You’re driving to work, the same route you take every day. But this time, a storm passes over: you’re suddenly faced with heavy rain and reduced visibility.
All of a sudden, the “accident score” meter on your car’s dashboard moves into the red. You ease off the gas, move out of the passing lane and your score drops down to amber—you can’t get it into the green due to the adverse weather conditions—and it stays that way as you finally pull safely into the parking lot.
Such accident-prediction systems may be coming to the next car you buy. Arm has joined forces with automotive technology innovator Pioneer to explore how artificial intelligence (AI) technology can be used to reduce accident rates using a technology called real-time predictive modeling. Our goal? Help drivers to modify their behavior, and—we hope—reduce accident rates.
What’s at the heart of real-time predictive modeling?
Advances in predictive modeling and predictive scoring—in retail customer personalization, fraud detection, finance, and insurance—are stimulating heightened interest in using AI in accident mitigation. Accidents are also an important emotional issue, both for their impact on victims and the feeling that many citizens have that more needs to be done to reduce the problem.
While fully autonomous vehicles are at least ten years away from mass deployment, advanced driver-assistance systems (ADAS) are likely to grow in complexity and capability as vehicles become more capable of monitoring and analyzing data in real time. The ADAS technology built into current-production vehicles generally combines data from the vehicle’s drivechain and safety systems with visual detection of objects and markings, such as lines painted on the road.
Customer data helps calculate individual scores
Professor Kazuya Takeda, a well-known expert on analyzing accident data and part of the team that developed Pioneer’s AI-driven ADAS ‘Intelligent Pilot’, has found that the factors most highly correlated with accidents include a variety of behavioral data specific to each driver—such as reaction time, age, overall approach to driving, and other individual influences.
Environmental factors such as weather changes can dramatically up the odds of a collision, as rain, snow or obscurants such as smoke or fog block views, and water or ice increase stopping times. In addition, limitations in the streets and driving surfaces themselves can create a risk; some roads are better designed and maintained than others. And even the most skilled drivers can be a little “off” in their reactions if they’re distracted, angry, or just having a bad day.
Combining data sources for an accurate picture
Capturing all of these variables in order to make decisions based on predictive modeling that positively affect the outcome of a journey is no mean feat. Yet Pioneer, using Arm technology, is looking at how it can be achieved – and in real time.
This level of real-time predictive monitoring requires leveraging many different technologies, including data management of sensors powering the Internet of Things (IoT) and AI. Pioneer is exploring this intersection of converging data through Arm Treasure Data’s enterprise Customer Data Platform (CDP) with Pelion Data Management. This platform crunches data from multiple sources in order to provide easy predictive customer scoring.
Using this technology, Pioneer has developed “YOUR SCORING,” an in-car display feature that shows drivers a real-time, constantly changing estimate of their accident risk using predictive modeling. The score is based on external map and road-condition data as well as the behavior of the driver. By combining and analyzing these data sources to create an overall picture of a journey in real time, YOUR SCORING is able to help drivers reduce their risks.
Arm’s AI and data management capabilities enable Pioneer to quickly gather and analyze numerous data feeds (such as street layout, traffic signals, telematics and third-party data) and apply machine learning (ML) techniques to deliver a single score through the system. At any given moment, the system could be processing reaction time data from the driver, factoring in previous driving history and combining this information with terrain and current weather data from the nearest IoT sensors.
A personalized AI co-pilot
But while arriving at an accurate accident-risk score is one thing, delivering this real-time, potentially life-saving information in the most well received, actionable way is another important technical feat as well. Data plays a major role in this respect as well, and displays could be the next frontier for personalization.
Research has shown that people vary quite a bit in how they react to different types of displays and instructions. A simple thing like the design of the risk display can affect some people’s willingness to take instructions. These displays and user interfaces can now be personalized to people’s preferences and refined by the AI’s monitoring of a driver’s compliance as the display is varied.
For example, preference for one gender over another in existing voice navigation systems is often subject to complex individual and cultural preferences. Stanford University professor and BMW consultant Clifford Nass famously shared that at one point, BMW recalled an early German version of its female-voiced navigation system when it found that many German men showed a strong preference for male voices. In other areas, female voices are preferred for some types of assistants, while male-voiced systems are preferred for others. The next generation of driving technology might be individually tailored in many dimensions to meet such individual customer behaviors and preferences.
So, if you’re hearing a lot from the backseat drivers in your life, stay patient and calm. Someday, they might relinquish their roles to a smarter—and potentially much less annoying—AI. And unlike your current backseat chorus, the AI won’t ask “Are we nearly there yet?”—because it will already know.