Making Roads Safer by Making Drivers Better

By July 26, 2016Blog

safe-driverIt was the summer of 1995. Five of us, all students at Berkeley, were driving on 101 South to Daly City (south of San Francisco). We were heading over to play a Cricket match in the Northern California Cricket Association league for the Berkeley Cricket Club, a pastime I happily engaged in while in research-avoidance mode in grad school. I remember it was chilly and a typically foggy day — wasn’t it Mark Twain who said that the coldest winter he spent was a summer in San Francisco? — par for the course, with nothing amiss. Suddenly, without warning, my friend driving the car lost control, the car swerved, spun around twice, hit the center barrier hard, and came to a stop straddling two lanes.

Miraculously, and luckily, no one was hurt. We were dazed and stunned… and extremely lucky to be alive. I was in the middle of the back seat, and, for some reason, not wearing a seat belt. It could have been much worse.

Millions of people are less lucky. Far less lucky.

    • The World Health Organization estimates that 50 million people each year are injured or disabled in road crashes.
    • Crashes caused nearly 1.25 million deaths last year, 3400 each day.
    • Crashes are the leading cause of death among young people between the ages of 15 and 29, and the second leading cause of death among children between the ages of 5 and 14.
    • Today, crashes are the 9th leading cause of death worldwide (across all ages), but if technology and medicine continue to advance at present rates, they will become the 7th leading cause of death by 2030.
    • Crashes are the leading cause of death for US nationals traveling abroad.
    • Most crashes — about 88% according to a recent naturalistic driving study — occur due to driver error or impairment, rather than due to vehicle failures or faulty roads.

Our mission. All these sobering facts bring us to our mission at CMT: to make roads safer by making drivers better. For decades, society’s approach toward driving safety has been one of penalties (“sticks”) and restraints. Fines for speeding. Penalties for breaking laws. Fines for not wearing seat belts. Penalties for drunk driving. Higher insurance premiums if one is involved in a crash. And so on. Most people will agree that penalties are essential, and some of these may need tighter enforcement (e.g., drunk driving), but the data on crashes, injuries, and fatalities indicates that these methods aren’t reducing crashes quickly enough. In fact, for the first time in years, crash-induced fatalities have risen in the United States over the past 1-2 years.

In 2010, a few months after we started CMT based on research from the CarTel research project at MIT’s CSAIL, we felt that a different, complementary approach involving incentives and positive reinforcement (“carrots”) ought to be tried to make drivers better. Over the past few years, working with a great set of partners around the world, we have pioneered a new area of driving safety using behavioral incentives. The solutions we have developed are now deployed to a fast-growing set of users via our partners: insurance companies (for traditional usage-based and our pioneering behavior-based insurance programs), cellular service providers, automobile makers, and governments. We call our program DriveWell, and we bring it to users via different mobile apps. The picture below shows our current global footprint, with users in all populated continents.

image-2

Our belief then was that safe drivers are made, not born. This hypothesis has now been substantiated with over 1 billion miles of real-world data from deployed programs in many countries. We are on a meaningful journey to make our roads safer by developing programs with our partners that combine technology, economics, and policy. The elements of this program include technological advances in three areas:

    • Mobile sensing and the “Internet of Things” (IoT): acquiring data from position, inertial, and other sensors on mobile devices efficiently
    • “Big data” analytics and inferencing using machine learning and statistics
    • Behavioral economics to convert data into insights and measurable actions.

Why this blog? It appears to be Marketing 101 for companies these days to set up a blog. We did not want to do that as a mere marketing exercise, and so didn’t start one for the 4+ years we have been in operation with real users (our first users came on board in early 2012). But now, thanks to our large user and partner base, we have many interesting things to say about the state of safe driving and its future. We want each of our articles to say something unknown before. We plan to share with the world ideas, data, and insights on topics such as:

    • What behavioral incentives have worked, and why; what improvements in driving quality have we observed in the field?
    • Which towns and cities have the best and worst drivers as assessed by our scoring algorithm? Who improves the most?
    • Does a behavior-based insurance program reduce the number of crashes and reduce the severity of claims? And if so, by how much?
    • How smartphones – rightly implicated in a number of traffic crashes – are actually also making driving safer.
    • How bad is phone-induced distraction and how to make it less so?
    • How IoT, sensing, and “connected vehicles” are changing safe driving.
    • How to build a real-time crash detection and notification system to bring roadside emergency assistance to a crash scene quickly.
    • How does driving performance vary between different states and cities in the US, and between different countries and locations in the world? Who has the best and worst drivers? Whose drivers are improving the most?
    • How to make accurate inferences from noisy sensor data, prone to outages and other errors? For example, accurate map-matching, handling map errors, acceleration inference, etc.
    • Why do sensors consume a lot of battery energy, and how to cleverly harness them in battery-efficient mobile sensing?
    • What are the challenges in classifying transport modes to ensure that only legitimate driving is measured?
    • How to build a scalable, distributed system to process billions of sensor data points per day.
    • War stories: what components can fail in an end-to-end telematics system, and how to overcome them?
    • . . . and many more.

A sample finding. To whet your appetite, I share below one result from EverDrive, a consumer app we launched recently with our partner, EverQuote. The EverDrive app is available for iOS and Android. It provides easily-digestible information about your driving quality, social leaderboards to compete with people in your town and state, and a way to invite friends and family to compete with for bragging rights. From time to time, the app also has a driving contest: we recently concluded the first one, Massachusetts v. New York, which ran for two weeks between June 12 and 25, 2016. The app has users in all 50 US states.

The graph below shows the mean “risk points” across tens of thousands of users for three of the factors used in assessing driving quality: hard braking, unsafe speeding, and phone distraction. The graph shows each user’s “day 1” score normalized to 1. The x-axis is the number of days into the program; the y-axis is the risk points for each factor. Lower means safer driving on that factor. Across all users, the risk points fall as the number of days in the program increases, and for the best users (top 20%, who have a score greater than 95 out of 100), the drop in the phone distraction factor is especially noteworthy. But even across the board, for users with scores > 50 (98% of all users), the fall-off in risk points is strong. This result suggests that people are indeed becoming safer drivers.

Result-Graphs
I hope you will stay tuned in the coming days and weeks as we tell you these stories, backed with data, insights, and new questions. And I welcome your comments on the posts from our talented and dedicated team.