Commentary
Stanford AI Teaches Autonomous Cars To Learn From Mistakes
- by Chuck Martin , Staff Writer, April 1, 2019
Artificial intelligence is being used for controlling autonomous cars.
Stanford University researchers developed a system for controlling a vehicle by integrating past driving experiences to aid how a car can perform in extreme or unknown circumstances.
Using a Volkswagen GTI and Stanford’s autonomous Audi TTS, the car performed like an existing autonomous control system and an experienced racecar driver, according to the study.
The researchers created a neural network to integrate data from past driving experience, using a closed course study of racing performances of the first five turns at the Thunderhill Raceway Park in California. The automated vehicle and human attempted to do the course in the minimum amount of time, approaching speeds of 95 miles per hour.
The research focused on autonomous vehicles adapting from normal conditions to driving in ice and snow.
Self-driving systems typically rely on real-time evaluations of their surroundings, such as identifying objects ahead, while the Stanford study incorporated data from recent maneuvers and past driving experiences. For example, the system included data from trips around an icy test track near the Arctic Circle.
“If we can develop vehicles that have seen thousands of times more interactions than we have, we can hopefully make them safer,” states one of the researchers.
Chuck Martin is Editor of the AI & IoT Daily at MediaPost and writes the daily Connected Thinking column. He is a NY Times Business bestselling author and his latest book is “Digital Transformation 3.0” (The New Business-to-Consumer Connections of The Internet of Things.) He is CEO of Net Future Institute and a frequent keynote speaker internationally. Follow him on Twitter @chuckmartin and contact him at Chuck@MediaPost.com.
Trips around a race track without other cars.............FANTASTIC. This is just one more Dumb and Dumber idea. So what happens if it hasn't had this experience yet, you know, a case of Virgin Auto-tonomousity. Never had the experience , so don't have anything in the memory bank to draw from. When a car slams on the brakes in front of me , I don't think,
"I remember this same scenario back in August of 78".
Come on, just this weekend , airlines got shut down again. Not engine problems, not strikes...................but computer failure.
After reading this article:
https://www.aitrends.com/ai-insider/airplane-autopilot-systems-self-driving-car-ai/
My conclusion is it takes more AI to self drive a car than a plane , because they will never allow take-off, landing and uplift without human guidance. Level 1-4 self driving cars requires a human to be on stand-by. (?) So WHAT THE HECK IS THE POINT??
Level 5, total Auto-tonomous.
There is an average of just over 100,000 flights a day. That's a lot of AI information out there. But airplanes are usually miles apart at many different altitudes. How many 100,000's of cars are there just on I-5 from San Diego to LA, 5 lanes deep, bumper to bumper, people merging and exiting, couches and ladders on the roads and you are going to tell me that all of those cars connected to the same AI source can not have an Epic Fail.
You're stupid if you think not. Auto-tonomous D.B.A Dead Before Arrival.