Even more than the Twitter feed of CEO Elon Musk, what gets the most attention about Tesla is its ever-evolving driver-assist technology, Autopilot (with Autosteer functionality) and Full Self-Driving (FSD). Do their names overpromise their capabilities? Will FSD someday deliver a system that’s actually self-driving? A big reason why Car and Driver leased a Tesla Model 3 in October 2019 was to test that technology. That’s why we chose to equip our Midnight Silver 2019 Model 3 Long Range with Tesla’s optional Full Self-Driving package. Although we paid for FSD over a year a half ago, our car still doesn’t have the tech. While we wait for Tesla to release FSD en masse, we developed a driving route to test the capabilities of any driver-assist technology. This will let us compare Tesla at its current level of driver assistance against a future capability for self-driving that we expect to get soon via an over-the-air update.
This content is imported from YouTube. You may be able to find the same content in another format, or you may be able to find more information, at their web site.
The Society of Automotive Engineers (SAE) defines self-driving technology as a set of levels, from 0 to 5, and self-driving capability starts at Level 3. At that point, the car can in some scenarios manage most aspects of driving, including monitoring the environment. But the driver is prompted to take over when a situation arises that the car can’t handle on its own.
Tesla’s Autopilot, likewise, requires drivers to make regular steering-wheel or control inputs, so drivers can’t fully check out—at least yet. (Although, it is possible to deliberately trick the system into thinking the driver is turning the steering wheel by hanging a weight from it.) During our test we took our hands off the wheel at various moments for demonstration. If the system determines the driver is out of the loop for too long it shuts itself off.
All new Teslas since 2020 are equipped with the Full Self-Driving computer, Musk says, but owners must purchase the $10,000 Full Self-Driving Capability package (which cost $6000 when we purchased it) to get its features. And it could be subscription based in the future. Soon, Musk says, owners will gain access to the technology the name promises, although, even Tesla admits, it still won’t be fully self driving. But still, many accidents have occurred while drivers were using the existing technology, and the National Highway Traffic Safety Administration (NHTSA) has investigated numerous instances.
Our loop introduces numerous difficulties for driver-assist systems to navigate, including roundabouts (humans in our area haven’t figured these out yet, either), 90-degree highway curves, cloverleaf interchanges, complex highway-to-highway merges, and urban areas with lots of pedestrian and vehicle traffic. Our aim is to repeat this loop once Full Self-Driving becomes available (as well as its subsequent updates over time) and show how the car’s performance improves—or gets worse—over time.
For this first test, we drove the 70-mile loop around our Ann Arbor, Michigan, headquarters under optimal weather conditions.
Here’s where our 2019 Model 3 Long Range stands now:
Autopilot and Navigate on Autopilot (March-April 2021)
Autopilot allows the car to steer, accelerate, and brake within its lane. It uses the car’s eight cameras, 12 ultrasonic sensors, radar, and onboard computer to read lane paint on the road and detect surrounding cars. Buyers who pay for Full Self-Driving also get an additional functionality called Navigate on Autopilot, which makes automatic lane changes around slower traffic on the highway, and can steer itself through highway interchanges.
The driver-assist system has evolved over the roughly 28,000 miles we’ve spent with our Model 3. It’s added the ability to recognize stop lights, stop signs, traffic signs painted on the road, cones, trash cans, and pedestrians, which it displays to the driver on the left side of the central touchscreen. In the spring of 2020, Traffic Light and Stop Sign Control Beta was added to cars that had opted for Full Self-Driving, which allows the car to stop for traffic control. However, drivers need to press the accelerator pedal or gear stalk to prevent the car from stopping at green lights, and the system had to be re-engaged manually after each stop, which made us question whether or not this was indeed progress. Intervention isn’t necessary when following a car through an intersection.
Tesla’s system has an advantage over General Motors’ Super Cruise hands-free driving system because it can be engaged off the highway. Super Cruise currently works on more than 200,000 miles of mapped highways in the U.S., though GM says that its goal is to have hands-free driving in 95 percent of scenarios. The system uses a camera behind the steering wheel to monitor if the driver stops paying attention to the road. After a series of warnings, the system shuts down.
Our route includes neighborhood streets and some roads with faded lane paint, and Autopilot doesn’t currently engage in those locations. Granger Avenue in Ann Arbor is a particularly tricky road for Autopilot (pictured below), and we’re curious to see how FSD handles it in future tests. Vehicles are meant to share a middle lane with bike lanes on both sides. With Autopilot engaged, our Model 3 went smack dab down the middle, hogging the center of the lane. Thankfully, there was no other traffic or cyclists present on our initial test loop.
Navigate on Autopilot is still in Beta, and when activated in the Autopilot menu, an alert reads: “Navigate on Autopilot does not make your Model 3 autonomous. Like other Autopilot features, the driver is still responsible for the car at all times.” The system engages from on ramp to off ramp, even navigating highway interchanges, when a destination is set in the car’s navigation system. The system will find the fastest route and change lanes to pass slower vehicles. Drivers can set a lane change confirmation or have the car switch lanes automatically behind slower vehicles.
We found that in most cases the car merges onto the highway too late, way at the end of the merging lane, and there was one instance exiting M-14 onto N Main St (pictured above) where the car did not decelerate enough, causing us to quickly take over to avoid a potential collision with the guard rail. The system also keeps the car in the passing lane too long after passing a vehicle that’s traveling at a slower speed in the right lane.
We were impressed when the system navigated us from I-94 to north on U.S. 23 without any intervention. The tricky cloverleaf includes a two-lane merge to travel north on 23 (pictured above), with cars coming from 23 south and crossing lanes. However, there wasn’t much traffic during our test. Traveling east on I-96 to south on U.S. 23 requires crossing three lanes of traffic, and the system could not complete all three lane changes in time for the exit to U.S. 23 south. We had our automatic lane changes, which include disabled, mild, average, and Mad Max settings, set to average and will continue to do so throughout our test.
With the features the Full Self-Driving package offers today, the system isn’t anywhere near something we’d call “self-driving”. The system bobbled more than 10 maneuvers and driving situations on our loop, some of which required us to quickly retake control. Although, we were impressed with its capability while taking certain highway interchanges, lane changes, and use on two-lane roads.