Tesla’s Path Prediction Improves Autopilot
Credit to Author: Johnna Crider| Date: Sat, 07 Sep 2019 16:33:05 +0000
Published on September 7th, 2019 | by Johnna Crider
September 7th, 2019 by Johnna Crider
In tests done on a curvy country road by YouTuber ALL Electric, you can see how Tesla’s path prediction has been improving Autopilot.
As you can see, ALL Electric has been running the same route repeatedly on Autopilot in order to track improvement over time in a single physical setting or route. In the video, he mostly compared this test (#7) to the first test, but also compared to test #6 at times.
The driver, Will, tested V. 2019.28.3.1 on the Model 3, which is the latest version of the software and the second version that “passed” their test. When he says passed the test, he means that he didn’t have to take over at all while testing out the curves on the road.
“Very impressive to see multiple versions in a row get the same result and require no intervention,” Will states in the video description.
He starts out by driving down an easy road and even over some railroad tracks. The car glides over them fine. He mentions that he would love to see if Autopilot would stop when the railroad crossing gates drop down, but that situation hasn’t arrived yet.
Tesla’s Autopilot sees a person crossing in front of the Model 3 in a realistic manner as the car is stopped.
While in traffic, a group of firefighters were out collecting donations and Tesla’s Autopilot saw the one standing on the center line in the road as the car approached. The driver then disengaged to donate funds. While disengaging, Autopilot saw the fireman running across the front of the car.
Will had to disengage only for a stop sign and to make a left turn, since Autopilot doesn’t yet handle those situations. (That feature is coming soon.)
The video shows flashbacks of test number 1 repeatedly to highlight how things have changed by test number 7. Both videos show the same road do a good job of clearly pointing out differences. For example, in the first comparison, there is an aggressive left turn. In test #1, Will increased the speed manually just a small amount to take on a right bend then a left turn just as a car passed. The Tesla hugged the left side of the lane.
In test #7, the video shows a major improvement in which the car stayed in the center of the lane while taking on those curves as opposed to crossing out of the lane in the very first test. Right after this, Will mentions how Tesla’s Path Prediction improves Autopilot.
“Now it is doing a lot better, and that’s thanks to this Path Prediction that they unveiled on Autonomy Day.”
Tesla Vision, which is the computer vision system built into NVIDIA’s CUDA, uses Path Prediction to improve the system between the two tests.
Tesla Vision uses path prediction to accurately predict how a road will extend, even when it can’t see around the corner pic.twitter.com/09qPkpqwSC
— Tesla (@Tesla) April 23, 2019
The video shows another curve comparison between the two tests where in test #1 Autopilot seems to head for the embankment while trying to take on the curve and the driver had to take over. In test #7, Autopilot manages that curve perfectly and the driver says, “Autopilot system is actually improving from test #1 through test #7.”
In another curve, which is a sweeping big right turn, Autopilot stays in the center of the lane and manages that while following a blind turn in both tests. However, in test #1, it hugs the lane more than in the more recent test.
The main improvement in these tests shows that Autopilot is learning how to stay in the center of the lane while executing the curves. There were a couple more curves that they tested and it shows the improvement of Autopilot as well as other areas that Autopilot still needs to improve in — something that, with each update, Tesla seems to be doing well.
One of the most important things one should consider while using Autopilot is that the driver should be ready to take over at all times. Yes, Autopilot makes driving easier and less stressful, but until it is 100% perfected, drivers need to always pay attention to the road. This means that if you see a giant fire truck in front of you and the car doesn’t show any signs of stopping, take over. This is not a joke and not something to be complacent about.
Drivers who appreciate Autopilot and where Tesla is going should be especially careful to drive responsibly.
Johnna Crider Johnna Crider is a Baton Rouge artist, gem and mineral collector, and Tesla shareholder who believes in Elon Musk and Tesla. Elon Musk advised her in 2018 to “Believe in Good.” Tesla is one of many good things to believe in. You can find Johnna on Twitter