A Full-er Stack: Building a data pipeline inside and outside a vehicle

The Ather 450® smart scooter has about 46 sensors onboard which continuously generates data points about different components and parameters on the scooter. These data points are used to continuously assess the riding patterns and behavior, to provide various insights into your ride such as ride statistics, predicted range, auto cut off indicators and many more. On the backend, these data points are collected and fed to various systems. This article talks about the data pipelines behind the scenes and what goes on in making a scooter intelligent.

We had one of the earliest owners report that there was inconsistent charging experience with the vehicle. After a service visit to the location, which found nothing amiss, the software team began to debug the issue based on charging data. The analysis found that the inconsistent experience was only after 9 PM. The service team then visited the location in the night to find that the owner was using a device only at night, which caused the line-neutral voltage to spike. This experience led the team to develop an algorithm to minimize charging issues, which was then deployed as an OTA.

Curious how this was possible? Read on to find out how data pipelines on an intelligent connected scooter work!

Scooters on the road have more constraints than a regular system that works with data. Unlike phones, tablets, and computers, scooters are used for a much longer duration without needing a hardware upgrade, and scaling up the hardware is usually very difficult. So how do you build a data-intensive intelligent scooter with the constraints of automotive-grade embedded hardware? Read more to find out.

What’s inside an intelligent, connected scooter?

The Ather 450 electric scooter has two kinds of features - critical and non-critical. Critical features are those which affect rideability and safety, and non-critical features are the ones that affect user experience.

Lithium-Ion batteries used in the current generation of electric vehicles are complicated and the world is still trying to understand how exactly they work. Batteries are known to vary the power output depending on the charge levels. Maintaining the same consistency in performance is important for regular usage of the scooter, but at the same time, the battery performance has to be maximized. In case there’s an adverse scenario such as the battery pack or motor overheating, it is critical to be able to manage performance to make sure you can get home, but at the same time make sure that the components are not getting damaged in the process. Some of the critical features that are being tracked include the vehicle states (ride modes, low power mode), component health and uptime monitoring, battery state of charge, and control of peripherals such as the headlights, indicators, etc.

Non-critical functions include auto indicator off, navigation, trip/efficiency meters, range prediction, cloud connectivity, and data sync.

These data points enable a few clever features. Have you noticed how when you’re navigating, the cursor is always on a road or pathway and never in the middle of a structure? The software team worked really hard to provide this experience. GPS coordinates are not precise and often are off by a few meters. Since navigation on a scooter is most likely to be used on a road, the GPS data is combined with the speed to make sure the navigation is a good experience.

The electronics and sensors along with the motor controller, charger and the Battery Management System (BMS) feed data to the dashboard continuously, which is then sent to the cloud.

How do you diagnose a problem in a scooter?

Regular scooters on the road are not inherently smart. They’re all mechanical parts that are put together without any monitoring once they’re on the road. If you notice an issue with your regular scooter or bike, it’s usually because of unusual sounds, vibrations or the ‘feel’. And it’s up to you to prove to the service center that the issue exists and to be able to reproduce the issue.

How can data help transform the ride experience?

Having all these data points helps build a superior and connected experience for owners. The data points are being used for component health and uptime monitoring, predictive maintenance and a ton of features such as range prediction, ride statistics, location monitoring, navigation, charge level monitoring, time left to charge and service diagnostics. This helps reduce the turnaround time in service and reduce the time to diagnose and resolve issues.

Further, there is an IMU (Inertial measurement unit), which tracks everything from the lean angle to how fast you’re going, the orientation of the vehicle and more. The IMU enables a lot of smart features such as the auto cut-off indicators, better navigation, and opens up a lot of possibilities for features in the future. Using multiple data points such as the lean angle, change in acceleration, orientation and more, the turn indicators go off automatically when you’ve made a turn.

Interaction with hardware brings its own issues with data sanity. There are data sanity checks to ensure that there’s clean data at all the processing endpoints.

Are cloud-connected automobiles viable in an era of patchy networks?

The whole data pipeline was designed keeping areas with patchy networks in mind. To start off, the data is always stored locally on the device. This enables the dashboard to sync the data when the network is available again. Even among the data sync, some data points are prioritized over others. For instance, if you’re parked in a basement, it is okay to get the ride stats a little later on the mobile app, but the location and the charge level are prioritized and synced right away.

Critical features on the scooter warrant their own hardware that handles all the crucial vehicle functions. These handle the vehicle state, peripheral control, and communication with other components. These components themselves are continuously monitored for uptime. Since these are critical functions, the data processing capabilities are minimal.

The main processor meanwhile, handles all the data and performance-intensive tasks such as the display, cloud connectivity and all the number crunching. Using an automotive-grade chipset has its advantages in terms of the reliability and stability of the components, but comes with lower processing and storage capabilities. This makes CPU and memory management critical for the stability of this system.

The data system also has a few nifty workarounds to make sure the data is available when you need it the most. Vehicle state data and component uptime are critical to diagnose vehicle issues remotely and to enable features such as predictive maintenance. Reducing the size of the data is important and hence data-intensive tasks are performed on the device if possible. The system also tries to send data only on change, rather than at regular frequencies.

Have you seen any benefits from the data already?

In several cases, the availability of these data points has helped accelerate the diagnosis and the remedy to issues. Older houses in Bangalore and Chennai are notorious for having poor grounding or irregular electricity supply. Very early on, data on charging patterns helped identify cases where the excessive load on the electricity supply grid caused issues with the charging point. This was quickly diagnosed and rectified to be more tolerant in a subsequent update. Many other bugs were also quickly identified before riders started experiencing them, based on patterns in the data collected.

If you’re curious how Ather Energy® leverages the cloud to enable these data processing capabilities, Google Cloud has put together a case study on this: https://cloud.google.com/customers/ather-energy/

What are the possibilities in the future?

Having these data points from one or two scooters might not be significant, but having data from thousands of scooters on-road, over a few months can generate incredible insights about riding in a city. The data collected from these scooters allows an intelligent layer to be added to your commute, where we can possibly map out routes in a city that have potholes or have very uneven surfaces or consider gradients and actual conditions on the road to give a more accurate estimate of the range.

Questions? Post your queries below and @chaitanya.hegde will be available and answer them!


A post was merged into an existing topic: Sentry mode