Driver Score - Why, What and How?

Why do we need it? πŸ€”

Zoomcar runs on cars. We have 10K cars running Pan India with over 6 million delighted customers. One of the major challenges we face is servicing and maintenance of our cars. With our fleet scattered all over the country in different geographies, maintenance becomes even more difficult.

In general, commercial vehicles have higher service and maintenance cost than personal vehicles. To give you an example, A personal Maruti Swift has almost 12K per year SAM cost. On the other hand, a commercial Maruti Swift incurs almost 70K per year.
That is how much the difference is 😲.
Moreover, the type and age of the vehicle impact the maintenance cost significantly.


How did we solve this problem? πŸ‘©β€πŸ”§

To overcome this challenge, we created a driver scoring system where we measure the driving performance of the driver. For each trip we measure the driver score on a scale of 1️⃣0️⃣0️⃣ (yeah, like school exams)

You can book your Zoomcar and get your score right now. The scores help us measure the health 🩺 of the car. Better scores mean, smooth treatment of vehicle and vice versa.

  • More time spent hard accelerating/ hard decelerating = more vehicle deterioration
  • More time spent idling (engine speed>0, vehicle speed = 0) = more vehicle deterioration
  • Less time changing gears = less time riding the clutch = better clutch health

So, how does it work? πŸ› οΈ

Let us dive into the driver scoring system.
The first question that comes into mind is: How do we say that someone is driving good or bad. Once we can distinguish between good and bad behavior then scoring becomes a trivial job. To do this we took inspiration from real life and of course our millions of happy customers πŸ₯³.

Let us take the simplest example, If the driver drives smoothly and co-passengers sitting on shotgun or back seats can sleep πŸ’€ without moving an inch, then the driver is driving great. On the other hand, if passengers are getting frequent jerks during driving, there is sudden acceleration, braking or a sharp turn out of nowhere at high speed then we call it bad driving πŸ€’.

Although, there are external factors that affect the driving behavior, like bad traffic, rain, sudden movement on road etc. We took care of πŸ€— all these situations while designing our driver scoring service.

Now that we understand what good / bad driving is. What’s left is to measure this behavior through a mathematical model and then map the behavior to score.


So, let’s do that. 🏁

As discussed in the last section driver score is all about translating driver behaviors into score.

There are 2 steps to driver score

β€’ Measuring driving behaviors
β€’ Transforming driving behaviors into score


Measuring driver behaviors πŸ”¬

Let’s start from the fundamentals. To measure any event or object there are established processes.

For object measurement we use the existing methodologies and units like Meter, Kilogram etc βš–οΈ.

For event measurement however, we perform experiments and look at the evidence πŸ•΅οΈβ€β™€οΈ.

Pretty simple so far. Detecting driver behaviors falls in the second category where we need to look at the data and subsequently interpret the behavior.


Let’s talk about the data πŸ“‹πŸ’­

Cars are one of the most fascinating pieces of machinery right now. They are living breathing organisms 🦠 (metaphorically). It is like a human where each part/component has unique responsibilities and all of them have to function together for a smooth ride.

Let’s get down to the parts of the car. Most obvious ones are clutch, brake, accelerator, engine, tires, battery etc. Apart from these there are thousands (literally) of other parts.

To measure the driving behavior, we’ll be taking help of data coming from these parts. To give you a rough estimate we get close to 200 million data points from our cars each day.

This data includes vehicle speed, engine speed, acceleration, brake etc. you get the idea πŸ’‘.


How do we measure driving behavior? πŸš™

All set now, we know the objective, we have the data, the only thing remaining here is to measure. Let’s do it. πŸ‘Š

Going back to our earlier discussion when we distinguished good driving from bad driving, we talked about attributes of driving like, lots of jerks, uneven driving being bad while smooth, less jerky driving being good.

To measure this smoothness of driving we have created an index (it’s just a number) which measures how much a driver is deviating from the ideal driving behavior. Lesser the deviation, smaller the index and smoother the drive.
That is, it. We have measured it. πŸ™Œ


For example, there are 2 drivers and here is their driving behavior:


As you can see Driver 1 is driving smoothly. Whenever she has to increase and decrease speed, she is doing it smoothly. On the other side, there is a lot of variation in Driver 2’s driving. Speed is irregular and there are more instances of speeding and braking. As a result, Driver 1 will have a better score.

Now to be more accurate and fairer to our customers, we take care of external factors on which the driver does not have any control like car type, condition, sudden events etc.

You must be thinking all this index and smoothness is good but what about obvious events that you can visibly see happening like sudden braking, over speeding etc 🀷.

They are taken care of, all the sudden events are recorded and you might see a notification or a voice message from us to nudge you about the same πŸ–– (yes, Vulcan).

Here we are done with the measurement of the driving behavior. Let’s have a look at the scoring process.


Transforming driving behavior into score πŸ€–

Once you have the behavior mapped out in numbers as we discussed in the previous section, allocating score is straightforward.

Smooth drive leads to less variation (index) and higher score. The rasher one drives, higher the index and subsequently lower the score.

To elaborate on events, harsh events of braking, speeding acceleration lead to penalty in your score and vice-versa.

Now, you might be asking what happens if you are in traffic and you must apply brakes continuously when suddenly an animal pops up out of nowhere. The scoring system takes care of such events by normalizing these events with your overall score.
So, if you are driving well and you apply brakes suddenly, do not worry you won’t be penalized for that.

πŸ€– (Beep bop: it’s a kind hearted machine)


Tying it all together 🍻

We started with the problem of significantly higher service and maintenance cost of commercial vehicles. To solve this problem, we created the driving scoring system which measures driving behavior. Health of the asset depends heavily on driving behavior. So, by tying the scores back to the vehicle we can estimate the health of the asset.

We keep the drivers informed about their scores, by nudging them to do better driving, we drive our service and maintenance cost down πŸ’°.

So go on and book a 🚘 Zoomcar πŸš™ today to get your score and btw, there are rewards too for higher scores πŸ….


Driver Score Service

Contributors


Building immersive AR experience with React Native

User Engagement: