automatic lane centering

One of the two primary features of Level 2 ADAS (Advanced Driver Assistance Systems) is Automatic Lane Centering (ALC) – the car's capability to steer itself in the center of a lane. Sometimes you hear people saying 'these systems are all the same'. The results here demonstrate the opposite, with some functioning impressively and others, so poorly you shouldn't consider them. 3rd-party critiques of ALC generally range from purely anecdotal comments to casually structured tests which are actually still subjective. In the last few years, Moblr has been unique in developing its own measured and reproducible ALC testing, including its methodology, test equipment, software and analysis. The test section of road we use consists of a 14-mile mix of straight and curving freeway driven at 70 mph in consistent lighting (click on the video for more). The results are boiled down to the histograms you see below; basically, bell curves showing the various places in the lane the car tends to be (the curve's peak being its most favored one). What follows are all the result's we've tested, but with the first one – for comparison – showing how a human driver (me, trying my best) performs. (In this case, you can see my left-of-center bias in the lane). This offset from the center also appears in many of the ALC examples, likely due to miscalibration of their system's camera. An optimal graph would have the vehicle's positions appear as a very narrow spike exactly in the middle of the lane; the wider and more spread-out the graph, the more the car wandered around. The associated scores are calculated by penalizing how far the data is distributed away from the lane's center. If you're interested in this data, or our ADAS-testing, drop us a message.