TechPrufer
Get started
Grab
Grab logoGrab · Feature trail

Driver Telematics and Safety Analysis

1 post · Mar 2022 – Oct 2022

What this trail covers

Grab has developed an in-house telematics engine that uses mobile phone sensors (accelerometer, gyroscope, GPS) to collect and analyze driver-partner driving behavior. This system detects events like harsh braking, acceleration, cornering, and unsafe lane changes to improve passenger safety and inform insurance premium calculations. The architecture includes on-device processing for crash detection and backend processing for generating safety reports after each ride. Key technical challenges addressed include sensor data time synchronization, hardware axis bias, sensor noise, and data compression for network efficiency. The system aims to extend telematics to all bookings across Grab verticals and add more on-device rules for event detection.

All Grab trails
  1. Using mobile sensor data to encourage safer drivingThis post dives deeper into Grab's telematics technology for encouraging safer driving. It details the challenges of using mobile sensor data, including accounting for phone orientation relative to the vehicle and handling variations in data quality across different phone models and vehicle types (two-wheel vs. four-wheel). The post outlines a customized processing algorithm that uses Euler angles (roll, pitch, yaw) derived from sensor data to infer vehicle acceleration. It also describes robust methods for detecting unsafe driving instances (acceleration, braking, cornering) using thresholding and peak finding algorithms, with fail-safe checks to minimize false positives. Speeding detection is achieved by comparing GPS speeds with map data and aggregated speed statistics, with a detailed explanation of how representative speed profiles are inferred by re-weighting ping counts and filtering for free-flowing traffic.
  2. How telematics helps Grab to improve safetyThis post details the development of an in-house telematics engine that uses mobile phone sensors (accelerometer, gyroscope, GPS) to collect and analyze driver-partner driving behavior. It describes the architecture with on-device processing for crash detection and backend processing for safety reports. Key implementation details include sensor data time synchronization through interpolation and decimation, and using Fourier Transform for compression. Learnings highlight challenges with sensor data frequency, vehicle type differences, hardware axis bias, sensor noise, time synchronization, and data compression. Future plans include extending telematics to all bookings and adding more on-device rules.