As an AP1 Model X owner, I didn’t expect Autopilot to evolve much over the last 12 months. Tesla software engineers had their hands full writing code for AP2 vehicles, so I figured Autopilot updates for AP1 vehicles would simply enter a dormant phase. That certainly did not turn out to be the case! While racking up some 30k Autopilot miles over the last 12 months, I experienced a continuing stream of improvements.
First off, the “Hold Steering Wheel” alerts began disappearing on long, straight stretches of divided lane freeways – sometimes for as long a 50 miles at a stretch. But the alerts became more frequent on two-lane highways, and began appearing very frequently in construction zones (triggered by either cones or construction warning signs). They even became somewhat driver-dependent (when fatigued or distracted I would get them much more often). More recently, they now sometimes appear ahead of highway intersections, and when oncoming traffic appears on a two-lane highway.
Camera vision has steadily improved. AP1 originally got tripped up by faded stripes, by glare on wet roads, when driving into the sun, and on rainy nights. Improvements in the camera’s dynamic range have continued to take place, and the software’s ability to discern lane markings through glare, water, and dirt has dramatically improved in the last year. About six months ago, on a rainy late-night freeway trip, I noticed that Autopilot can now track the lane stripes better than I can when there is headlight glare from oncoming traffic.
Autopilot’s early tendency to slow down too much or too abruptly when cornering has also seen big improvement. Originally, the camera would read the recommended cornering speed on a highway or freeway, and Autopilot would reduce the vehicle’s set speed by the difference between the speed limit and the cornering speed. And if it encountered cornering Gs that were too high, it would slow abruptly. The first problem got solved by an update that allows Autopilot to track the cornering speed of actual drivers, allowing it to take precedence over the posted cornering speed. I discovered that tapping the accelerator briefly as Autopilot begins to slow for a corner will override the slowdown, so I played around with that over a curvy 50-mile stretch of freeway. Several weeks later when I driving the same route, Autopilot remembered the corners that had overrides, and maintained its speed setting. More recently, the over-reaction to cornering Gs saw a big improvement. Autopilot is now more willing to corner aggressively (at least on dry roads), and when it does reach its acceleration limit, the reaction is not as abrupt.
Holding an appropriate position within the lane during corners has also gotten a lot better. Starting about a year ago, Autopilot began hugging the inside track as human drivers do through corners (this can sometimes be a bit unnerving if there’s a truck in the next lane over, but on the whole it feels more natural). Autopilot also developed a better grasp on the lane positions of other vehicles during the cornering process, leading to fewer false braking events.
A few weeks ago, another big cornering improvement occurred, allowing Autopilot to handle even sharper corners. I decided to see how good it was on the Reno side of the Mt. Rose Highway, a super-curvy road that climbs 4000 feet over 12 miles. This road has several corners marked at 20 mph, and the posted speed limit is 45 mph. But even when you stick to that limit the cornering G-forces can be relatively high. For several years I’ve tried in vain to train Autopilot on this road, a process that generates dozens of overrides, carries the risk of being pulled over for a sobriety test, and requires a firm grip with two hands on the wheel. So imagine my surprise when Autopilot handled 90% of the course with relative ease. I still go a little slower than I’d like around the 20 mph corners, but so do many other drivers on that road.
Like other Tesla drivers, I found “Radar 2.0” to be a cool feature. One time coming out of a mountain pass in heavy fog, the car ahead of me remained visible on the dash, even as it got obscured watching through the windshield. I’ve felt more comfortable having a radar-driven emergency braking system, though my wife found the occasional false-braking events to be very annoying. But over the last few months it seemed like the false braking events largely disappeared. At times I wondered if the radar’s emergency braking role had been diminished.
Those concerns were put to rest two weeks ago while I was commuting home on old 395 just south of Reno. Shortly after using Autopilot’s lane-change feature to get around a slow-moving vehicle, Autopilot slowed the vehicle abruptly just as it was returning to the right-hand lane. The lane was clear as far as I could see (at least 250 yards), so I began to override the slowdown by pressing down on the accelerator. Then suddenly, I was seeing a large obstacle that the radar already knew about. An empty 30-foot-long trailer – visually obscured by trees – was lurking just around the bend, and it was blocking the right lane entirely. I had just enough time to swerve into the left lane and miss it. Without Autopilot’s reaction, I might have plowed into it at a relatively high rate of speed. Thank you Tesla software engineers!
