DATA IN LOOP - Big Data Engineering Pipeline Development for ADAS

DATA IN LOOP - Big Data Engineering Pipeline Development for ADAS

Problem Overview

One of the biggest challenges in Intelligent Mobility is Driver Assistance and Autonomous Driving Functions development and adoption.

Unlike typical software and systems development, ADAS functions development and validation is a complicated process wherein the AI models need to be validated in a variety of real-world scenarios. Moreover, the explainability and interpretability of the models on various environmental scenarios are the core fo model adaptation at scale.

To achieve this, we need to adapt what we call a DATA IN LOOP framework.

DATA IN LOOP DEFINITION – Continuous up-gradation and adaptation of ADAS Functions by capturing real-world data from various drives across driving conditions. The primary Steps include:

  • Data Acquisition and Annotation
  • Data Conversion and Transformation
  • Data Availability and Observability
  • Data Governance
  • Data – Analysis and Management Frameworks

Challenges

ADAS functions and Data generation from various drive conditions in the core to develop next-generation algorithms/functions. However, acquiring the real-world drive measurement data is a huge challenge. The entire lifecycle of data movement is a high time-consuming and costly process in terms of data storage, transformation, and management.

Approach

While working with multiple global OEMs in terms of conceptualizing the process of ADAS – Functions development and Data Lifecycle helps us understand the major bottlenecks and data management challenges.

We seek to streamline the process of the entire Data Lifecycle by automating the entire data conversion and staging pipelines using the modern Big Data Engineering Ecosystem.

This is a continuous development and improvement framework to build next-generation frameworks to integrate various State of Art Models and Sensors/Data Acquisition Systems at ease.

Impact

Our customers have successfully reduced the TAT for real-world data availability from 2 weeks to mere 24 Hours. In addition, the reduction in data redundancy in individual Systems/Networks has helped reduce Memory, Bandwidth and Power consumption at a large scale.