As reported in the MIT Technology Review, this picture represents an ‘edge case’ where the sensors of an on-board autonomous vehicle were able to successfully identify a number of different objects but failed to understand that some were just images as part of a vehicle advertising mask. The camera and image processing identified cars, a person, and multiple bicycles, but most objects were not real. It simply could not tell the difference.
Such situations with important objects could cause the vehicle to act in an undesired way. In the right circumstances it could make a vehicle suddenly brake or swerve. Such ‘edge cases’ represent challenges for technologists working to make smart cars safe to operate without the need of human oversight.
The lesson learned from this image is that more sensors are needed to verify objects. The use of cameras, ultrasound, lidar, and radar can provide comprehensive data needed to distinguish and validate different objects. This is why many autonomous vehicles have an array of sensors.
Vehicles sensors, artificial intelligence (AI), and actuators all must act in concert to understand the dynamic environment in order to navigate safely. Overlapping sensors as well as a number of checks and balances is required in the AI logic to support safe autonomous operation.
This is no easy challenge. It takes tremendous engineering, testing, AI learning, and continuous improvements to adapt to new and unforeseen challenges. The very best technology, thought-leadership, and innovation is necessary to address 'edge cases' that will arise.
Trust will become paramount to the acceptance and proliferation of autonomous transportation.
But how will prospective customers and passengers know if the combination of technology is safe? Most of the autonomous driving magic is hidden away, embedded in the code of AI engines, and will not be understood by the average consumer. Nor will the nuances of sensor hardware. The result will be a rise in brand reputation. Consumers don’t need to know all the details in order to believe in safety. Many car brands are already known for superior crash safety, without the consumer being able to understand the myriad of systems that are involved.
The same will be true for autonomous vehicles. Suppliers and vehicle manufacturers will differentiate based on their bundle of safe and secure driving technology. Trust will be the new currency in future branding wars. Those who can achieve the best results will flaunt it, to attract buyers. Those without will be left in the rear-view mirror.
Image Credit: https://www.technologyreview.com/s/608321/this-image-is-why-self-driving-cars-come-loaded-with-many-types-of-sensors/
Interested in more? Follow me on LinkedIn, Twitter (@Matt_Rosenquist), Information Security Strategy, and Steemit to hear insights and what is going on in cybersecurity.
Now that's interesting! AI is pretty amazing but oh so unpredictable
Cool post, thanks! The major question is how much safer need the autonomous driving to be until mainstream adoption? Alreday now it's 100x safer?
Not sure. It is a relative question. There aren't any defined standards/test as yet, for government approvals.
Now this is getting cool. Surely the radar sensor should be enough to clarify whether or not what is perceived to be an object between the two cars actually exist?
I mean, when satellites can measure changes in elevation on the ground using InSAR sensors, then surely a car can spot an object in between it and another larger object just by comparing the distances between where the radar signals bounces off the large object (the car) and where it bounces of the smaller object (the bicycle) and conclude that it is indeed a bicycle if itsdistance is significantly lower.
Cool topic, keep posting!
Yes and they still make mistakes from time to time
Thanks so much for this useful post. I do definitely agree that more sensors are needed. However, the sensors have to work well with other technologies such as AI to be able to avoid accidents. It's still rather early for self-driving cars. I have no doubts that the technology will rapidly improve over time.
Now that LiDAR systems are becoming cheaper, most of these problems will be solved. One of the biggest problems now is that LiDARS ar big and extremely expensive (and ugly). But this technology is heavily evolving, clearing the path to autonomous driving for the common people
Can't even imagine how this will work. It's pretty hardcore engineering to tell AI to do everything right heh.
Most mistakes have been human error.
yes I can imagen. I heard of a city gonna test with those cars in the city only and no others and then you have to park ooutside the city and these cars bring you to you destination in the city.
Interesting! Which city?
It's very interesting.
The technology is growing up very fast, always surprise us :)
That's pretty amazing. Google has also been working on their automatic car. The world in changing.
It's mostly likely error of human. Only pro would understand :D Anyway thanks for sharing this :)
♣♣♣♣♣I will follow back those people who will follow me, I just thought that's a fair deal.♣♣♣♣♣
lol~
An ad in a car is quite rare though. But you're right, the AI has to be almost 100% sure and thus has to use other sensors. Very interesting article!