Jack Bowers has more than 240,000 Tesla miles driven, including two cross-country trips. He is the publisher of Fidelity Monitor & Insight (a DIY investment newsletter) and an investment strategist for two affiliated advisory firms.
Photo: Mt. Rose Highway/Google Maps
First off, the “Hold Steering Wheel” alerts began disappearing on long, straight stretches of divided lane freeways – sometimes for as long a 50 miles at a stretch. But the alerts became more frequent on two-lane highways, and began appearing very frequently in construction zones (triggered by either cones or construction warning signs). They even became somewhat driver-dependent (when fatigued or distracted I would get them much more often). More recently, they now sometimes appear ahead of highway intersections, and when oncoming traffic appears on a two-lane highway.
Camera vision has steadily improved. AP1 originally got tripped up by faded stripes, by glare on wet roads, when driving into the sun, and on rainy nights. Improvements in the camera’s dynamic range have continued to take place, and the software’s ability to discern lane markings through glare, water, and dirt has dramatically improved in the last year. About six months ago, on a rainy late-night freeway trip, I noticed that Autopilot can now track the lane stripes better than I can when there is headlight glare from oncoming traffic.
Autopilot’s early tendency to slow down too much or too abruptly when cornering has also seen big improvement. Originally, the camera would read the recommended cornering speed on a highway or freeway, and Autopilot would reduce the vehicle’s set speed by the difference between the speed limit and the cornering speed. And if it encountered cornering Gs that were too high, it would slow abruptly. The first problem got solved by an update that allows Autopilot to track the cornering speed of actual drivers, allowing it to take precedence over the posted cornering speed. I discovered that tapping the accelerator briefly as Autopilot begins to slow for a corner will override the slowdown, so I played around with that over a curvy 50-mile stretch of freeway. Several weeks later when I driving the same route, Autopilot remembered the corners that had overrides, and maintained its speed setting. More recently, the over-reaction to cornering Gs saw a big improvement. Autopilot is now more willing to corner aggressively (at least on dry roads), and when it does reach its acceleration limit, the reaction is not as abrupt.
Holding an appropriate position within the lane during corners has also gotten a lot better. Starting about a year ago, Autopilot began hugging the inside track as human drivers do through corners (this can sometimes be a bit unnerving if there’s a truck in the next lane over, but on the whole it feels more natural). Autopilot also developed a better grasp on the lane positions of other vehicles during the cornering process, leading to fewer false braking events.
A few weeks ago, another big cornering improvement occurred, allowing Autopilot to handle even sharper corners. I decided to see how good it was on the Reno side of the Mt. Rose Highway, a super-curvy road that climbs 4000 feet over 12 miles. This road has several corners marked at 20 mph, and the posted speed limit is 45 mph. But even when you stick to that limit the cornering G-forces can be relatively high. For several years I’ve tried in vain to train Autopilot on this road, a process that generates dozens of overrides, carries the risk of being pulled over for a sobriety test, and requires a firm grip with two hands on the wheel. So imagine my surprise when Autopilot handled 90% of the course with relative ease. I still go a little slower than I’d like around the 20 mph corners, but so do many other drivers on that road.
Those concerns were put to rest two weeks ago while I was commuting home on old 395 just south of Reno. Shortly after using Autopilot’s lane-change feature to get around a slow-moving vehicle, Autopilot slowed the vehicle abruptly just as it was returning to the right-hand lane. The lane was clear as far as I could see (at least 250 yards), so I began to override the slowdown by pressing down on the accelerator. Then suddenly, I was seeing a large obstacle that the radar already knew about. An empty 30-foot-long trailer – visually obscured by trees – was lurking just around the bend, and it was blocking the right lane entirely. I had just enough time to swerve into the left lane and miss it. Without Autopilot’s reaction, I might have plowed into it at a relatively high rate of speed. Thank you Tesla software engineers!
Jack Bowers has more than 240,000 Tesla miles driven, including two cross-country trips. He is the publisher of Fidelity Monitor & Insight (a DIY investment newsletter) and an investment strategist for two affiliated advisory firms.
Photo: Mt. Rose Highway/Google Maps
Last edited by a moderator:
