Tesla's "Enhanced Auto Pilot" EAP, an amateur analysis

(Disclaimer: I am not a traffic engineer or investigator.  I just took delivery of a Tesla Model 3 a few weeks ago, and am interested in understanding Tesla's "Enhanced Auto Pilot", in light of recent press about questions of the reliability of Auto Pilot.) 
Tesla vehicles built since 2017 come with AP 2.0 or 2.5 hardware, a collection of cameras, sensors, radar, and computer power that can perform certain vehicle control functions.  All vehicles have the SAFETY-related functionality (for example Emergency Automatic Braking), and basic Cruise Control (which does not adjust from the set speed based on traffic).  But for more advanced functionality (Traffic-Aware Cruise Control, Steering Assist, Self Parking, Summon,) it is necessary to purchase "Enhanced Auto Pilot" ("EAP").  Tesla presents EAP as "beta" software, and instructs drivers to keep their hands on the wheel, and be ready to take control if the Auto Pilot fails to perform as expected.  If the system doesn't detect the driver's hands on the wheel, on the Model 3 it will display a warning message, flash a blue alert on screen, and eventually sound an audible warning, and failing to detect the driver's hands on the wheel, it will then disable Autopilot for the rest of the session (that is, until the vehicle stops and is put into Park.)

Recent attention has been drawn to how Autopilot can have problems identifying lanes and gore points, based on faded or missing painted lines.  One driver posted a video of a particular lane split in Chicago with poorly maintained lane markings where the car ended up traveling down the gore point headed for the crash attenuator on the concrete barrier.  Below, I have captured still frames showing when the AP was mis-identifying gore point markings as lane markings. 

The important part of these graphics is in the screen, below the speedometer, representing the car, and what the Autopilot recognizes as the travel lane, and the lane lines.  It shows the travel lane as a lighter gray path, the adjacent lane or shoulder as a darker gray path; it represents a paint line recognized as a lane marker as a solid blue line (whether it is a solid or broken line). When it does NOT recognize a lane delineator, there is NO blue line. (The blue line is almost green in these screen shots from YouTube.)

In this frame, at :26 on the timeline, through the windshield, we see the solid line on the left, and the broken line on the right. AP recognizes both correctly as the lane delineators, and is maintaining a target cruise control speed of 59 MPH.
 In this frame, at :27, the gore point is beginning, but with only the left line of the gore point being clear. AP recognizes the left gore line as the left lane delineator, and does not recognize a right-hand delineator, since the broken line is no longer within the normal distance for a lane, and the right-hand gore line is too faded to be recognized. AP still has (false) confidence in the left-hand line, and starts to follow that line into the gore point.
 In this frame, at :28, AP continues along the left-hand gore line further into the gore point, still with no right-hand line being recognized. The small chevrons are not recognized, or are ignored.
 In this frame, at :30, the car is fully in the gore point, but still recognizing the left gore point line as the left-hand lane delineator.
 In this frame, still at :30, we see the left and right gore lines about to be recognized as lane markers. 
 In this frame, the last one at :30, we see AP is now recognized both the left and right gore point lines as left and right LANE LINES, and still travelling at 59 MPH.
 In this frame, at :31, the DRIVER, appropriately monitoring the performance of AP, recognizes the problem, and applies the brake, which will disengage AP and start to slow the vehicle.
 In this frame, still at :31, the driver having braked (note the brake lights on the graphic of the car), the AP has disengaged, so AP is no longer attempting to recognize lane lines, and the speed is dropping. 
 In this last frame, at :35 seconds, the car has slowed to 4 MPH, about to stop just short of the crash attenuator. 

Link to video on YouTube:


Posted with permission of


One might think that AEB (Automatic Emergency Braking) should recognize the threat and engage.  However, AEB is designed to work identifying impending collisions with MOVING VEHICLES. The radar is designed to filter out stationary objects, which would often be ghost or reflected signals, resulting in false panic stops.

What I hope people take away from this analysis is confirmation that AP can be a helpful tool, but still requires the driver to be attentive, particularly where lane markings are not clear.  As Tesla points out, is is a work in progress.