sudden harsh braking, we have the capability to bring in context—both inside and outside the vehicle—to understand what was happening the moment distraction or another high-risk event occurred.
What’s done on the cloud versus the device?
Cheng: When distraction or other high-risk events occur, we upload these events to the cloud, where they’re automatically visible to the fleet and safety managers of a given fleet via a secure web application. This data is aggregated and filtered according to the event, whether it was a collision, driver distraction, or tailgating. Nauto then synthesizes the data into actionable insights for operations and safety leaders, so they can identify high-risk drivers and effectively coach them to improve performance with the full context of the road ahead and the driver’s actions behind the wheel.
Because we have more compute flexibility in the cloud, that power enhances and complements the AI on our devices. Also from the cloud, we have an aggregated view of high-risk driving behaviors. Our machine learning and deep learning models “learn” from this aggregation. The more Nauto-equipped fleets, the more we’re able to add to our aggregate data and therefore more accurately detect distraction and other high-risk events.
Where do you think Nauto technology will have the most impact?
Cheng: The fact that our technology is both on the edge in the device and in the cloud provides a closed-loop safety pipeline. By using AI on the device, we’re able to detect distractions and immediately warn drivers in real-time without being intrusive to drivers. We believe that using AI to ensure the safety of drivers shouldn’t be at odds with protecting their privacy. That’s why only a fraction of driving time—such as distractions or other high-risk events—is uploaded to a secure web application in the cloud, where it’s automatically visible to fleet managers via a secure web application. In doing so, we’re able to provide the safety insights that matter most, while ensuring the privacy of drivers. With this knowledge of driver behavior, we can also apply these learnings—in partnership with OEMs—to the development of safe, effective autonomous vehicles.