ADAS Functions – Validation - Big Analysis Framework

ADAS Functions – Validation - Big Analysis Framework

Problem Overview

The typical Functions Validation and Component Testing process in Automotive has taken a setback while adapting to modern Software and Models. Unlike Deterministic Algorithms, modern functions use a Stochastic Modeling based approaches which requires a paradigm change in validating the components in a real-world test scenario.

Challenges

With the advent of AI and Intelligent Systems, modern Software and Models have moved towards a more Stochastic based approach in comparison to older Rule Based Methods. Major challenges include:

  • Data availability and Annotation.
  • Uncertainty and Explainability of Models.
  • Model Drift.
  • Anomalous Behaviors of Data Acquisition Devices [Analog and Digital Systems], Model Output and Actuators.

Approach

Leveraging the availability of data at ease [Thanks to the Big Data Ecosystem], several Data Analytics frameworks have been built wherein, ADAS Functions Validation is the breakthrough. Validation of various Intelligent Systems/Components is approached through the following processes:

  • Requirement generation and KPI
  • Data acquisition and Annotation
  • Anomaly Detection, Scenario Mining, and Statistical Analysis
  • Event Tagging and metadata generation for future reference
  • Report Generation

Tech Stack


Impact

Our customers were able to validate the behavior of the ADAS/AD algorithms in the real-world environment through various performance metrics and identify various anomalous behaviors of the functions based on various driving conditions. This helps in understanding the gaps in the model design and implementation and handling edge-cases that can be found in the real-world driving environment.