An ADAS Research Report on Subaru’s EyeSight System
Date: March 2020
Research Project: Understanding the failure rate of an uncalibrated windshield (After Replacement)
Investigator: Ricardo Da Cruz
Prepared by: Ricardo Da Cruz
Statement of Purpose:
This case study is designed to examine how the Eyesight system(s) available in current Subaru models react to not being calibrated after windshield replacement.
Research Questions:
The study examines three major questions:
1. What is the failure rate for object recognition prior to calibration?
2. What is the failure rate for object recognition after calibration?
3. What differences exist between OE and Autel?
TL;DR:
Uncalibrated
51.6% Failure rate at night
10.0%-33.3% Failure rate during day (Percentage range is dependent on on data points below)
Uncalibrated runs, day or night, were plagued by inconsistent system crashing while failing.
Calibrated
0% Failure rate at night – 97% Accuracy Calibration
0% Failure rate during day – 98% Accuracy Calibration
Calibrated runs crashed system significantly less than uncalibrated.
OE Manufacturer information is mandatory when performing these services, regardless of tools used.
What does this mean??
If you replace a windshield and don’t have it calibrated, your Eyesight system in your Subaru is likely 50% less effective at night and at least 10% less effective during the day.
Study Methodology:
To place a recognizable target in the path of the vehicle. Record and observe how the vehicle would react when presented with different external circumstances. Record findings at night and day.
Testing was carried out in an area awaiting redevelopment with ample room to bring a vehicle to speeds required for the Eyesight system to function as needed.
Testing was conducted with a variety of cruise control and non-cruise control variables. Vehicle would be driven at varying speeds at the target with documentation of each test pass.
All driving was recorded.
Tools Used:
2018 Subaru Impreza – Accident vehicle with windshield replacement. No codes stored; no calibration performed prior to testing.
- Service Management Software: CAR
- Subaru OE Information Subscription – 3 times.
- Identifix
- Autel Scan Tool w/ADAS Subscription
- ADAS Hardware w/Calibration Target & Hardware
- Battery Tender
- Target – was designed/built inhouse. Made of simple materials that could be found at any Target and Home Depot. Designed to limit any potential vehicle damage caused by impact. Made to represent a person approximately 12 years or older in height. Needed to be replicated in less than an hour if broken beyond repair. Materials available upon request.
Overall Research Findings:
From our testing the following is clear:
1. Nighttime testing, No Calibration:
Conditions: Cold, Clear, 9pm
- 61 test runs
- 47 center line passes
- 24 clear passes, all center line
- 20 clear fails, 18 center line fails, 2 Left Bias (LB) or Right Bias (RB)
- 4 center line swerves, late warnings, swerved – High speed
- 1 Moving target, high speed, good warning time
- 6 All Systems Down – 2 at night
- 38 Night Runs: Failure test here was object recognition
- 23 Day Runs: Failure test here was object recognition
- 31 night runs at center, 16 fail 51.6% | 15 pass. 48.4%
2. Day Time testing, No Calibration:
Conditions: Cold, Clear, 9am
- 16 day runs at center, 6 that are questionable because of much higher speeds
- Removing those 6, has 10 remaining, 2 fail, one is questionable based on the results from calibration sample tests, the other is no longer question based on those same results.
- 10.0% failure rate assumed in this calculation, 90.0% passing rate assumed – I wonder if this has to do with the percentage rates you get on the machine when calibrating?
- Leaving the 6 but removing 1 (included in the original 6 count) where the target moved (wind), 15 successful day runs at center. 10 passing, 5 failing for target hits.
- 33.3% failure rate assumed in this calculation, 66.7% passing rate assumed
3. Nighttime testing, CALIBRATED:
Conditions: Cold, Cloudy with Light Rain, 7pm – It got boring…
11 runs done, object recognition was 100%.
4. Day Time testing, CALIBRATED:
Conditions: Cold, Clear, 11am
21 runs done, object recognition was 100%.
Last 5 runs were done at 30mph+ on cruise control.
5. OE Information & Setup Findings
- A little difficult to navigate if you’ve never used, much like all other OE systems.
- Instructions were thorough, though difficult to attain a 100% reading.
- Lack of information on understanding exactly what is the % reading and what it means.
- OE Software is out of reach because of price.
6. Autel Information & Setup Findings
- Calibration frame setup is not the best system for efficiency.
- Subaru target does not line up with center of calibration frame.
- Height leveling is a painful exercise that’s inaccurate and wastes time overall.
- Target height is not the same as Subaru spec of 1000mm.
Video & Photo Documentation Available Upon Request.
From our research in general the following is clear:
1. The market still does not understand these systems, how to price their repair, or what’s really at stake or involved.
Calling 4 facilities yielded the following results.
- Subaru Dealer #1 – $465 for Eyesight calibration, blind spot calibration is done at the same time according to this dealer.
- Subaru Dealer #2 – $165 for Eyesight calibration and didn’t know what blind spot was, but came back on the phone and said the same exact price for all the services. $165 each.
- Local collision facility – $250, claimed to “do a lot of work for Safelite”. Did not ask about blind spot.
- Safelite – $600
This information may not be used or republished without written consent.
