Tesla is using machine learning software to fix the problems it has been having with the autopilot, or self-driving, program.
Even though Consumer Reports recently withdrew its support for the Model S because of reliability issues, the problems will become a memory quickly with machine learning. Using data collected from real-world driving to identify problems and explain why they occurred will quickly get the feature working well.
Many videos have been released that show how the autopilot features can go wrong in practice, like that the vehicles have difficulty staying in the right lane on a poorly paved road. Tesla CEO Elon Musk addressed some of the problems in a tweet, saying that the new Autopilot update includes “curve speed adaption, controller smoothness, better lane holding on poor roads, improved fleet learning!” TreeHugger points out that would address nearly all the problems that have been noted.
The drivers themselves also apparently created the problem in part by driving the vehicles on roads where autopilot use is not recommended, or not keeping their hands connected to the wheel.
Musk also tweeted about Consumer Reports, explaining that they included many older vehicles in the survey and that the problems have already been addressed in the newer models.
Tesla’s “improved fleet learning” machine learning software uses tens of thousands of vehicles to gather as much information about their driving experience as possible. That information is then used to update the programming of the autopilot. It’s a powerful tool that makes these temporary setbacks irrelevant.
