Case study

Delivering smart real-time suggestions to a drivable prototype, to raise the feature adoption of ACC among the drivers.

Industry: Automotive

Customer: German car maker

Delivering smart real-time suggestions to a drivable prototype, to raise the feature adoption of ACC among the drivers.

Industry: Automotive

Customer: German car maker

Competence domains:

  • Product ideation & strategy
  • Data analysis & AI solutions
  • Cloud architecture & Big Data processing
  • Integration & deployment into production systems


A significant percentage of newly developed in-car features are under-valued and under-used by drivers. Our customer wished to address this by providing smart suggestions to drivers to increase the use of these features during the drive. The scope of this project included the introduction of one feature to drivers, namely Adaptive Cruise Control (ACC). To ensure the correct application and personalization of each suggestion, the client decided to involve CloudMade’s machine learning toolbox.


Areas of responsibility

CloudMade deployed the following components into the customer’s cloud:

  • A pre-learned crowdsourced map of ACC-friendly zones (see “ACC Fleet Learning Map” below);
  • An ACC recommendation flow chart accompanied by mockups of the head unit widget screens presenting ACC suggestions (see “Intelligent UX” below);
  • REST API, consuming a set of signals from a vehicle, including the GPS position, ACC activation/deactivation, suspension status, speed, steering wheel angle, driver ID, and other signals that we refer to as a context. The API matches the driver’s location to an ACC friendliness map, captures the driver’s previous interaction and personal preferences history, and decides whether an ACC recommendation is appropriate at that time. If yes, the API triggers an ACC activation suggestion on the in-car head unit widget.

The client’s team developed an in-vehicle solution that:

  • Collects the above described in-car context and transits them to the cloud every 10 seconds;
  • Renders HMI sequences on a head unit according to sequence codes received from the Cloud. HMI sequences include text, images, text to speech (TTS), and audio tones. Each HMI sequence displays multiple screens through which the user navigates by pressing head unit buttons.

ACC Fleet Learning Map

Fleet Learning is a solution set that allows the system to learn the road segments that are most and least suitable for driving with ACC and make it available in each car based on observed system activation patterns by a fleet of cars.
The suitability of every ACC Friendly Zone for semi-autonomous driving depends on recency and frequency of system activations on each given road segment by the fleet of drivers.
The hybrid architecture used for Fleet Learning technology maintenance and accessibility allows aggregating and managing data from the vehicle fleets in a cost-effective and expedient manner.

Intelligent UX

Having to figure out how to correctly activate and configure the ACC is one of the burning pain points contributing to the decline of ACC usage.
Our intelligent ACC solution is accompanied by a targeted UX that flows seamlessly, explaining ACC activation and resumption flows via an unobtrusive interface.
The system picks up the most optimal communication channel for presenting ACC recommendations depending on the:

  • Actual ACC status (entirely off or in suspension mode);
  • Number of ACC suggestions per journey;
  • The driver’s manual inputs.

The system uses various combinations of head unit pop-ups, audio tones, and voice notifications to deliver the best ACC coaching experience and reduce negative impressions.


Cloudmade provides full end-to-end solution development from the co-ideation and integration architecture definition to the support for the vehicle during demo drives.

The solution was split into 4 main components:

  • In-vehicle UX/UI
    Written in proprietary framework, owned by the OEM.
  • Cloud individual predictions
    AWS as a cloud provider; REST API written in Java with Spring framework; CloudMade’s C++ SDK with Java wrappers; DynamoDB for context state handling; Fargate as a compute engine; S3 for data storage.
  • Cloud anonymized learning for geo data
    Spark; SparkML; custom Java-written ML algorithms, developed by CloudMade’s DS team; Rundeck/Airflow for pipeline orchestration; HDFS as a target storage for the analyzed dataset (HBase and Parquet data formats).
  • Distribution of the geo data
    CloudMade’s Python REST API, which could distribute data for the specific tile and layer, base on the initial request from the vehicle; Docker container; Fargate as a compute engine.

Tell us about your challenge. At CloudMade we are ready to provide expertise and support.

Contact us