Driverless Cars: (Mostly) Safer Than Humans- Struggle with Turns
I was scanning the morning news when the following headline caught my eye: "Driverless cars are mostly safer than humans – but worse at turns." Yes, that certainly tracks with my experience. I've been using Tesla's "Supervised Full Self Driving" (FSD) since March.
Elon gave everyone one month free and then I received another three months for referring my husband when he bought a Tesla. (It was a win-win for both of us. He upgraded his truck for a much better ride, and he stopped borrowing my car all the time.)
Supervised Full-Self Driving
I wrote about FSD when I was first trying it out. I found the software to be buggy but promising. To be precise, a Tesla vehicle is not a fully autonomous vehicle. Instead, it is classified as a Level 2 autonomous system under the SAE (Society of Automotive Engineers) levels of driving automation. (See my article on the history of self-driving cars.)
As each month has gone by, the software in my car has continued to improve. I went from issuing a half dozen bug reports a day to just one or two. The reports I've issued have been acted upon, as new features are being continually added to the car. The driving experience keeps getting better and better. I'll keep providing feedback, and the software will continue to improve.
Turns, Turns, Turns
Now, that being said, there still are some areas where FSD makes me uncomfortable- such as turning. I'm comfortable letting the car exit the highway on its own, but turning left at the end of the ramp? Ugh. That makes my stomach flutter. It still feels like the car is struggling to find the lane. Not a big deal because my hands are always on the wheel and I can correct, but I wish I didn't have to do it.
Roundabouts
The biggest struggle the car has is figuring out roundabouts and our town is lousy with them; a nearby stretch of road has three in the span of a mile. FSD continually struggles to figure out the correct lane. It loves to chose the turning lane when we are going straight. This is not as terrifying as it sounds. It is always obvious when the car is making the wrong choice. I can easily take over as the human. It's akin to watching a toddler telegraph a move before they do it. You just need to be watching their every move—hence the term "Supervised Full Self-Driving."
So, that is a semi-autonomous vehicle. What about the fully autonomous ones?
Autonomous Vehicles: Safety and Challenges
The recent study covered by New Scientist sheds light on the nuanced safety dynamics of driverless cars compared to human-operated vehicles. The findings, based on a large-scale accident analysis led by researchers from the University of Central Florida, suggest that while autonomous vehicles generally offer a safer alternative in routine driving scenarios, they falter in more complex conditions such as low-light environments and during turns.
Key Insights from the Study:
Researchers Shengxuan Ding and Mohamed Abdel-Aty advocate for enhancing sensor integration and weather adaptation technologies to mitigate these risks. The study emphasizes the need for improved safety measures, particularly during turning maneuvers and under varying lighting conditions, which remain significant hurdles for current autonomous driving technologies.
The Challenge of Data and Bias
The research also underscores a critical challenge in the autonomous vehicle arena—data reliability. Experts like Missy Cummings and Eric Teoh have pointed out potential biases in accident reporting and the limited scope of data, which can skew perceptions of autonomous vehicle safety. This calls for a more comprehensive and unbiased data collection and reporting system to better understand and refine the safety capabilities of driverless cars.
Final Thoughts
As autonomous vehicles continue to be tested and deployed across various U.S. cities, the findings from this study serve as a crucial checkpoint. They not only highlight areas of significant progress but also outline the pivotal challenges that need to be addressed to align the integration of autonomous vehicles with public safety and acceptance. As the technology progresses, so too must the strategies for its integration, ensuring that safety remains at the forefront of this automotive revolution.
As the driver of a Level 2 vehicle, I'll continue to happily argue with my car on roundabouts. With each new weekly software update, I'll await that magical day when the car figures out the correct lane.
领英推荐
Read the article from NewScientist here.
Crafted by Diana Wolf Torres, a freelance writer harnessing the combined power of human insight and AI innovation.
Stay Curious. #DeepLearningDaily
Vocabulary Key
FAQs
Additional Resources for Inquisitive Minds:
Decoding the Brains Behind the Wheel: The Role of Deep Learning in Autonomous Vehicles. Deep Learning Daily.
What is Tesla's "end-to-end neural network?" Deep Learning Daily.
What Happens When The Second EV Comes Home? Deep Learning Daily.
What Happens When AI Doesn't Understand Us? Deep Learning Daily.
FSD Beta 12 and Tesla's "Meet the Engineers" Event Deep Learning Daily.
How AVs Could Transform Our Cities (and Your Commute) Deep Learning Daily.
Eyes on the Road: The Privacy Implications of Camera-Equipped Cars. Deep Learning Daily.
Navigating the Ethical Roadmap: The Complex Intersection of Autonomous Vehicles and AI Ethics. Deep Learning Daily.
A Journey Through the Past, Present, and Future of Self-Driving Cars. Deep Learning Daily.
#SelfDrivingCars #ArtificialIntelligence #DeepLearning #TeslaFSD #AutonomousDriving #TeslaTechnology #AITraining #HumanFactor
Yet here we are less than 2 months later, and I would say that behavior in roundabouts is a major strength in FSD 12.5.1.1 and its ability to handle unprotected left turns is generally better than mine, or at least, when we team up, togther we do a better job. The exception cases are when human drivers do something unexpected like be surprisingly polite or excessively aggressive.