A French startup has all the Artificial Intelligence technologies needed to avoid many automous car accidents such as the recent UBER accident

Logo NEXYAD 3D (200x82) 01

We all saw videos of the UBER VOLVO accident. The dashcam video shows that visibility conditions were very poor and that the pedestrian is dressed in black (pullover) and crosses a road at night without any light and out of pedestrian crossing areas. This attitude is pure suicade and you wouldn't teach your children to behave this way. But even if this pedestrian shouldn't have been in the middle of the road at night with no light (knowing that it is not possible that the pedestrian didn't see the lights of the car ...), we all expect from an autonomous vehicle not to crash the pedestrian.

It seems that the camera was not used as a sensor by the autonomous car (it seems that the VOLVO UBER car uses lidar like Velodyne or other equivalent one). But poor visibility conditions brings a factor of risk anyway (even if your autonomous car is not based on artificial vision) because human beings (human drivers and pedestrians that share road infrastructure with autonomous cars) are vision-based "systems". It means that when you detect a lack of visibility, you should slow down, event if your sensors can "see" anyway : if you are the only one to "see", there is a danger : accident is avoided because everyone can see everyone. If someone cannot see, then accident avoidance is on the shoulders of the only one that can see and risk is increased. The French startup NEXYAD proposes an algorithm called VisiNex that measures in real time "visibility" distance and quality. If visibility is low, then slow down.

See VisiNex demo : 
demo 1 : VisiNex applying on an automobile cam : https://www.youtube.com/watch?v=41MW65cmT-I
demo 2 : VisiNex applying on a road monitoring cam : https://www.youtube.com/watch?v=_Lp48uGpihI

This company NEXYAD also proposes an artificial intelligence algorithms for map and sensors fusion that sorts all information and alerts on the the most dangerous criteria. On the UBER accident scene, it is almost sure that the Velodyne lidar sensor "saw" an interdistance and a "time to collision" that was rapidly decreasing, plus there were an intersection (high slope road connecting on the left : probably where the pedestrian came from with the bicycle). SafetyNex would have alerted for vigilance both the human driver and the robot driver (that would have slowed down).

SafetyNex computes driving risk 20 times per second and is a way to close the decision circuit on a SOTIF measurement : even if you think that your decision making system is very good (and probably it is the fact on the UBER/VOLVO robot car : the Autonomous Driving system seems to to be very good), you should always measure driving risk (as a consequence of your driving behaviour), just to make sure, and apply the very simple algorithm : "if risk is to high, then slow down the car and alert the human driver". See demo of SafetyNex embedded in a smartphone : https://www.youtube.com/watch?v=nQpn7x2s-x0 

NEXYAD has started to integrate SafetyNex into ADAS and TELEMATICS products of big OEMs worldwide, and we think that this technology is now ready for application in the robot taxi/shuttle domain (ask sales@nexyad.net for any detail).

Autonomous cars must take into account visibility conditions (using VisiNex) EVEN IF they are not computer-vision based because human beings are, and they must be "AWARE" of the risk they take (usig SafetyNex), dynamically, in real time. With those features, autonomous cars will be much safer.