CHANNELS

Aerospace

Automotive

Automated & Autonomous Vehicles

Battery & Electrification Technology

Defense

Drone TV

Electronics

Manufacturing & Prototyping

Materials

Medical

Motion Control

RF & Microwave Technology

Robotics & Automation

Sensors & IoT

Test & Measurement

Aerospace
Automotive
Automated & Autonomous Vehicles
Battery & Electrification Technology
Defense
Drone TV
Electronics
Manufacturing & Prototyping
Materials
Medical
Motion Control
RF & Microwave Technology
Robotics & Automation
Sensors & IoT
Test & Measurement
Aerospace & Defense
Search Results

Developing Truly Autonomous (and Safe) Vehicles

The push toward truly autonomous vehicles has been hindered by the cost and time associated with safety testing, but a new system developed at the University of Michigan shows that AI can reduce the testing miles required by 99.99%. It aims to enable manufacturers to more quickly verify whether their AV technology can save lives and reduce crashes. Watch this video to see what happens when vehicles trained by AI perform perilous maneuvers.

“The safety critical events—the accidents, or the near misses—are very rare in the real world, and often time AVs have difficulty handling them,” said Henry Liu  , U-M professor of civil engineering and director of both Mcity and the Center for Connected and Automated Transportation, a regional transportation research center funded by the U.S. Department of Transportation.

Topics:
Automotive Autonomy

More From SAE Media Group

    Transcript

    00:00:01 We have billions of years of evolution that  back us in performing the daily driving task.   Teaching computers how to replace us in this  task has proven to be quite complicated.   There are seemingly endless rare safety critical  situations that occurs on the road that cannot be   handled properly by autonomous vehicles. In  real world testing environments it can take   billions of miles for an autonomous vehicle to  repeatedly experience rare dangerous events. This   makes it time-consuming and expensive to validate  their safety performance, slowing down companies   deploying AVs that are fully autonomous.  Now Michigan Engineering researchers are   using artificial intelligence to train virtual  vehicles to perform adversarial behaviors that   can challenge autonomous vehicles. Instead of  testing autonomous vehicle in a naturalistic   driving environment where our safety critical  events are rare we challenge autonomous vehicle  

    00:00:59 using intelligent testing environment which  safety critical events are much more frequent.   The virtual vehicles are taught to constantly  challenge autonomous vehicles as they drive   around a test track performing maneuvers  such as cutting in or failing to yield at   a roundabout. To ensure the frequency of safety  critical events the researchers use the dense   deep learning approach to train the neural  networks that make maneuver decisions. In our   default neural network to learn effectively we  remove the non-septic critical data and feed the   neural network with only the safety critical  data so that we can accelerate the testing   efficiency but guarantee the testing accuracy. By  exposing an autonomous vehicle to rare events more   frequently the intelligent testing environment  can reduce safety testing miles by 99.99 percent. If we're going to do this in the time frames  that we want to we're going to need a way for  

    00:02:04 our software to experience all of the things  that we might encounter over a lifetime of   driving a car. We can demonstrate to consumers  that autonomous vehicle has the potential to   have comparable performance like human drivers. We  are not there yet but we are on the right track.